Freedom and Determinism

Freedom and Determinism
edited by
Joseph Keim Campbell,
Michael O’Rourke,
and David Shier
Topics in Contemporary Philosophy

Topics in Contemporary Philosophy
Editors
Joseph Keim Campbell, Washington State University
Michael O’Rourke, University of Idaho
David Shier, Washington State University
Editorial Board Members

Kent Bach, San Francisco State University
Michael Bratman, Stanford University
Nancy Cartwright, London School of Economics
Richard Feldman, University of Rochester
John Martin Fischer, University of California, Riverside
Nicholas F. Gier, University of Idaho
Philip J. Ivanhoe, Boston University
Michael McKinsey, Wayne State University
John Perry, Stanford University
James Rachels, University of Alabama at Birmingham
Stephen Schiffer, New York University
Harry Silverstein, Washington State University
Brian Skyrms, University of California, Irvine
Peter van Inwagen, University of Notre Dame
edited by Joseph Keim Campbell, Michael O’Rourke, and David Shier
A Bradford Book
The MIT Press
Cambridge, Massachusetts
London, England
© 2004 Massachusetts Institute of Technology
All rights reserved. No part of this book may be reproduced in any form by any elec-
tronic or mechanical means (including photocopying, recording, or information
storage and retrieval) without permission in writing from the publisher.
This book was set in Stone Serif and Stone Sans by Interactive Composition Corpo-
ration and was printed and bound in the United States of America.
Library of Congress Cataloging-in-Publication Data
Freedom and determinism / edited by Joseph Keim Campbell, Michael O’Rourke, and
David Shier.
p. cm.—(Topics in contemporary philosophy)
“A Bradford book.”
Includes bibliographical references and index.
ISBN 0-262-03319-4 (hc. : alk. paper)—ISBN 0-262-53257-3 (pbk. : alk. paper)
1. Free will and determinism. I. Campbell, Joseph Keim, 1958– . II. O’Rourke,
Michael, 1963– . III. Shier, David, 1958– . IV. Series.
BJ1461.F755 2004
123'.5—dc22
2003066491
10 9 8 7 6 5 4 3 2 1
Contents
Freedom and Determinism: A Framework 1

Joseph Keim Campbell, Michael O’Rourke, and David Shier
Determinism, Freedom, and Agency 19
1 Determinism: What We Have Learned and What We Still

Don’t Know 21
John Earman
2 Freedom and the Power of Preference 47

Keith Lehrer
3 Agency, Responsibility, and Indeterminism: Reflections on Libertarian

Theories of Free Will 70
Robert Kane
4 Trying to Act 89
Carl Ginet
The Metaphysics of Moral Responsibility 103
5 The Sense of Freedom 105

Dana K. Nelkin
6 Libertarian Openness, Blameworthiness, and Time 135
Ishtiyaque Haji
7 Moderate Reasons-responsiveness, Moral Responsibility, and

Manipulation 151
Todd R. Long
vi Contents
8 Which Autonomy? 173

Nomy Arpaly
9 The Transfer of Nonresponsibility 189
John Martin Fischer
The Compatibility Problem 211
10 Van Inwagen on Free Will 213

Peter van Inwagen
11 Compatibilist Options 231
John Perry
12 Freedom and Contextualism 255
Richard Feldman
13 Buddhism and the Freedom of the Will: Pali and Mahayanist

Responses 277
Nicholas F. Gier and Paul Kjellberg
14 After Compatibilism and Incompatibilism 305

Ted Honderich
Index 323
Freedom and Determinism: A Framework
Joseph Keim Campbell, Michael O’Rourke, and David Shier
1 The Traditional Problem of Freedom and Determinism
Thoughts about freedom and determinism have engaged philosophers

since the days of ancient Greece.1 On the one hand, we generally regard
ourselves as free and autonomous beings who are responsible for the ac-
tions that we perform. But this idea of ourselves appears to conflict with a
variety of attitudes that we also have about the inevitable workings of the
world around us. For instance, some people believe that strict, universal
laws of nature govern the world. Others think that there is an omnipotent
God who is the ultimate cause of all things. These more global views sug-
gest that each particular event—including each human action—is causally
necessitated, and so they suggest a conflict with the claim that we are free.
Hence, the problem of freedom and determinism is, at base, a problem
about reconciling attitudes we have toward ourselves with our more gen-
eral thoughts about the world around us. It is a problem about locating our
actions within those streams of events that make up the broader universe.
Freedom is usually discussed within the context of theoretical concerns
about the nature of moral responsibility. For it is a basic assumption that
some kind of freedom—call it “moral freedom”—is a necessary precondi-
tion for our being accountable for our actions. Moreover, even those who
endorse moral nihilism, the claim that no one is ever morally responsible for
anything, usually do so because they also believe that we lack moral free-
dom. Consequently, the assumption of freedom plays a role in our beliefs
about the appropriateness of moral praise and blame. We find it absurd to
blame a rock that happens to crash through our living room window but
acceptable to blame the child who threw the rock. And we would consider
such blame more legitimate were the rock knowingly and intentionally
2 J. K. Campbell, M. O’Rourke, and D. Shier
thrown by an adult with normal cognitive capacities. In trying to uncover

the basis for these differences in attitude, we encounter other, more funda-
mental distinctions in moral psychology between action and passion, be-
lief and desire, reason and emotion, and control and compulsion. Not
surprisingly persons in the fields of ethics, philosophy of psychology, and
philosophy of law all share an interest in understanding the nature of
moral freedom. This remains true for moral nihilists since some under-
standing of the nature of moral freedom is implicit in its denial.
There are a variety of kinds of determinism that have been offered as
potential threats to our freedom. For instance, there is logical determinism,
the view that all propositions—including those reporting our future
actions—are either true or false. There is also theological determinism, ac-
cording to which an omniscient God knows about the future in complete
detail. T. S. Eliot (1943) notes a problem between freedom and temporal
determinism, which claims that time is another dimension like any of the
other three spatial dimensions, so that the difference between what is in
your past and what is in your future is a lot like the difference between
what is to your left and what is to your right. Lastly, there is causal deter-
minism, which claims that the past facts, together with the laws of nature,
entail all future facts.2 Each of these determinisms is a global thesis, making
a claim about all propositions. It is then suggested that this global property
carries with it the kind of necessity that is itself a threat to our freedom.
Of course, the term ‘freedom’ is also ambiguous. There is political freedom
and freedom of religion. These and other freedoms are characterized in a neg-
ative way, as the absence of certain constraints on one’s activities or beliefs.
Since there are many kinds of constraints, there are many negative free-
doms. But the kind of freedom that metaphysicians are interested in—call it
‘metaphysical freedom’—can also be described in a positive way, as an active
power to do things that are up to us. In this sense, metaphysical freedom
seems more fundamental than the other, merely negative freedoms.3 Meta-
physical freedom may be even more fundamental than moral freedom, for
as the problems noted in the previous paragraph indicate, threats to our
freedom can be presented without mentioning attributions of moral praise
or blame. On the other hand, each of the determinisms noted above is also
a potential threat to our moral freedom. By and large, folks in the twentieth
century are concerned with moral freedom—the freedom-relevant condi-
tion necessary for moral responsibility—and causal determinism.
For this reason, debates about freedom and determinism center on a set
of distinct, though related, questions.4
1. What is the thesis of causal determinism?
2. What is the nature of moral freedom?
3. Is moral freedom compatible with causal determinism?
4. Is causal determinism true?
5. Are any persons morally free?
The first three questions are conceptual and, thus, the primary focus of
philosophers is directed towards answering those questions. A provisional
answer to (1) was given above, but more precisely, determinism is the con-
junction of the following two theses:
For every instant of time, there is a proposition that expresses the state of
the world at that instant;
If p and q are any propositions that express the state of the world at some
instants, then the conjunction of p with the laws of nature entails q. (van
Inwagen 1983, 65)
Note that there are no temporal restrictions on p and q. Still if p is a propo-

sition that expresses the state of the world at some time in the past, then
when conjoined with the laws of nature it entails each and every future
fact.
Until recently, all parties in the freedom and determinism debate ac-
cepted a short answer to (2), regarding the nature of moral freedom. First,
it was held that the kind of freedom required for moral responsibility was
free will. Second, it was accepted that free will requires that persons have
alternatives to at least some of their actions, where a person S has an alter-
native to an action iff S can do otherwise, or if S is able to do otherwise, or
if it is within S’s power to do otherwise. Call the identification of moral
freedom with free will, the ‘traditional view’ of moral freedom.5
The traditional view makes it rather easy to understand the philosophi-
cal problems surrounding freedom and determinism, for free will is also a
likely candidate for the more basic metaphysical freedom. According to the
traditionalist, there are not three distinct kinds of freedom—metaphysical
freedom, moral freedom, and free will. There is only one kind of freedom,
free will, and it is essential to moral praiseworthiness and blameworthiness.
Proponents of the traditional view tend to fall into well-defined groups,
depending on how they answer the remaining questions above. Compati-

bilists believe that determinism is consistent with the free will thesis—the
view that at least some persons have free will—whereas incompatibilists
believe that it is not. Soft determinists are compatibilists who accept both de-
terminism and the free will thesis, whereas hard determinists are incompat-
ibilists who endorse determinism but deny the free will thesis. Finally,
libertarians are incompatibilists who deny determinism and endorse the
free will thesis (van Inwagen 1983). The following chart should help the
reader to keep track of the theories noted in this paragraph.
(3) Compatibility? (4) Determinism? (5) Freedom?
Compatibilism Yes
Incompatibilism No
Soft Determinism Yes Yes Yes
Hard Determinism No Yes No
Libertarianism No No Yes
During the last century a few of the above theories began to look rather
archaic. First of all, certain results in quantum mechanics led many philoso-
phers eventually to reject determinism. According to the standard interpre-
tation of quantum theory, there are facts about the sub-atomic level—facts
like this electron has the property of “spin up”—that are not a logical conse-
quence of any conjunction of past facts and laws of nature. Subsequently,
many philosophers have abandoned the positions of soft and hard deter-
minism since they both entail a positive answer to (4), and the acceptance
of determinism.
In addition, philosophers began to question the relevance of the concept
of free will to moral praiseworthiness and blameworthiness. Motivation for
the traditional view of moral freedom stems from an acceptance of some-
thing like the following principle:
The principle of alternative possibilities (PAP): S is morally responsible
for action a only if she has, or had, alternatives to a.
But in an important and influential paper, Harry Frankfurt provides appar-
ent counterexamples to PAP, cases in which agents perform blameworthy
actions even though it appears that they could not have done otherwise.
Frankfurt writes:
Suppose someone—Black, let us say—wants Jones4 to perform a certain action. Black
is prepared to go to considerable lengths to get his way, but he prefers to avoid
showing his hand unnecessarily. So he waits until Jones4 is about to make up his
mind what to do, and he does nothing unless it is clear to him (Black is an excellent
judge of such things) that Jones4 is going to decide to do something other than what
he wants him to do. If it does become clear that Jones4 is going to decide to do some-
thing else, Black takes effective steps to ensure that Jones4 decides to do, and that he
does do, what he wants him to do. Whatever Jones4’s initial preferences and inclina-
tions, then, Black will have his way. (Frankfurt 1969, 148–149)
Perhaps Black wants Jones4 to rob a bank, and that Jones4 robs the bank
without the need of Black’s counterfactual intervention. It seems, then,
that Jones4 is morally responsible for his action but that he could not
have done otherwise. Call such examples ‘Frankfurt examples’. To ac-
cept that Frankfurt examples are genuine counterexamples to PAP is to ac-
cept a split between moral freedom and alternatives, for the Frankfurt
examples suggest that one can have the former without having the latter.
This means that the traditional view is wrong about the nature of moral
freedom.6
Lastly, some philosophers have abandoned the traditional view because
of the development of a number of persuasive formal arguments in support
of incompatibilism, presented primarily by Peter van Inwagen (1975, 1983,
1989) and Carl Ginet (1966, 1990). These are all versions of the Consequence
Argument.
If determinism is true, then our acts are the consequences of the laws of nature and
events in the remote past. But it is not up to us what went on before we were born,
and neither is it up to us what the laws of nature are. Therefore, the consequences of
these things are not up to us. (van Inwagen 1983, 16)
Van Inwagen (1983) offers three formal versions of the Consequence

Argument. In the third version—aptly called ‘the Third Argument’—he
constructs a modal operator ‘N’ where ‘Np’ means ‘p and no one has, or
ever had, any choice about whether p’ (1983, 93; 1989, 404). Intuitively, we
can think of this as the unavoidability operator. Van Inwagen claims that
according to the logic of this operator the following two inference rules are
valid:
(␣) From p deduce Np

(␤) From Np and N(p→q) deduce Nq,
where ‘ ’ represents broad logical necessity and ‘→’ represents material im-
plication (1983, 94; 1989, 227). From the above rules and definitions, to-
gether with reasonable assumptions about the past and the laws of nature,
one can show that no one has, or ever had, any choice about whether any
true proposition is true, given determinism. Thus, the Third Argument
apparently establishes incompatibilism.7
The three developments discussed above have led to a number of new
theories in the freedom and determinism debate. Among the more attrac-
tive of these is semicompatibilism, inspired by the rejection of PAP and the
success of the Consequence Argument (Fischer 1994). According to the
semicompatibilist, moral freedom is not essentially linked to alternatives.
The semicompatibilist can agree that free will is incompatible with deter-
minism but also contend that determinism and moral responsibility
are compatible. Moral freedom is best understood in terms of concepts like
guidance control and reasons-responsiveness that rely on features about the
actual causal history leading up to an action, instead of whether or not
there are any alternatives to the action.8
Libertarianism is still widely held.9 Moreover, many incompatibilists
deny the free will thesis altogether. Hard determinism is rarely endorsed,
since few philosophers accept determinism, but in its place are successor
views, like hard indeterminism, according to which incompatibilism is true,
determinism is false, but the free will thesis is false anyway (Pereboom
2001). Most successor views differ from traditional versions of moral ni-
hilism. Traditionalists begin with the Consequence Argument. Hard deter-
minists then argue that determinism is true, and thus all human actions are
unavoidable. Other traditionalists extend the Consequence Argument to
cover cases of indeterminism, as well. For instance, after presenting the Third
Argument one can go on to show that most propositions are unavoidable
even if one assumes the truth of indeterminism (van Inwagen 1983, 1989).
Proponents of successor views often believe that unavoidability is unimpor-
tant to moral freedom, given the Frankfurt examples. Instead, they claim
that moral freedom is essentially linked to origination, and claim that origi-
nation is impossible under the assumption of determinism and unlikely at
best even if indeterminism is true.10
In summary, the majority of contemporary philosophers agree that some
kind of freedom—moral freedom—is required for moral responsibility. But
they differ as to the nature of this freedom as well as some of the other nec-
essary conditions for moral responsibility. Proponents of the traditional
view continue to maintain that moral freedom is just free will, but a variety
of philosophers have rejected the latter notion altogether. This is primarily
due to the impact of the Frankfurt examples and formal arguments for
incompatibilism. Moreover, while debate about the compatibility of moral
freedom and determinism is still alive and well, most philosophers have
rejected determinism given quantum mechanics. Gone are the labels of
soft determinism and hard determinism, but the lion’s share of opinions on
the nature of freedom and determinism still fall into three main groups:
libertarians, moral nihilists, and compatibilists, including semicompati-
bilists. We now introduce the essays in this current volume.
2 New Work on Freedom and Determinism
As we have seen, there are essentially two projects that are important to
contemporary philosophers working in the area of freedom and determin-
ism: (a) understanding the nature of moral freedom and its relationship
with moral responsibility, and (b) determining whether or not moral free-
dom is compatible with determinism. That is, contemporary philoso-
phers are primarily concerned with questions (2) and (3) from our earlier
list. However, one cannot address these questions without a provisional
understanding of the general concepts involved—determinism and moral
freedom—as well as a clearer sense of the type of events relevant to our con-
cerns here, namely, actions. These presuppositional topics are the subject
of the essays in our first section, “Determinism, Freedom, and Agency.” We
then turn to the nature of moral freedom and responsibility in “The Meta-
physics of Moral Responsibility,” and finally, to questions about the com-
patibility of freedom and determinism in “The Compatibility Problem.”11
Determinism, Freedom, and Agency

Many standard ways of treating the problem of freedom and determinism
presume answers to questions (1) and (4) above. As it turns out, it is not
easy to say what it would be for the world to be deterministic, and even
less easy to see whether, on any plausible account of what determinism
amounts to, the world is in fact deterministic. In “Determinism: What We
Have Learned and What We Still Don’t Know,” John Earman briefly reviews
relevant parts of current physics with respect to their compatibility with
determinism. He begins by distinguishing between determinism and pre-
diction. While the latter may entail the former, the converse is false. Hence,
one cannot infer from an inability to predict future states of a system that
the time evolution of the system is indeterministic. Earman’s review covers

determinism and predictability not only of quantum systems, but also clas-
sical and relativistic systems. One of the interesting results he reaches is
that, contrary to naive intuition, classical Newtonian physics is more hos-
tile to determinism than either quantum mechanics or special relativity.
Another is that the prospects for a deterministic theory unifying general
relativity and quantum theory are rather grim. After noting various points
of tension between determinism and each of these parts of current theory,
Earman identifies ways in which extensions of the theories might save
determinism, but argues that, as things now stand, questions about the
truth or falsity of determinism remain open.12
In keeping with the introductory nature of the first section, the next two
essays in the volume offer two distinct theories of moral freedom: a com-
patibilist account and a libertarian one. Both essays are exceptionally broad
in their scope, covering over a half-century of scholarly research combined.
Keith Lehrer’s “Freedom and the Power of Preference” is a break from his
earlier views. In Lehrer (1980), the author argues that total integration of
preferences is sufficient for moral freedom. Since total integration is com-
patible with determinism, it follows that moral freedom is compatible with
determinism, too. Lehrer now believes that total integration is not suffi-
cient for moral freedom because the integrated structure might be pro-
duced by another person and, thus, it would not be a preference structure
of which the agent is the author. According to Lehrer’s new view, in order
for our actions to be free they must be the result of a preference structure
that the agent prefers to have, and one that the agent has because
he prefers to have it. This is sufficient, according to Lehrer, for the agent to
be the author of his own actions yet is still consistent with determinism.
In “Agency, Responsibility, and Indeterminism: Reflections on Libertar-
ian Theories of Free Will,” Robert Kane embraces an incompatibilist theory
of moral freedom. Kane is careful to distinguish between two require-
ments that are central to libertarianism. First, there is the condition of
alternative possibilities (AP), which claims that alternatives are necessary
for free will. Second, there is the ultimacy condition, which requires that
the agent be the ultimate source of her morally free actions. Kane notes
that although AP may be necessary for free will it is not sufficient. Thus,
an account of libertarian free will must add something else to the mere exis-
tence of alternatives. Kane adds the condition of ultimate responsibility,
which includes the ultimacy condition. But Kane distances himself from
other, more classic libertarian theories, like the libertarian agency theory,
which holds that agent causation is sui generis and not reducible to event
causation.13 Instead, Kane adopts a teleological theory that attempts to
explicate moral freedom in terms of the reasons or purposes of the agent.
Kane closes his essay with a lengthy response to allegations that the no-
tion of indeterministic action is incoherent.
The kinds of things that we are normally held responsible for are actions,
or their consequences, so the problem of freedom and determinism is es-
sentially a problem about reconciling free actions with determinism. Thus,
in addition to delving into the intricacies of the latter view, theorists often
engage in important work on the nature of human action. In “Trying to
Act,” Carl Ginet offers four conditions, each of which he claims is sufficient
for the truth of sentences of the form, ‘S tried to a’. Ginet falls short of of-
fering an analysis since he does not claim that any, or all, of the conditions
are necessary for the truth of such sentences. Still, Ginet notes some im-
portant features that all four conditions share. For instance, in each condi-
tion S tries to a by doing something else which S believes is connected in a
certain way with the possibility of doing a. To try to do a is to do something,
even if one fails to do a, so these are also conditions for the truth of sen-
tences reporting the performance of a certain class of actions.
The Metaphysics of Moral Responsibility

The papers in this section discuss question (2) in detail: What is the nature
of moral freedom, the freedom-relevant condition for moral responsibility?
Dana Nelkin’s “The Sense of Freedom” is a sustained discussion of the
following principle:
(R): Rational deliberators, in virtue of their nature as rational

deliberators, necessarily have a sense that they are free.
Establishing (R) is important for at least two reasons. First, it is one of the
most central and universally agreed upon beliefs that we have about moral
freedom. Second, philosophers like Immanuel Kant use (R) as an important
premise in arguments against moral nihilism. After considering and dis-
missing another version of (R), Nelkin concludes that the ‘belief-concept’
reading of it should be accepted. Thus, according to Nelkin, all rational
deliberators essentially believe that their actions are up to them in the
sense that they are accountable for those actions.
In “Libertarian Openness, Blameworthiness, and Time,” Ishtiyaque Haji

argues against a traditional belief in moral theory that praiseworthiness
and blameworthiness are essentially retrospective, that is, that we cannot
hold a person responsible for an action until after the action has been per-
formed. Against this, Haji argues in support of blame future, the view that it
is possible for S to be morally blameworthy for doing a at t at some time
prior to t. Haji’s intuitions run counter to those who believe in libertarian
openness (L-openness), the view that if S has free will then it was always pos-
sible for S to do otherwise at any time prior to performing an action. Pro-
ponents of L-openness argue that we cannot be responsible for our future
free actions since it is undetermined as to whether or not they will occur
until the action is actually performed. But Haji rejects this argument by
noting that one can construct Frankfurt examples in which S is morally re-
sponsible for performing some action even though the action is not L-open
to S. Haji goes on to discuss a variety of conceptions of moral responsibil-
ity that might lie behind the rejection of blame future. Haji critiques all of
these views and, instead, advocates the self-disclosure conception of moral
responsibility that is consistent with blame future.
The Frankfurt examples also play an important role in Todd Long’s
“Moderate Reasons-responsiveness, Moral Responsibility, and Manipula-
tion.” Long’s paper is a critique of John Martin Fischer and Mark Ravizza’s
view of moral freedom (1998). According to Fischer and Ravizza, agents in
the Frankfurt examples lack regulative control over their actions since they
cannot do otherwise. But such agents have guidance control over their ac-
tions as long as their actions are a result of mechanisms that are moderately
reasons-responsive (MRR). Moreover, it is guidance control, and not regula-
tive control, that is the freedom-relevant condition for moral responsibil-
ity, according to Fischer and Ravizza. Long offers some Frankfurt examples
that he thinks pose problems for Fischer and Ravizza’s theory. More specif-
ically, the examples are such that the counterfactual intervention involves
cases of direct manipulation of the inputs to the agent’s deliberative
processes by either adding or replacing different reasons for action. Intu-
itively, in such cases the potential actions in the alternative sequence
scenario—the actions that would result were the counterfactual interven-
tion to be operative—arise from the same mechanism as those performed
in the actual sequence scenario—the actions that do result, without the aid
of intervention. But an important assumption of Fischer and Ravizza is that
agents are morally responsible for their actions in the actual sequence sce-
nario yet these same agents would not have been morally responsible for
their actions were the counterfactual intervention to be enacted. Fischer
and Ravizza are then faced with a dilemma. Either they must concede that
there is more to moral responsibility than an action’s being the result of a
mechanism that is MRR or they must give up their belief that their theory
explains why agents in the Frankfurt examples would not have been
morally responsible for their actions had the counterfactual intervention
taken place. Long also suggests that agents may be morally responsible for
actions even if they undergo external manipulations that are quite severe.
Nomy Arpaly’s paper, “Which Autonomy?” is a general critique of the
concept of autonomy, one of the more popular candidates for moral freedom.
First, Arpaly discusses a variety of divergent accounts of autonomy that
philosophers have offered. This alone poses a problem for such theorists
since the diversity of such accounts has rendered the concept of autonomy
practically meaningless. Arpaly goes on to offer substantive criticisms of
each of these accounts and ends by advising that we might be better off deal-
ing more directly with moral praiseworthiness and blameworthiness and
abandoning the notion of autonomy altogether.
We close the second section with an essay by John Martin Fischer enti-
tled “The Transfer of Nonresponsibility.” Fischer’s contribution is part of a
long-standing debate—one in which he has exerted substantial influence—
on the Principle of the Transfer of Nonresponsibility (Transfer NR), a ver-
sion of which was put forth by van Inwagen. Van Inwagen’s principle is
formally identical to his rule (␤), noted above, only here ‘Np’ means “p and
no human being, or group of human beings, is even partly responsible for
the fact that p” (van Inwagen 1980). Both rule (␤) and Transfer NR can be
seen as closure principles of unavoidability and nonresponsibility, respec-
tively. Fischer formulates the principle as follows:
Transfer NR: If p obtains and no one is even partly morally responsible for
p and if p obtains, then q obtains, and no one is even partly morally
responsible for the fact that if p obtains, then q obtains, then q obtains,
and no one is even partly morally responsible for q. (Fischer, chapter 9)
Transfer NR has recently come under attack by semicompatibilists, like
Fischer, who argue that the Frankfurt examples are counterexamples to this
principle too. To use an example from Fischer’s essay, suppose that q is the
fact that an avalanche occurs and that the avalanche was started by an
agent, Betty, but that the avalanche would have occurred even if Betty had
not started it due to the fact that a glacier was eroding, p. No one is respon-
sible for the fact that p, and no one is responsible for the fact that if p, then
q, but it seems that Betty is responsible for the fact that q.14 Proponents of
Transfer NR have tried to alter the principle to meet this counterexample.
Fischer considers and responds to a variety of these reformulations. He con-
cludes that the newer versions of Transfer NR cannot be used to support the
incompatibilism of determinism and moral responsibility. At most, he says,
they lead only to a dialectical stalemate.
The Compatibility Problem

The final section of the volume contains essays on question (3) above, the
issue of whether or not some substantive notion of freedom is compatible
with the thesis of causal determinism. Like some of the other contributions
in this volume, “Van Inwagen on Free Will” offers an important retrospec-
tive on an influential body of work given from the author’s own point of
view. By now, many of van Inwagen’s views should be well known to the
reader since they have had some bearing on almost all of the issues dis-
cussed above. Van Inwagen adopts the traditional view of moral freedom as
free will and has offered compelling arguments in support of incompatibil-
ism as well as the incompatibility of determinism and moral responsibility.
In this essay, van Inwagen also discusses his thoughts on the Frankfurt
examples, which he still finds unconvincing, and his argument for restric-
tivism, the view that we are rarely, if ever, free. He closes with some criti-
cisms of the concept of agent causation.
John Perry’s “Compatibilist Options” is both a defense of compatibilism
and a direct response to one of van Inwagen’s arguments for incompatibil-
ism, the ‘First Formal Argument’ (1975; 1983, 69–78). According to Perry,
there are two distinct ways in which compatibilists can avoid a conflict be-
tween metaphysical freedom and determinism. They can adopt a soft view
of laws of nature, and accept that laws are little more than true universal
generalizations, or they can adopt a soft view of ‘can’.15 Perry advocates the
second strategy. He endorses a hard view of laws of nature and then argues
that metaphysical freedom and determinism are nonetheless compatible.
In his discussion of van Inwagen’s argument, he notes that it involves a
subtle ambiguity in the phrase ‘renders a proposition false’. He then rejects
the argument by providing a counterexample to one of its key premises,

the premise most similar to rule (␤).
Many contemporary philosophers have noted structural similarities
between rule (␤), Transfer NR, and the following:
The principle of epistemic closure: From S knows that p and S knows that
p entails q, deduce S knows that q.
This latter principle plays an important role in arguments in support of

skepticism, the view that we have no knowledge. Given the formal similari-
ties of these various closure principles, it is fashionable of late to try to
bring the technical apparatus from the epistemology debate to bear on the
problem of freedom and determinism. For instance, John Hawthorne
(2001) has recently put forth a view of contextualism according to which
sentences about freedom are context sensitive. Thus, in ordinary contexts
when an English speaker utters sentences of the form ‘S does a freely’, the
sentence comes out true even if determinism is true. In “Freedom and Con-
textualism,” Richard Feldman argues against Hawthorne’s theory. Among
his many reasons for rejecting contextualism, Feldman notes that the con-
textualist concedes too much to the incompatibilist for he admits that in
philosophical contexts it is usually false to say that ‘S does a freely’ given
determinism. In addition, Feldman claims that the contextualist fails to
take our incompatibilist worries about moral freedom seriously.
In “Buddhism and the Freedom of the Will: Pali and Mahayanist Re-
sponses,” Nicholas F. Gier and Paul Kjellberg provide a comparative discus-
sion of freedom and determinism from an Asian perspective. With few
exceptions, thinkers across the ancient world embraced determinism with
no recognition that it might undermine moral responsibility. According to
Gier and Kjellberg, this suggests that the issue of free will is a distinctively
modern one, initiated by Augustine and carried over into European philos-
ophy. Buddhist philosophers writing in the Pali language generally support
a form of compatibilism that compares favorably with Lehrer’s idea that
moral freedom lies in developing preferences that lead to the good life.
Mahayanist Buddhist philosophers writing in Sanskrit embraced forms of
idealism or skepticism that appear to dissolve personal agency altogether.
Anticipating French deconstruction, Nagarjuna, one of the most sophisti-
cated Buddhist philosophers, deconstructs the self and implies that the talk
of freedom versus determinism is simply competing rhetorics that have no
foundation in reality itself.
The volume ends with an original and controversial contribution by Ted

Honderich. In “After Compatibilism and Incompatibilism,” Honderich en-
dorses a theory of determinism, and then argues strongly for the startling
thesis that both compatibilism and incompatibilism are wrong. Each
claims that we have one settled conception of moral freedom, or one im-
portant conception, and this, he argues, is demonstrably false. The real
problem of determinism has seemed to him until recently to be the practi-
cal one of dealing with the fact that we not only have attitudes consistent
with determinism, but also attitudes inconsistent with it. But now he
thinks there is a further problem. Reflection on your past life can issue in
confidence in determinism but also in feelings of moral responsibility akin
to those that depend on indeterminism. On what fact can they rest?
Honderich does not solve this problem directly but he does suggest two av-
enues of pursuit: embrace a radical theory about the nature of consciousness,
or question longstanding assumptions about causation and explanation.
His paper is a useful stopping place for this anthology since it suggests that
no matter how much has been said about the issues of freedom and deter-
minism there is always something new to add to the debate.
Notes
1. See Aristotle (1963) along with writings by the Stoics and Epicureans in Long and
Sedley (1987). This is disputed by Gier and Kjellberg, chapter 13, who argue that the
problem is a modernist one.
2. For discussions of the problems of freedom and logical/theological determinism

in ancient times, see Aristotle (1963) along with writings by the Stoics in Long and
Sedley (1987). For more contemporary essays, see Fischer (1989) and Kane (2002,
part 1). A more precise definition of ‘determinism’ is provided below.
3. See van Inwagen (1998, 365–366). The term ‘metaphysical freedom’ is van
Inwagen’s. The term ‘active power’ comes from Reid (1983).
4. For a related, though somewhat different set of questions and problems, see van
Inwagen (1983, 1–2) and Kane, chapter 3.
5. Following van Inwagen (1983, 8), we use the term ‘free will’ out of respect for tra-
dition. By use of the term we do not suggest that there is some faculty, e.g., the will,
that has the property of being free. To say that S has free will is just to say that S has
alternatives, in the sense noted in this paragraph. This connection between the con-
cepts of free will and alternatives has been accepted by virtually everyone throughout
the history of philosophy. Frankfurt (1969, 1971) remains an exception.
6. Not all philosophers agree that the Frankfurt examples undermine PAP. See the
articles in Fischer (1986, part 2), Fischer and Ravizza (1993, part 3), and Kane (2002,
part 5). For discussions of the Frankfurt examples in this volume, see Haji, chapter 6,
Long, chapter 7, Fischer, chapter 9, and van Inwagen, chapter 10.
7. Let ‘P0’ represent a proposition that expresses the state of the world at some time
t0 before any human beings existed, let ‘L’ represent the conjunction of the laws of
nature, and let ‘P’ represent any true proposition. Here is van Inwagen’s Third Argu-
ment (1983, 93–104; 1989, 404–405).
(1) ((P0 & L) → P) assumption of determinism

(2) (P0 → (L → P)) from (1) by exportation
(3) N(P0 → (L → P)) from (2) by (␣)
(4) NP0 premise
(5) N(L → P) from (3), (4) by (␤)
(6) NL premise
(7) NP from (5), (6) by (␤)
8. Fischer (1982, 1987, 1994) and Fischer and Ravizza (1998). For other theories of
moral freedom along the lines of semicompatibilism, see Lehrer (1980) and the
articles in Fischer (1986, part 1), Fischer and Ravizza (1993, part 2), and Kane (2002,
part 4). This latter section includes discussions of the new compatibilists, who provide
naturalistic conceptions of moral responsibility inspired by P. Strawson (1962).
9. For contemporary discussions of libertarianism, see O’Connor (1995) and Kane

(2002, part 6). See also Kane, chapter 2, and van Inwagen, chapter 10.
10. For versions of the successor view, see Kane (2002, part 7), G. Strawson (1986),
Pereboom (2001), and Honderich, chapter 16. The term ‘successor view’ is Kane’s.
See also Kane, chapter 3, for a discussion of origination. Van Inwagen is not a moral
nihilist though he does embrace a related view, called ‘restrictivism’, defined below.
11. The expression ‘compatibility problem’ is from van Inwagen (1983, 2).
12. We thank Bruce Glymour for providing the basis of this paragraph.
13. See Reid (1983), Campbell (1957), Taylor (1963), and Chisholm (1964). For more
recent versions of the theory, as well as criticisms of it, see articles in O’Connor
(1995) and Kane (2002, part 6). Kane, chapter 2, and van Inwagen, chapter 10, con-
tain interesting criticisms of the concept of agent causation from the libertarian
perspective.
14. This is based on an example from Ravizza (1994, 78).
15. For compatibilists representing the former approach, see Lewis (1981) and
Lehrer, chapter 2. For compatibilists representing the latter approach, see Moore
(1912) and Lehrer (1976, 1980). Kane (2002, part 4) also contains some recent essays
of interest on compatibilism.
References
Aristotle. 1963. Categories and De Interpretatione. J. L. Ackrill, trans. Oxford:

Clarendon Press.
Campbell, C. A. 1957. On Selfhood and Godhood. London: George Allen and Unwin.
Chisholm, R. 1964. “Human Freedom and the Self.” The Lindley Lecture. Lawrence,
Kans.: University of Kansas, Department of Philosophy. Reprinted in Pereboom
(1997).
Eliot, T. S. 1943. “Burnt Norton.” Four Quartets. New York: Harcourt, Brace.
Fischer, J. M. 1982. “Responsibility and Control.” Journal of Philosophy 89: 24–40.

Reprinted in Fischer (1986).
. ed. 1986. Moral Responsibility. Ithaca, N.Y.: Cornell University Press.
. 1987. “Responsiveness and Moral Responsibility.” In F. Schoeman, ed.,

Responsibility, Character, and the Emotions: New Essays on Moral Psychology.
Cambridge: Cambridge University Press.
. ed. 1989. God, Foreknowledge, and Freedom. Stanford, Calif.: Stanford Univer-
sity Press.
. 1994. The Metaphysics of Free Will. Oxford: Blackwell.
Fischer, J. M., and M. Ravizza, eds. 1993. Perspectives on Moral Responsibility. Ithaca,
N.Y.: Cornell University Press.
. 1998. Responsibility and Control: A Theory of Moral Responsibility. Cambridge:

Cambridge University Press.
Frankfurt, H. G. 1969. “Alternate Possibilities and Moral Responsibility.” Journal of

Philosophy 66: 828–839. Reprinted in Fischer (1986). (All page references are to this
latter book.)
. 1971. “Freedom of the Will and the Concept of a Person.” Journal of Philoso-
phy 68: 5–20. Reprinted in Fischer (1986).
Ginet, C. 1966. “Might We Have No Choice?” In K. Lehrer, ed., Freedom and Deter-
minism. New York: Random House.
. 1990. On Action. Cambridge: Cambridge University Press.
Hawthorne, J. 2001. “Freedom in Context.” Philosophical Studies 104: 63–79.
Kane, R. 2002. The Oxford Handbook of Free Will. Oxford: Oxford University Press.
Lehrer, K. 1976. “‘Can’ in Theory and Practice: A Possible Worlds Analysis.” In M.

Brand and D. Walton, eds., Action Theory. Dordrecht: D. Reidel. Reprinted as “A
Possible Worlds Analysis of Freedom” in Lehrer (1990).
. 1980. “Preferences, Conditionals and Freedom.” In P. van Inwagen, ed., Time

and Cause: Essays Presented to Richard Taylor. Dordrecht: D. Reidel. Reprinted in
Lehrer 1990.
. 1990. Metamind. Oxford: Clarendon Press.
Lewis, D. 1981. “Are We Free to Break the Laws?” Theoria 47: 113–121. Reprinted in
Lewis (1986). Philosophical Papers, volume 2. Oxford: Oxford University Press.
Long, A. A., and D. N. Sedley, eds. and trans. 1987. The Hellenistic Philosophers:
volume 1: Translations of the Principle Sources. Cambridge: Cambridge University Press.
Moore, G. E. 1912. Ethics. New York: Oxford University Press.
O’Connor, T., ed. 1995. Agents, Causes, and Events: Essays on Indeterminism and Free
Will. Oxford: Oxford University Press.
Pereboom, D. 1997. Free Will. Indianapolis, Ind.: Hackett.
. 2001. Living without Free Will. Cambridge: Cambridge University Press.
Ravizza, M. 1994. “Semi-Compatibilism and the Transfer of Nonresponsibility.”

Philosophical Studies 75: 61–93.
Reid, T. 1983. Inquiry and Essays. R. E. Beanblossom and K. Lehrer, eds. Indianapolis,
Ind.: Hackett.
Strawson, G. 1986. Freedom and Belief. Oxford: Clarendon Press.
Strawson, P. F. 1962. “Freedom and Resentment.” Proceedings of the British Academy

48: 1–25.
Taylor, R. 1963. Metaphysics. Englewood Cliffs, N.J.: Prentice-Hall.
van Inwagen, P. 1975. “The Incompatibility of Free Will and Determinism.” Philo-
sophical Studies 27: 185–199.
. 1980. “The Incompatibility of Responsibility and Determinism.” In M.

Bradie and M. Brand, eds., Bowling Green Studies in Applied Philosophy, vol. 2: 30–37.
Bowling Green, Ohio: Bowling Green State University. Reprinted in Fischer (1986):
241–249. (All page references are to this latter work.)
. 1983. An Essay on Free Will. Oxford: Clarendon Press.
. 1989. “When Is the Will Free?” In J. Tomberlin, ed., Philosophical Perspectives

3: Philosophy of Mind and Action Theory. Atascadero, Calif.: Ridgeview. Reprinted in
O’Connor (1995).
. 1998. “The Mystery of Metaphysical Freedom.” In P. van Inwagen and

D. Zimmerman, eds., Metaphysics: The Big Questions. Oxford: Blackwell.
Determinism, Freedom, and Agency
1 Determinism: What We Have Learned and What We Still
Don’t Know
John Earman
1 Introduction
The purpose of this essay is to give a brief survey of the implications of the
theories of modern physics for the doctrine of determinism. The survey
will reveal a curious feature of determinism: in some respects it is fragile, re-
quiring a number of enabling assumptions to give it a fighting chance; but
in other respects it is quite robust and very difficult to kill. The survey will
also aim to show that, apart from its own intrinsic interest, determinism is
an excellent device for probing the foundations of classical, relativistic, and
quantum physics.
The survey is conducted under three major presuppositions. First, I take
a realistic attitude toward scientific theories in that I assume that to give an
interpretation of a theory is, at a minimum, to specify what the world
would have to be like in order for the theory to be true. But we will see that
the demand for a deterministic interpretation of a theory can force us to
abandon a naively realistic reading of the theory. Second, I reject the “no
laws” view of science and assume that the field equations or laws of motion
of the most fundamental theories of current physics represent science’s
best guesses as to the form of the basic laws of nature. Third, I take deter-
minism to be an ontological doctrine, a doctrine about the temporal evo-
lution of the world. This ontological doctrine must not be confused with
predictability, which is an epistemological doctrine, the failure of which
need not entail a failure of determinism. From time to time I will comment
on ways in which predictability can fail in a deterministic setting. Finally,
my survey will concentrate on the Laplacian variety of determinism
according to which the instantaneous state of the world at any time
uniquely determines the state at any other time.
22 J. Earman
The plan of the survey is as follows. Section 2 illustrates the fragility of

determinism by means of a Zeno-type example. Then sections 3 and 4 sur-
vey successively the fortunes of determinism in the Newtonian and the
special relativistic settings. The implications of ordinary nonrelativistic
quantum mechanics and relativistic quantum field theory for determinism
are taken up on section 5. Determinism in classical general relativistic
physics is discussed in section 6. Section 7 contains some, necessarily spec-
ulative, comments on how determinism may fare in a quantum theory of
gravity. Conclusions are presented in section 8.
2 Zeno’s Revenge: An Illustration of the Fragility of Determinism
Suppose that the world is populated with billiard balls. And suppose that
the laws of motion for this world consist precisely of the specifications that
when two balls collide they obey the standard laws of elastic impact and
that between collisions they move uniformly and rectilinearly. And finally
suppose that atomism is false so that billiard balls of arbitrarily small size
can exist. Then à la Zeno we can string a countably infinite number of unit
mass billiard balls in the unit interval. Assume that at t t* all the balls in
this infinite string are at rest and that coming from the right is a cue ball of
unit mass moving with unit speed (see figure 1.1a). In a unit amount of
time an infinite number of binary collisions take place, at the end of which
each ball is at rest in the original position of its left successor in the series
(see figure 1.1b).1 The time reverse of this process has all the balls initially
at rest. Then suddenly a ripple goes through the string, and the cue ball is
ejected to the right. Futuristic Laplacian determinism is violated since it is
consistent with the laws of elastic impact that the string does not self-excite
but remains quiescent for all time (Laraudogoitia 1996).2
(a)
(b)
Figure 1.1
Zeno’s revenge.
What We Have Learned and What We Still Don’t Know 23
I see no non–ad hoc way to save determinism in this setting. If we cher-

ish determinism we can only thank the Creator that he did not place us in
a world where atomism is false and where Zeno can have his revenge.
3 Determinism in Newtonian Physics
One theme that will be sounded again and again in this section is that
classical spacetimes provide unfriendly and even hostile environments for
determinism. A related theme is that determinism in classical physics is
inextricably linked to basic philosophical issues about the nature of space,
time, and motion. To illustrate the latter theme I assert that Laplacian de-
terminism implies that it cannot be the case that both (i) space is ‘ab-
solute’ in the sense that it is a ‘container’ for bodies (e.g., shifting all the
bodies in the universe one mile to the east results in a new state distinct
from the original state), and (ii) all motion is the relative motion of bod-
ies. The argument is simple. Assumption (ii) implies that only relative
particle quantities, such as relative particle positions, relative particle ve-
locities, relative particle accelerations, and so forth, and not absolute posi-
tion, velocity, acceleration, and so forth, are well-defined quantities. The
appropriate classical spacetime setting that supports this conception of
motion has three elements: planes of absolute simultaneity, which reflect
the observer-independent nature of coexistence; a metric (assumed to
be ⺕3) on the instantaneous three-spaces, which measures the spatial
distance between simultaneous events; and a time metric, which measures
the lapse of time between nonsimultaneous events.3 But if (i) is main-
tained in this spacetime setting, not even a weakened form of Laplacian
determinism can hold for particle motions. In coordinates adapted in the
natural way to the spacetime structure, the symmetries of the spacetime
have the form
x → x R(t)x a(t) (1a)
t → t t const (1b)
where R(t) is a time dependent orthogonal matrix and a(t) is an arbitrary

smooth function of time. We can choose a(t) and R(t) such that a(t) 0
R(t) for t t* but a(t) 0 or R(t) 0 for t t*. Since a symmetry of the
spacetime should be a symmetry of the laws of motion, the image under (1)
of a solution of the equations of motion should also be a solution. But for
24 J. Earman
*
t=t
particle world lines
Figure 1.2
An apparent violation of Laplacian determinism.
p ∉ D+ (Σ)
t
space invader
Σ (plane of absolute simultaneity)
Figure 1.3
Domain of dependence in classical spacetime.
the chosen forms of a(t) and R(t) the two solutions will agree for all t t*
but will disagree for t t* since the two solutions (as indicated by the solid
and the dashed world lines of figure 1.2) entail different future positions for
the particles in the container space.
To save determinism one can reject (i) and claim that the alleged viola-
tion of determinism is spurious on the grounds that once the container
view of space is rejected there is no temptation to see the dashed and solid
lines of figure 1.3 as different future histories rather than as different repre-
sentations of the same history. To make good on this point of view it would
have to be shown how to concoct deterministic and empirically adequate
equations of motion that are formulated entirely in terms of relative
particle quantities. As the history of mechanics shows, this is not an easy
row to hoe. But that is a story I don’t have time to recount here.
The alternative way to save determinism is to reject (ii) and beef up the
structure of the spacetime by adding, say, inertial structure to make well-
defined quantities like absolute acceleration. This additional structure
linearizes the transformations (1) to
x → x Rx vt c (2a)
t → t t const (2b)
where R is now a constant orthogonal matrix and v and c are constants.

The transformations (2) are, of course, the familiar Galilean transforma-
tions. In this spacetime setting (also known as neo-Newtonian spacetime)
the above construction that undermines Laplacian determinism doesn’t
work since if a transformation from (2) is the identity map for all t t* it is
the identity map for all time.
However, in neo-Newtonian spacetime other threats to determinism
arise. Consider Newtonian gravitational theory written as a field theory.
The gravitational potential ␸ is governed by the Poisson equation
2␸ 4␲␳ (3)
where ␳ is the mass density. The gravitational force acting on a massive test
body moving in the Newtonian gravitational field is proportional to ␸.
Even the weakest form of Laplacian determinism fails because if ␸ is a solu-
tion to (3), then so is ␸ ␸ f, where f(x, t) is any function linear in x. By
choosing f(x, t) to be 0 for all t t* but different from 0 for t t*, we pro-
duce solutions for which the test body feels exactly the same gravitational
force and has exactly the same motion in the past but feels different forces
and, hence, executes different motions in the future. The nondeterministic
solutions can be killed by excluding the homogeneous solutions to (3).
This exclusion amounts to a declaration that the Newtonian gravitational
field has no degrees of freedom of its own and is only an auxiliary device
for describing direct particle interactions. Let us then turn to the pure
particle description of Newtonian gravitation.
Consider a finite number of point mass particles interacting via Newton’s
1兾r 2 force law. Let’s simply ignore problems about collision singularities
by focusing on solutions that are collision-free. Nevertheless, after many
decades of work, it has been established that noncollision singularities can
occur; that is, drawing on the infinitely deep 1兾r potential well, the parti-
cles can accelerate themselves off to spatial infinity in a finite amount of
26 J. Earman
time (Xia 1992). The time reverse of such a process is an example of ‘space
invaders’, particles appearing from spatial infinity without any prior warn-
ing. To put it crudely, you can’t hope to have Laplacian determinism for
open systems, and for the type of interaction under discussion, the entire
universe is an open system.
To save determinism from this threat, three approaches can be taken. The
first, and least interesting, is to impose boundary conditions at spatial
infinity to rule out space invaders. This smacks of making determinism true
by creating a postulate of wishful thinking. The second is to maintain
the idealization of point mass particles while adding to Newton’s 1兾r 2 attrac-
tive force a short-range repulsive force which doesn’t affect predictions for
particles with large spatial separations, but which is such that the total po-
tential well is no longer infinitely deep. One would then have to show that
the combined force law gives rise to a well-posed Laplacian initial value prob-
lem. I am not aware of any results to this effect, but I see no in-principle ob-
stacles to achieving them. The third alternative is to move from point mass
particles to a particle with a finite radius, and to postulate that when two par-
ticles collide they obey, say, the laws of elastic impact. Even if it is assumed
that atomism is true in a form that excludes the Zeno examples of section 1,
this tack can run aground on at least two shoals. (a) If triple collisions occur
the result is generally underdetermined since there will be more unknowns
than there are governing equations. (b) Even if attention is restricted to bi-
nary collisions the uniqueness of solutions can fail if an infinite number of
particles are present in the universe. Turn off the gravitational interactions
of the particles and suppose that they interact only upon contact. Lanford
(1974) constructed a solution4 in which all the particles are at rest for all t t*
but for any t t* all but a finite number of particles are in motion. Thus,
the equations of motion don’t determine whether a quiescent past is to be
extended into the future by a ‘normal’ solution in which the particles con-
tinue to be quiescent or by an ‘abnormal’ solution in which the particles ap-
pear to self-excite. Determinism can be saved either by banning infinities of
particles or by imposing boundary conditions at infinity which prevent a too
rapid increase in the velocities of particles as one goes out to infinity. Either
move smacks of making the world safe for determinism by fiat. A more in-
teresting saving move would be to show that analogues to Lanford’s solution
cannot be constructed if elastic collisions are replaced by a smooth short-
range repulsive force. I am not aware of any results to this effect.
Let us now leave particles to consider pure field theories. A familiar field
equation in Newtonian physics is the Fourier heat equation
⭸
2 ␬ ⭸t (4)
where ␬ is the coefficient of heat conduction. That (4) fails to be invariant

under the Galilean transformations (2) is no cause for concern since the
-field is supposed to be a property of a medium, and this medium picks
out a distinguished rest frame. What is of concern is that disturbances in
the -field are propagated infinitely fast. As a result, (4) admits smooth so-
lutions * with the by now familiar determinism-wrecking property that
* 0 for all t t* but is non-zero for t t*—the field theoretic version of
space-invading particles. Because of the linearity of (4), if is a solution, so
is *. Once again determinism can be saved by paring down the set of
solutions by imposing boundary conditions at infinity. Perhaps a more in-
teresting move is to declare that heat is nothing but the kinetic energy of
molecules and, thus, that the indeterminism in (4) is not disturbing since
(4) is not a fundamental law. The fate of determinism then reverts to prop-
erties of the fundamental laws, which are taken to be the laws governing
particle motion. Even if one is inclined to follow this line, it should not be
allowed to disguise the disturbing point that the Newtonian setting is in-
imical to deterministic field laws. The natural language for formulating
laws governing field quantities is that of partial differential equations. But
the type of pde that admits existence and uniqueness for a Laplacian initial
value problem are of the hyperbolic type, and hyperbolic pdes require a
null cone structure for spacetime that exists naturally in the relativistic set-
ting but can only be artificially introduced in the Newtonian setting.
Although it is outside of my main focus, I will close this section with a
few comments on epistemological matters. Ontological determinism is
compatible with sensitive dependence on initial conditions. When mea-
surement procedures for ascertaining the values of state variables are not
error-free—as would seem to be the case for any actual measurement—
sensitive dependence on initial conditions means that the link between
determinism and prediction is weakened and even broken. Furthermore,
a strong form of sensitive dependence on initial conditions (positive
Liapunov components) plus the compactness of phase space implies
‘chaos’ in the form of higher-order ergodic properties, such as the Bernoulli
property (see Belot and Earman 1997), which means that on a macroscopic
28 J. Earman
scale a deterministic system can behave in a seemingly random and sto-

chastic fashion. This raises the issue of whether and how critters such as us
can be justified in believing that the stochastic behavior we are observing is
due to indeterminism in the form of an irreducibly stochastic element or to
deterministic chaos (see Suppes 1993).
4 Determinism in Special Relativistic Classical Physics
The prospects for determinism brighten considerably when we leave classi-

cal spacetimes for Minkowski spacetime, the spacetime setting for special
relativistic theories. The combination of the null cone structure for
Minkowski spacetime plus the prohibition of superluminal propagation kills
space invaders and solves in a non–ad hoc way the open systems problem.
Fudge-free examples of Laplacian determinism—no boundary conditions at
infinity needed—are now possible. Indeed, physicists are so convinced that
Laplacian determinism is the norm in this setting that they sometimes use it
to draw a distinction between fundamental and nonfundamental fields. A
‘fundamental field’ (such as the source-free electromagnetic field and the
scalar Klein–Gordon field) is one whose field equations (respectively, the
source-free Maxwell equations and the massive Klein–Gordon equations)
have a Laplacian initial value problem that admits global existence and
uniqueness proofs: given the appropriate initial data on a Cauchy surface (a
spacelike hypersurface which intersects each timelike curve without an end
point) there exists one and only one global solution (i.e., a solution valid for
all of Minkowski spacetime) whose restriction to the chosen Cauchy surface
agrees with the given initial data. The failure of field equations to admit even
local existence and uniqueness proofs is taken as an indication of an incom-
pleteness of description. The failure of global existence and uniqueness—
say, because the solutions develop singularities after a finite amount of
time—is taken as an indication of an illicit idealization of description.
A feel for why the relativistic setting is friendlier to Laplacian determin-
ism than the Newtonian setting can be gained by introducing the notion of
the domain of dependence D(兺) of some subset 兺傺 M of points 兺 of the
spacetime manifold M. The future (respectively, past) domain of depen-
dence D(兺) (respectively, D(兺)) is defined as the set of all p 僆 M such that
every past (respectively, future) inextendible causal curve through p
intersects 兺. The total domain of dependence D(兺) of 兺 is then defined as
D(兺) 艛 D(兺). The idea is that, if the laws of physics are cooperative, the
state on 兺 suffices to fix the state throughout D(兺); but for a point q 僆 D(兺)
it is hopeless to try to determine the state at q from the state on 兺 since
events at q can be influenced by a causal process that does not register on
兺. Using D(兺) we can give a general definition of a Cauchy surface that ap-
plies to general relativistic spacetimes (to be discussed in section 6) as well
as to Minkowski spacetime and also—with suitable allowances—to classical
spacetimes: it is a global time slice 兺 (i.e., a spacelike hypersurface without
edges) such that 兺 is achronal (i.e., is not intersected more than once by
any future directed timelike curve) and such that D(兺) is the entire space-
time M. The t const level surfaces of any inertial time t for Minkowski
spacetime are, of course, Cauchy surfaces. But the level surfaces of absolute
simultaneity of Newtonian spacetime are not Cauchy (see figure 1.3); indeed,
in this case D(兺) 兺 for 兺 as t const.
If determinism fares better in Minkowski spacetime, prediction certainly
does not. The basic problem is that the very null cone structure that helps
to make the special relativistic setting friendly to determinism makes it im-
possible to acquire the information needed for a prediction prior to the oc-
currence of the events to be predicted. To make this precise a few additional
definitions are needed. Define the causal (respectively, chronological ) past of
a point p 僆 M, J(p) (respectively, I(p)) to be the set of all q 僆 M such that
there is a future-directed causal (respectively, timelike) curve from q to p.
Then for a point p 僆 M take the domain of prediction P(p) of p to be the set
of all q 僆 M such that (i) every past inextendible causal curve through q en-
ters J(p), and (ii) I(q) 債 I(p).5 Condition (i) is needed to assure that an ob-
server at p can, in principle, have causal access to all the processes that can
influence the events at q, and condition (ii) is needed to assure that from
the perspective of an observer at p the events to be predicted at q have not
already occurred. The reader can now verify that for any point p of
Minkowski spacetime, P(p) .
5 Determinism in Nonrelativistic Quantum Mechanics and Relativistic

Quantum Field Theory
To illustrate how using determinism to probe the foundations of physics

can lead to interesting results, consider the following little puzzle. No field
equation for a scalar field ␺ that is first order in time and Galilean invariant
30 J. Earman
can be Laplacian deterministic. For we can choose a Galilean transforma-

tion (2) with the property that it is the identity on the slice t 0 but non-
identity for t 0. By Galilean invariance this transformation will carry a
solution of the field equation to another solution, but the new solution has
the same initial data ␺ (x, 0) as the original solution but different values for
t 0. The puzzle concerns the Schrödinger equation
⭸␺
Ĥ␺ ih៮ ⭸t (5)
where Ĥ is the Hamiltonian operator. For a particle with mass m moving in

h៮2 2
an external potential V(x), Ĥ 2m V. This equation seems to contradict
what was said above since it is first order in time, is (presumably) Galilean
invariant, and is Laplacian deterministic in that ␺(x, 0) uniquely deter-
mines ␺(x, t) for t 0. The resolution is that ␺ does not behave like a scalar
under a Galilean transformation. In fact, the transformation properties of ␺
needed to guarantee Galilean invariance of (5) implies a ‘superselection
rule’ for mass which means that states corresponding to different mass
cannot meaningfully be superposed.
In some respects quantum systems behave more deterministically and
more predictably than their classical counterparts. As an example of the
former, consider again the problem of a finite number of point mass parti-
cles interacting via Newton’s 1兾r 2 law. The quantum Hamiltonian operator
Ĥ for this system is (essentially) self-adjoint, which implies that the evolu-
tion operator Û(t) : exp(ih៮Ĥt) is unitary and is defined for all t. QM has
magically smoothed away the both the collision and noncollision singu-
larities of classical mechanics. This magic does not work for all of the sin-
gularities of classical mechanics. For example, one could try to overcome
the Lanford type singularity discussed above in section 3 for infinite
billiards by modeling the collisions of the balls with a short range repulsive
force and then quantizing. Unfortunately, there is no well-defined quan-
tum dynamics for such a system if the repulsive force is sufficiently strong
(Radin 1977).
QM also overcomes one form of unpredictability that haunts classical
mechanics since, in one sense, QM does not recognize any sensitive de-
pendence on initial conditions. From the linearity and unitarity of the time
evolution operator, it follows that 储␺1(0) ␺2(0)储储U(t)(␺1(0) ␺2(0))储
储U(t)␺1(0) U(t)␺2(0)储储(␺1(t) ␺2(t))储: in words, if at t 0 two states are
nearby, they remain nearby for all t 0 in the Hilbert space norm 储 • 储. This
simple fact has caused some consternation since it isn’t evident how, con-
sistent with the correspondence principle, chaos can emerge from QM in
some appropriate classical limit (Belot and Earman 1997).
None of the remarks so far touches the core problem of determinism in
QM. But that problem is difficult to discuss, or even to formulate, because
it is bound up with a contentious issue about the nature of quantum ob-
servables; namely, under what conditions do the quantum observables, as
represented by self-adjoint operators on a Hilbert space, take on definite
values? One answer endorsed by standard textbooks on QM is that an ob-
servable A takes the definite value a at t iff the state ␺(t) is an eigenstate of
the operator Â with eigenvalue a.6 This eigenvector-eigenvalue rule—or any
value assignment rule which says that an observable has a specified definite
value just in case the state vector has some specified characteristic—
together with deterministic Schrödinger evolution for ␺ would mean that
Laplacian determinism holds for quantum systems.7
The well-known difficulty with the eigenvector-eigenvalue value assign-
ment rule is that it implies the unacceptable result that, after an interaction
with an object system, the pointer on the dial of a measuring instrument
that couples to the observable whose value for the object system we wish to
ascertain has no definite value if the initial state of the object system was
a superposition of eigenstates of the observable being measured. One at-
tempt to overcome this embarrassment involves ‘state vector reduction’: at
some juncture during the measurement process, Schrödinger evolution
ceases and the state vector jumps into a simultaneous eigenstate of the
object observable and the macro pointer position observable that serves
as an indicator of the value of the object observable. The violation of the
Schrödinger equation is, of course, a violation of Laplacian determinism;
but statistical determinism is maintained by the assumption that the
propensity of the state vector to collapse into a given eigenstate is governed
by the Born rule probability calculated from the state vector just before
collapse.
The collapse solution to the measurement problem comes at a high cost.
The original theory with an embarrassing consequence has been replaced
by a nontheory: the objection is not simply that the state vector reduction
is a miracle—a violation of the Schrödinger equation—but that since ‘mea-
surement’ is a term of art whose application is not specified by the theory,
exactly when and under what circumstances the reduction takes place is
32 J. Earman
left dangling. Some physicists have proposed to bridge this gap by describ-
ing a mechanism for state vector reduction by means of a nonlinear wave
equation or a stochastic differential equation (see Ghirardi et al. 1986 and
Pearle 1989). The extant state vector reduction schemes are nondetermin-
istic; but it is an open question whether they must have this property.
The major alternative route to a solution of the measurement problem in-
volves a modification of the eigenvector-eigenvalue value assignment rule
rather than of the state vector dynamics. The different paths that branch off
this no-collapse route can lead to quite different conclusions about deter-
minism, as is illustrated by the Bohm interpretation, the modal interpreta-
tions, and the many worlds/many minds interpretations. According to the
first, particles always have definite positions, and these positions evolve
deterministically via an equation of motion that is parasitic on the
Schrödinger evolution of the state vector. Assuming that all laboratory mea-
surements can be reduced to recording positions, the Bohm interpretation
offers a deterministic explanation of how and when other quantum me-
chanical observables, such as spin, take on definite values. It is shown that
the orthodox quantum mechanical probabilities are recovered for all times
on the assumption that the initial probability distribution at t 0 over par-
ticle positions q is equal to the quantum mechanical prescription 冷␺ (q, 0)冷2.
By contrast, modal interpretations don’t seek to preserve determinism but
settle for the more modest aim of breaking the eigenvector-eigenvalue link
open wide enough to make sure that measurements have definite outcomes.
This aim is accomplished by a value assignment rule which makes use of a
special decomposition of the vector for composite systems (the ‘biorthogo-
nal decomposition’). This decomposition identifies a privileged class of
observables which are said to possess definite values without saying which
particular values are possessed. Some modal theorists want to add to their
value assignment rule a dynamics for possessed values. There is a plethora of
such dynamical schemes (see Dickson 1998). Typically these schemes are not
deterministic. However, a recent proposal by Ax and Kochen (1999) restores
determinism by construing the phase of the state vector as an additional
hidden parameter—the different unit vectors in a ray of Hilbert space are
taken to correspond to different members of an ensemble, and the apparent
indeterminism of QM is due to a random distribution of initial phases.
Many worlds interpretations come in two main versions: the literal ver-
sion according to which a measurement event corresponds to a physical
splitting of the world observer (qua physical object), and the figurative
version according to which it is not the physical world but the mind or
mental state of the observer that splits. On the literal version determinism
fails if there is a fact of the matter about which postmeasurement observer
(qua physical object) is identical with the premeasurement observer. By the
same token, measurement is not an indeterministic process if genidentity
of observers is denied. A parallel conclusion holds for the many minds ver-
sion, with minds in place of bodies. It is hard to take any of this seriously,
not so much because of the metaphysical extravagances, but because it is
not at all evident that the many worlds/many minds interpretation does re-
solve the measurement problem: it doesn’t explain exactly when and under
what circumstances the splitting takes place, and it doesn’t explain why the
splitting takes place in some bases of the Hilbert space but not others.
There is no easy way to summarize the status of determinism in QM, but
two points need to be underscored. The proponents of the Bohm interpreta-
tion claim that it recovers all quantum mechanical predictions, statistical and
nonstatistical, about experiments. If this is correct, it means that in a world of
ordinary nonrelativistic QM the fate of Laplacian determinism would have to
be decided by nonempirical factors. The second remark concerns the claim,
sometimes heard, that determinism is defeated by various no-go results for
hidden variables—for example, those of the Kochen–Specker type and those
that flow from the Bell inequalities. What these no-go results show is that,
consistent with certain plausible mathematical restrictions on value assign-
ments and/or with the statistical predictions of QM, some set of quantum
observables cannot be assigned simultaneously definite values lying in the
spectra of these observables. By itself this hardly defeats Laplacian determin-
ism; nor is it even a terribly surprising conclusion—after all, many classical
quantities are best construed as having a dispositional character in that they
only take on definite values in limited contexts. That it is impossible to view
the values of quantum observables as supervening on more fundamental
quantities whose temporal evolution is deterministic would require a differ-
ent kind of proof, a proof whose existence is highly dubious in view of the
example of the Bohm interpretation.
We saw that moving classical mechanics from the Newtonian setting to
the special relativistic setting improves the fortunes of determinism. The
same cannot be said for quantum mechanics; for the fortunes of deter-
minism in QM are bound by the knots of the measurement problem and
34 J. Earman
the value assignment problem, and in the relativistic setting these knots are
drawn even tighter. State vector reduction is supposed to take place instan-
taneously. Transposing this notion to the relativistic setting seems to
require either breaking Lorentz invariance or relativizing states to spacelike
hyperplanes (Fleming 1996), neither of which is an attractive option. The
relativistic setting is also unkind to the modal interpretation because it is
difficult to reconcile the value assignment rule of this interpretation with
Lorentz invariance (Dickson and Clifton 1998). Nor does the Bohm inter-
pretation fit comfortably with relativistic quantum field theory (QFT) since
the ontology of this theory is best construed not in terms of particles but in
terms of local field observables. The general idea of Bohmian dynamics can
be applied to observables other than particle position. But in the absence of
a demonstration that in measurement contexts the field observables chosen
for Bohmian treatment will exhibit the localization needed to explain mea-
surement outcomes, one of the major reasons for finding the Bohm inter-
pretation attractive has been lost (see Saunders 1999). Furthermore, when
QFT is done on curved spacetime the dynamics for the quantum field may
not be unitarily implementable (see Arageorgis et al. 2002), making it harder
to construct a field theoretic analogue of the Bohm dynamics for particle
position.
6 Determinism in Classical General Relativistic Physics
In Einstein’s general theory of relativity (GTR) we have to deal not with one
spacetime setting but many since different solutions to the Einstein gravi-
tational field equations (EFE)
1
Rab 2 Rgab ␬Tab (6)
give different spacetime structures. (Here gab is a Lorentz signature metric,

Rab is the Ricci tensor [which is constructed from gab and its derivatives],
R : Tr(Rab) is the Ricci curvature scalar, and Tab is the stress-energy tensor
[which describes the distribution of the matter-energy].) A model of GTR
is a triple M, gab, Tab, where M is a four-dimensional manifold (without
boundary) and the metric gab is defined on all of M. A dynamically possible
model is one that satisfies (6) for all points of M. (Additionally, one may
want to impose on Tab various so-called energy conditions that guarantee,
for example, nonnegative energy densities.)
null cones
t
timelike curve
Σ
Figure 1.4
Behavior of null cones in anti–de Sitter spacetime.
To first approximation we can take a spacetime M, gab of a dynamically

possible model as a fixed background on which a test field propagates. And
we can ask whether this propagation exhibits Laplacian determinism. For
some spacetimes coming from dynamically possible models of GTR the
issue of global Laplacian determinism for a test field cannot even be stated
since there may not exist any global time slices, as is the case with Gödel
spacetime. In other cases the issue can be posed but has a negative answer be-
cause the spacetime admits global time slices but none of them is a Cauchy
surface, as illustrated by the (covering space) of anti–de Sitter spacetime (see
figure 1.4) where, in effect, the space invaders of Newtonian physics have re-
turned in a general relativistic guise. Here again we have an illustration of
the moral that determinism succeeds only with a little—or a lot—of help
from its friends. Here the friends must find a non-question-begging way to
exclude such pathologies from the causal structure of general relativistic
spacetimes.
Treating spacetime as a fixed backdrop against which nongravitational
physics takes place is inimical to the spirit of GTR which implies that space-
time structure is dynamical. Strictly speaking, the evolution of the test field
has to be treated as part of the general problem of the coevolution of the
metric and matter fields as governed by the coupled Einstein matter field
equations. For the sake of simplicity I will initially concentrate on the
initial value problem for the source-free (Tab ⬅ 0) EFE
1
Rab 2 Rgab 0 (7)
Specifying the metric field and its normal derivative on some space-
like slice 兺 does not suffice to determine, via (7), the values of the field at
36 J. Earman
points of M to the future or the past of 兺. Indeed, specifying gab on 兺 and

the entire causal past of 兺 does not suffice to determine gab at points to the
future. For if M, gab is a solution to the (7) and d : M → M is any diffeomor-
phism of M, then M, d*gab is also a solution.8 By now you know the trick:
choose d such that d id for all points of M on or to the past of 兺 but d
id for points of M to the future of 兺. Thus we get two solutions of (7) such
that the metric fields are the same in the past but different in the future.9
Determinism can be saved by requiring of it only that the values of gen-
uine physical magnitudes or ‘observables’ be determined, via the field
equations, from initial data. What then are the observables of GTR? If we
have faith in determinism we can close the circle and require that whatever
counts as an observable be such that its future and past values be deter-
mined by appropriate initial data (Bergmann 1961). From the above con-
struction we see that the observables of GTR must be diffeomorphically
invariant quantities.10 This means that none of the familiar local field
quantities—such as the metric field or scalar fields formed from the metric
field—are observables. And in some situations, such as solutions with com-
pact time slices, not even quasi-local quantities such as the integral of the
Ricci curvature over a time slice are observables.
Here I want to emphasize the contrast between the Newtonian and gen-
eral relativistic cases. In the Newtonian case it was possible to save deter-
minism from the kind of threat under discussion either by abandoning
the notion that the spacetime manifold is a container for events or alter-
natively by retaining this notion and adding sufficient absolute back-
ground structure to the spacetime. But in the general relativistic setting
that eschews absolute objects, there is no choice; making GTR a safe
haven for Laplacian determinism necessitates a radical revision of the sur-
face level ontology and ideology of GTR as a theory of tensor fields on
manifolds. One of the disconcerting features of the revision is that it
seems to entail a completely frozen block universe—not only is there no
shifting nowness, but there is a total absence of ordinary change since the
values of observables of GTR do not change with time (see my 2002). So
as not to sidetrack the discussion into these interesting but controversial
matters, I will continue to discuss determinism in GTR in terms of the
vocabulary of the surface level structure, which means that the statement
of uniqueness results for the initial value problem will have to contain an
escape clause of ‘up to a diffeomorphism’.
Suppose then we are given a three manifold 兺 and initial data on 兺 for
the source-free gravitational field. Does this data fix an appropriately
unique solution to the source-free field equations (7)? The answer is yes, at
least locally: there exists a unique (up to diffeomorphism) maximal devel-
opment M, gab of the initial data for which 兺 is a Cauchy surface; that is, M,
gab cannot be extended as a solution to the source-free equations in any
way that keeps 兺 a Cauchy surface.11
Extending this local result, which gives no information about the size
of the maximal solution for which 兺 is a Cauchy surface, to a global result—
especially to the very strong global result that requires the unique maximal
development for which the initial value hypersurface is a Cauchy surface
to be maximal simpliciter—can run into various kinds of problems. First,
we might have made a poor choice of the initial value hypersurface; for
instance, we might have chosen 兺 to be the spacelike hyperboloid of
Minkowski spacetime pictured in figure 1.5. Obviously the maximal devel-
opment of the initial data, induced on 兺 by the Minkowski metric, for
which 兺 is a Cauchy surface—namely D(兺)—can be properly extended as a
solution of the source-free field equations. Second, and more interestingly,
there might be no good choice for the initial value surface, as in the case of
an inextendible spacetime, which is a solution of (7) and does not possesses
any global Cauchy surfaces. This disturbing possibility can arise for two rea-
sons. (a) The causal structure might be like that illustrated in figure 1.4. (b) A
singularity might develop in a finite time from the regular initial data, as is
indicated schematically in figure 1.6. Singularities can arise, of course, in
the Newtonian and special relativistic settings, but because of the existence
null
surfaces
D+ (Σ)
Figure 1.5
A bad choice of initial value hypersurface.
38 J. Earman
singularity
D+(Σ)
Σ
Figure 1.6
A naked singularity.
of a fixed background metric these singularities are easily characterizable—

for example, they occur at regions of spacetime where some physical field
‘blows up’ or becomes discontinuous. But in the general relativistic setting
there is no fixed background metric, and the singularities at issue are singu-
larities in the metric structure of spacetime itself. Since by definition a
spacetime is a pair M, gab where gab is defined at all points of M, we cannot
think of singularities as objects that have spacetime locations. Attempts
have been made to characterize spacetime singularities in GTR as boundary
points that are attached to the manifold M; but these constructions all seem
to involve counterintuitive features—for example, the boundary points
may not be Hausdorff separated from interior points. And to make things
even more complicated, the singularities that lead to a breakdown of
Laplacian determinism need not be of the intuitive sort involving, say, a
‘blow up’ or wild oscillation of curvature scalars. In fact, one could simply
dub any spacetime pathology which prevents moving from the local to
global versions of Laplacian determinism in GTR a ‘naked singularity’. Then
the task would be to classify and characterize these pathologies.12
Roger Penrose’s cosmic censorship conjecture surmises that the patholo-
gies of naked singularities are not as bad in practice as they might seem in
principle. Cosmic censorship comes in two versions, weak and strong. The
weak version asserts that, generically, if naked singularities do develop
from the gravitational collapse of suitable matter fields, then they will be
contained inside black holes so that those fortunate observers who remain
outside the black hole event horizon cannot ‘see’ the singularity and are,
blackhole singularity
interior
ℑ+
ι0
horizon
collapsing
matter
r=0
(a)
singularity
ℑ+
horizon ι0
collapsing
matter
r=0
(b)
Figure 1.7
Weak and strong cosmic censorship in black hole formation.
thus shielded by the one-way causal membrane of the horizon from what-
ever nondeterministic effects emanate from the singularity.13 The strong
version of cosmic censorship conjectures that, generically, naked singulari-
ties do not develop from suitable matter fields, so that even the unfortu-
nate observers who fall into a black hole do not ‘see’ the singularity. The
strong and weak versions are illustrated respectively in figures 1.7a and
1.7b, which represent black hole formation in spherical gravitational
collapse.14 The weasel phrases ‘generically’ and ‘suitable matter fields’ are
essential to warding off potential counterexamples. For instance, the weak
version of cosmic censorship can be violated using collapsing dust matter.
40 J. Earman
But it seems wrong to lay the blame for the resulting naked singularity on
the process of gravitational collapse itself since, even in Minkowski space-
time singularities in dust matter fields can develop from regular initial data.
Thus, ‘suitable’ matter fields should be restricted to the ‘fundamental’
matter fields for which global existence and uniqueness properties hold in
Minkowski spacetime (see section 4). But this restriction is not by itself
enough to save weak cosmic censorship since special initial configurations
of the Klein–Gordon field (certainly a fundamental field) can lead to naked
singularities in gravitational collapse. Making precise the notion of non-
special or generic initial conditions would require either a suitable measure
on the space of solutions to Einstein’s equations (in which case, generic
except for a set of measure zero) or a suitable topology (in which case,
generic complement of a set whose closure has an empty interior). Since
it is not clear what the relevant measure or topology is, the cosmic censor-
ship conjecture is also unclear. Despite these difficulties however, a grow-
ing body of inductive evidence—for example, from the stability of black
holes—suggests that some interesting form of weak cosmic censorship is
probably true (Wald 1998). Opinion on the strong cosmic censorship
conjecture is more mixed, in part because it is harder to state in a form that
is both general and free of obvious counterexamples. Proving a general
form of either the weak or strong version seems to be beyond the capa-
bilities of present techniques for investigating the global existence and
uniqueness of the solutions to nonlinear pdes.
Supposing that cosmic censorship fails, it remains unclear how the for-
tunes of determinism are affected. Classical GTR places no restrictions on
what nondeterministic influences can emerge from the naked singularities.
But this does not mean that anything goes. If we are optimistic, we can
suppose that even at the prequantum level there are additional lawlike
regularities, beyond those codified in Einstein GTR, that govern these
influences. And a quantum theory of gravity may add further restrictions
or, even better, get rid of the singularities altogether.
In closing this section I will note that while the fortunes of determinism
becomes more perilous in passing from special to general relativistic
physics, the fortunes of prediction perk up. For example, in various general
relativistic spacetimes it is possible to have nonempty domains of predic-
tion; indeed, it is possible for there to be points p 僆 M such that P(p) M
(see Hogarth 1993).
7 Implications of Quantum Gravity
Among the challenges facing physics in the twenty-first century, that of

producing a quantum theory of gravity is one of the most important and
the most difficult. At the present moment there are two leading approaches:
the loop formulation, which is a version of the canonical quantization pro-
gram that aims to produce a quantum theory of gravity by quantizing GTR,
and string theory, which does not start with classical GTR but seeks to
explain it as an outcome of some low-energy limit of vibrating string or
branes. Both programs hold out the hope that the singularities of classical
GTR will be smoothed away, which would be a boon for determinism. This
hope is fueled by the ability of the quantum to smooth away classical sin-
gularities. In section 5 we saw a demonstration of this ability at work in or-
dinary QM which smooths away some of the collision and noncollision
singularities of Newtonian point mass mechanics. Another demonstration
comes from the fact that quantum particles shot at the timelike singularities
(which violate cosmic censorship) of some general relativistic spacetimes
have a well-defined temporal evolution (Horowitz and Marolf 1995). How-
ever, there is an argument to the effect that we shouldn’t hope that quan-
tum gravity will smooth away all of the singularities of classical GTR; for if,
for instance, a quantum theory gravity smoothed away the singularity of
the negative mass Schwarzschild solution, the theory would admit a non-
singular solution with an unstable ground state, which is a physical disaster
(Horowitz and Myers 1995). String theorists can remain optimistic on the
basis of the fact that their theory must entail a stable ground state; thus,
insofar as string theory smooths away classical singularities, it must also
contain a mechanism that excludes solutions such as a smoothed version
of the negative mass Schwarzschild solution (Johnson et al. 2000).
In the absence of a completed theory of quantum gravity, some insight
into what may result from combining GTR and QM can be obtained from at-
tempting to do QFT on a curved spacetime background. In even the simplest
case of a linear scalar field, determinism is crucial to the attempt: the stan-
dard construction of the algebra of field observables relies on the determin-
istic character of (real) solutions to the Klein–Gordon equation, which holds
when the spacetime background is globally hyperbolic ( admits a Cauchy
surface). When global hyperbolicity fails, there is no accepted procedure for
constructing the field algebra, and when the reason for failure of global
42 J. Earman
ℑ+
Σ2
ι0
horizon
Σ1
r=0
Figure 1.8
Black hole evaporation.
hyperbolicity is the presence of nasty causal structure—such as closed time-

like curves—there may exist no global field algebra consistent with the nat-
ural demand that the algebra, when restricted to sufficiently small globally
hyperbolic neighborhoods, should agree with the algebra obtained by ap-
plying standard construction to those neighborhoods (see Fewster 1999).
The next step up the ladder to quantum gravity involves the so-called
semiclassical approximation in which the expectation value of the (renor-
malized) stress-energy tensor for quantum fields on a curved spacetime is
inserted on the right hand side of the EFE (6), and the backreaction effects
on the spacetime metric are calculated. Hawking found that the presence of
quantum fields means that the black hole is not black but instead radiates
with a thermal spectrum and that the backreaction effect of this Hawking
radiation is to cause the black hole to evaporate (Wald 1994). If the evapo-
ration is complete, the likely outcome—insofar as it can be described in
terms of classical general relativistic spacetime structure—is a naked singu-
larity (see figure 1.8). As a consequence, a pure quantum state at the time
兺1, before black hole formation, evolves into a mixed state at the postevap-
oration time 兺2. This pure-to-mixed transition is necessarily nonunitary,
and the ‘information loss’ it involves has been the subject of a heated dis-
cussion in the physics literature (see Belot, Earman, and Ruetsche 1999 for
a review). If the pure-to-mixed transition survives in the final quantum
theory of gravity, then so does a semblance of singularities as a sink for the

missing information (Wald 1999).
8 Conclusion
One might have hoped that this survey would provide an answer to the
question: If we believe modern physics, is the world deterministic or not?
But there is no simple and clean answer. The theories of modern physics
paint many different and seemingly incommensurable pictures of the
world; not only is there no unified theory of physics, there is not even agree-
ment on the best route to getting one. And even within a particular theory—
say, QM or GTR—there is no clear verdict. This is a reflection of the fact that
determinism is bound up with some of the most important unresolved foun-
dations problems for these theories. While this linkage makes for frustration
if one is in search of a quick and neat answer to the above question, it also
makes determinism an exciting topic for the philosophy of science.
Acknowledgment
I am grateful to Rob Clifton, John Norton, and Laura Ruetsche for a num-
ber of helpful suggestions.
Notes
1. There is, of course, a violation of conservation of energy and momentum in the over-
all process even though both energy and momentum are conserved in each collision.
2. This example also illustrates a violation of even the weakest version of Laplacian
determinism according to which the future state of the world is determined by its
entire past history.
3. For more details on various classical spacetime structures, see my (1992).
4. Solution in the sense that only binary collisions occur and each such collision
obeys the laws of elastic impact.
5. This is a slight modification of the definition given in Geroch (1977). Other defin-
itions of prediction for general relativistic spacetimes are studied in Hogarth (1993).
6. This rule implies that an observable with a pure continuous spectrum never takes
on a definite value. This awkwardness can be overcome by talking about approxi-
mate eigenstates.
44 J. Earman
7. At least on the reasonable understanding that the failure of the theory to

determine the nonexistent value of an observable does not represent a failure of
determinism.
8. A diffeomorphism of M is a one-one map of M onto itself that preserves the

differentiable structure (e.g., if M is a C manifold, the mapping must be C ). d*gab
denotes the dragging along of gab by d.
9. This is a version of the notorious ‘hole argument’ which led Einstein to abandon
his search for generally covariant field equations from 1913 until late 1915; see
Norton (1987).
10. In 1916 Einstein took the ‘observables’ of GTR (although he did not use this
term) to be spatiotemporal coincidences, such as the intersection of two light rays
(see Einstein 1916 and Howard 1999). In order to be adequate to the content of GTR,
Einstein’s notion of coincidence observables has to be extended to fields.
11. M, gab is said to be a (proper) extension of M, gab iff the latter can be isometri-
cally embedded as a (proper) subset of the former. For the details of the initial value
problem in GTR, see Wald (1984).
12. For a more detailed discussion of spacetime singularities, as well as the cosmic
censorship hypothesis (introduced below), see my (1995).
13. Being visible is not sufficient to make a singularity ‘naked’ in the relevant sense;
for example, the initial big bang singularity in Friedmann-Robertson-Walker space-
times is not ‘naked’ because all of these spacetimes possess Cauchy surfaces.
14. Figures 1.7 and 1.8 use the conventions of conformal diagrams, which preserve
causal relations but distort metrical relations in order to bring infinities in to a finite
distance; namely, null directions lie at 45° with respect to the vertical; ℑ stands for
future null infinity, the terminus of outgoing light rays; and ␫o stands for spacelike
infinity. The center of the spherical symmetry is labeled by r 0.
References
Arageorgis, A., J., Earman, and L. Ruetsche. 2002. “Weyling the Time Away: The
Aon-unitary Implementability of Quantum Field Dynamics on Curved Spacetime.”
Studies in History and Philosophy of Modern Physics 33: 151–184.
Ax, J., and S. Kochen. 1999. “Extension of Quantum Mechanics to Individual

Systems.” Quant-ph/9905077.
Belot, G. and J. Earman. 1997. “Chaos Out of Order: Quantum Mechanics, the Cor-
respondence Principle and Chaos.” Studies in History and Philosophy of Modern Physics
28: 147–182.
Belot, G., J. Earman, and L. Ruetsche. 1999. “The Hawking Information Loss
Paradox: The Anatomy of a Controversy.” British Journal for the Philosophy of Science
50: 189–229.
Bergmann, P. G. 1961. “Observables in General Relativity.” Reviews of Modern Physics

33: 510–514.
Dickson, M. 1998. “On the Plurality of Dynamics.” In J. Hellman and R. Healey, eds.
Quantum Measurement: Beyond Paradox. Minneapolis, Minn.: University of Minnesota
Press.
Dickson, M. and R. Clifton. 1998. “Lorentz Invariance in Modal Interpretations.”

In D. Dieks and P. Vermaas, eds. The Modal Interpretation of Quantum Mechanics.
Dordrecht: Kluwer.
Earman, J. 1992. World Enough and Spacetime: Absolute vs. Relational Theories of Space
and Time. Cambridge, Mass.: MIT Press.
Earman, J. 1995. Bangs, Crunches, Whimpers, and Shrieks: Singularities and Acausalities
in Relativistic Spacetimes. Cambridge, Mass.: MIT Press.
Earman, J. 2002. “Thoroughly Modern McTaggart: Or What McTaggart Would Have

Said if He Had Learned the General Theory of Relativity.” Philosophers’ Imprint.
Http://www.philosophersimprint.org.
Einstein, A. 1916. “Die Grundlage der allgemeinen Relativitätstheorie.” Annalen der

Physik 49: 769–822. English translation from W. Perrett and G. B. Jeffrey eds. The
Principle of Relativity. New York: Dover, 1952.
Fewster, C. J. 1999. “Bisolutions to the Klein–Gordon Equation and Quantum Field

Theory on 2-Dimensional Cylinder Spacetimes.” Classical and Quantum Gravity 16:
769–788.
Fleming, G. N. 1996. “Just How Radical Is Hyperplane Dependence?” in R. Clifton

ed. Perspectives on Quantum Reality. Dordrecht: Kluwer Academic.
Geroch, R. 1977. “Prediction in General Relativity.” In J. Earman, C. Glymour, and

J. Stachel eds. Foundations of Space-Time Theories, Minnesota Studies in the Philosophy
of Science 8. Minneapolis, Minn.: University of Minnesota Press.
Ghirardi, G. C., T. Rimini, and T. Weber. 1986. “Unified Dynamics for Microscopic
and Macroscopic Systems.” Physical Review D 34: 470–491.
Hogarth, M. 1993. “Predicting the Future in Relativistic Spacetimes.” Studies in History

and Philosophy of Science 24: 721–739.
Horowitz, G. T., and D. Marolf. 1995. “Quantum Probes of Spacetime Singularities.”

Physical Review D 52: 5670–5675.
46 J. Earman
Horowitz, G. T., and R. Myers. 1995. “The Value of Singularities.” General Relativity
and Gravitation 27: 915–919.
Howard, D. 1999. “Point Coincidences and Pointer Coincidences: Einstein and the
Invariant Content of Space-Time Theories.” In H. Goenner, J. Renn, J. Ritter, and
T. Sauer, eds. The Expanding Worlds of General Relativity. Einstein Studies, volume 7.
Boston, Mass.: Birkhäuser.
Johnson, C. V., A. Peet, and J. Polchinski. 2000. “Gauge Theory and the Excision of
Repulsion Singularities.” Physical Review D 61: 086001.
Lanford, O. E. 1974. “Time Evolution of Large Classical Systems.” In J. Moser, ed.

Dynamical Systems, Theory and Applications. Lecture Notes in Mathematics, volume 38.
Berlin: Springer-Verlag.
Laraudogoitia, J. 1996. “A Beautiful Supertask.” Mind 105: 81–83.
Norton, J. D. 1987. “Einstein, the Hole Argument and the Reality of Space.” In
J. Forge, ed. Measurement, Realism, and Objectivity. Dordrecht: D. Reidel.
Pearle, P. 1989. “Combining Stochastic Dynamical State Vector Reduction with

Spontaneous Localization.” Physical Review A 39: 2277–2289.
Penrose, R. “Gravitational Collapse: The Role of General Relativity.” Revisita del

Nuovo Cimento. Serie 1, 1 Numero Speciale: 252–276.
Radin, C. 1977. “The Dynamical Instability of Non-Relativistic Many-Body Systems.”

Communications in Mathematical Physics 54: 69–79.
Rovelli, C. 1991. “What Is an Observable in Classical and Quantum Gravity?”

Classical and Quantum Gravity 8: 297–316.
Saunders, S. 1999. “The `Beables’ of Relativistic Pilot Wave Theory.” In J. Butterfield and
C. Pagonis, eds., From Physics to Philosophy. Cambridge: Cambridge University Press.
Suppes, P. 1993. “The Transcendental Character of Determinism.” Midwest Studies in

Philosophy, volume 18. Notre Dame, Ind.: University of Notre Dame Press.
Wald, R. M. 1984. General Relativity. Chicago: University of Chicago Press.
Wald, R. M. 1994. Quantum Field Theory in Curved Spacetime and Black Hole Thermody-
namics. Chicago, Ill.: University of Chicago Press.
Wald, R. M. 1998. “Gravitational Collapse and Cosmic Censorship.” In C. V.

Vishveshwara, B. R. Iyer, and B. Bhawal, eds., Black Holes, Gravitational Radiation, and
the Universe. Dordrecht: Kluwer Academic.
Wald, R. M. 1999. “Gravitation, Thermodynamics, and Quantum Theory.” Classical

and Quantum Gravity 16: A177–A190.
Xia, X. 1992. “The Existence of Noncollision Singularities in the N-Body Problem.”

Annals of Mathematics 135: 411–468.
2 Freedom and the Power of Preference
Keith Lehrer
It is the power of preference that leads us to think and aver that things are
up to us or in our power. Our preferences are the source of our freedom. But
we must have our preferences because we prefer to have them and not
because they are imposed on us. We have reasons for our preferences. Our
reasons for our preferences must be our reasons because we prefer them to
be. We must be agent and author of our preferences for them to be free. Our
preferences, however, seem to be part of the causal order. So how can we
have our preferences because we prefer to have them? How can things
be reasons for our preferences because we prefer that they be? How can we
be the agent and author of our preferences, when our preferences are part
of the causal order? How, on the contrary, can we have preferences because
we prefer to have them if they are not part of the causal order? When we
reflect on how preferences empower us, we understand that we must em-
power them. But how? These are the questions I shall seek to answer. The
keystone to the answer is a kind of special preference that loops back onto
our preferences including itself.
There is a minimal notion of freedom expressed by ‘can’ and ‘could
have’. It goes this way. I do something freely if and only if I could have
done otherwise. Extending the minimal notion of freedom to preference, I
prefer something freely if and only if I could have preferred otherwise. The
appropriate meaning of ‘could have’ required to articulate our freedom
cannot be assumed to be contained in antecedent usage, however. On the
contrary, we must construct the requisite meaning needed for an adequate
theory of freedom.
Let me return to a simple analysis, recognizing that this is theory
construction, not semantics. We are trying to construct, then, an analysis
of ‘can’ and ‘could have’ that will enable us to build a theory of freedom.
48 K. Lehrer
To accomplish this, I shall need a notion of preference that I have advo-

cated. You may wish to substitute some notion of preference you consider
more satisfactory. I distinguish between desire and preference (Lehrer
1997). Desire I think of as something that arises in us without our bidding
and often against our will. I suddenly desire to eat some chocolate as I am
writing. Should I stop writing and get some chocolate? I have Ghirandelli
semisweet chocolate in the kitchen. I have also eaten quite bit of this
chocolate today. Then I have a desire to look at my paintings. I am a
painter, and I like to look at my paintings. But once I do, I absorb myself in
them for a while. So here I am, trying to write, distracted by desires to eat
chocolate and look at my paintings. What a nuisance these desires are.
They suddenly appear, these uninvited distractions. Go away, I want to
write. But they do not go away. I prefer not to satisfy them, and, in fact, I
continue to write. What I prefer is up to me. What I desire is not.
I evaluate my desires and prefer to satisfy some and not others. So pref-
erence is typically the positive evaluation of something desired. But not
always. Sometimes I consider some course of action and prefer to undertake
that course of action without any desire to do so. I prefer to prepare the ma-
terials to pay my taxes, but I have no desire to do so. None whatsoever. It is
rather, as I consider the consequences of preparing or not preparing the
materials to pay my taxes, I evaluate preparing the materials as preferable
to not doing so, and I prefer to prepare them without any desire to do so.
There are many such cases of odious or burdensome obligations. Though I
prefer to do what I am obliged to do, I have no desire for it. What this re-
veals is that preference is a different kind of state than desire, though both
may influence what I do. I have proposed that preferences are states of a
different level of mentality. So, I think of desire as arising without my bid-
ding or approval, the states of a first-level mind that are not up to me, as
contrasted with preferences that are up to me. What I prefer as I evaluate
what I desire or consider, either reflectively or spontaneously, at the level of
the metamind, is where optionality and freedom enter the picture. Prefer-
ence like choice, which is the immediate expression of preference, is a
condition of our freedom.
This discourse could become fully and completely a discussion of prefer-
ence and what it is like. I have, however, discussed preference elsewhere
and so, having noted the distinction between desire and preference, it is
the application of the notion in a theory of freedom that will be my focus.
Freedom and the Power of Preference 49
Some have suggested that ‘I can do X’ might be analyzed conditionally as

‘I will do X if I prefer to do X’ and ‘I could have done X’ as ‘I would have
done X if I had preferred to do X’.1 Now, in the normal case, my preferences
will be up to me. Consequently, analyzing ‘I can’ and ‘I could have’ in this
way will yield the result that when I can do something in the way analyzed,
then it is in my power to do it. Further, if I both can do something and can
do otherwise instead, then it is up to me whether I do it. But that result is
only a material consequence, for it is possible that what I prefer is not in
my power, not up to me. Consequently, we have not conceptually insured
even a minimal doctrine of freedom with the satisfaction of the condition-
ally analyzed notion of ‘can’ and ‘could have’.
Before rushing too quickly to the conclusion that the analysis is unsatis-
factory, it is important to notice the importance and insight of it. It is the
kind of freedom that we lack when in chains or imprisoned. It is the kind
of freedom that people have fought and died for. If it is not all that there is
to freedom, it is a great deal. Moreover, since preference reflects our evalu-
ation of options, it is typically and normally an expression of choice.
Whether I choose X or not-X typically depends on what I prefer. So, in the
typical case, the conditional analysis of ‘can’ and ‘could have’ will capture
a notion of freedom in a way that a conditional analysis in terms of desires
would not. A person driven hither and thither by desire may appear more
passive than active, more controlled than empowered. But preference is
different. What we prefer reflects our evaluations of what is preferable,
desirable, good for ourselves, and, ideally, good for others. Preference amal-
gamates these evaluations in a choice disposing state.
However, even if the conditional analysis gives us most of what we value
in freedom and perhaps most of what makes us uncomfortable as we deal
with our freedom, a problem remains. Our preferences might be controlled
by another and not be in our power or up to us. We must say something
about freedom of preference as well as freedom of action based on the
conditional analysis of freedom.
I suggested in earlier work (Lehrer 1990) influenced by Frankfurt (1971),
that we iterate the conditional analysis to provide an account of freedom of
preference. Putting the matter in terms of ‘can’, we might add to the analy-
sis of ‘I can do X’ as ‘I will do X if I prefer to do X’ the further conditional
that ‘I will prefer to do X if I prefer to prefer to do X’, or to introduce a nota-
tion reflecting levels of concatenation of preference ‘I will prefer (1) to do
50 K. Lehrer
X if I prefer (2) to do X’. Thus, ‘I prefer to do X’, expresses first-level prefer-

ence, and in our level-specified notion becomes ‘I prefer (1) to X’, while ‘I
prefer to prefer to do X’ expresses second-level preference, and in our nota-
tion becomes ‘I prefer (2) to do X’. More generally, ‘I prefer (n 1) to do X’
is equivalent to ‘I prefer to prefer (n) to do X’. To avoid a problem arising
about freedom at level (2) or beyond, we might add the perfectly general
conditional, ‘I will prefer (n) to do X, if I prefer (n 1) to do X’ requiring
that it hold for all levels of n. There is no vicious regress in the condition,
since I do not have to do anything for the set of conditionals to be true.
They can all be true. Moreover, there is no problem in principle of process-
ing such higher order conditionals since, as the general conditional reveals,
to process up a level I would only need to process whether I prefer what I
preferred at the previous level, and the answer might be obviously, “Yes.”
We are capable of processing things computationally, a step at a time,
without grasping the whole sequence in a flash of intuition.
However, as some have correctly pointed out, van Inwagen and
Segerberg,2 it might be that someone controls what I prefer at all levels and
what I would prefer at all levels if my preferences were different so that my
preference at all levels are in his power and not in mine. Our detractor who
will introduce a demon or manipulator imagined to control my preference,
perhaps by inserting a device, call it, as I have, a “braino,” in my brain so
that when he types ‘pX(1)’ on his computer, I prefer (1) to do X and the
same for any other level. I have two comments. First, there are no such
manipulators. Once brainos start getting produced and installed, we shall
have a problem about our freedom that we do not have at the moment. So,
the analysis may have the virtue of being materially adequate and illumi-
nating in our current state of technology. It could reveal the way in which
our preferences can be in our power or up to us. They can be what we prefer
them to be.
However, it is important to go beyond the actual case to obtain further
illumination and security for our analysis against preference control. More-
over, the way to accomplish this is not only clear but avoids the necessity
of postulating an infinite hierarchy of preferences, even hypothetically.
What is needed for a satisfactory analysis is the satisfaction of two condi-
tions. The first is that I prefer to have the preferences I do have concerning
X however far up the preference hierarchy they might extend or not
extend. The second is that this preference to have the preferences I have
is one that I have, not because it is manipulated by another, but just because
I prefer to have it. Let us call the preferences we have at various levels con-
cerning doing X my preferences structure for X. Now consider my preference
to have my preference structure for X. Let us call that my power preference
for X. I propose as one condition of freedom of preference that I prefer X
and I have my preference structure for X because I prefer to have it.
There are special features of the power preference and of my having that
preference because I prefer to have it. First, my power preference for having
my preference structure for X is itself a preference in that preference struc-
ture because it is itself a preference concerning X. Thus, it is a preference for
having a preference structure that includes that preference. The preference
refers to itself, as it is a preference for itself as well as other preferences in
the structure. It is, therefore, an unstratified or level-ambiguous preference.
On one standard account of level-ambiguous states, they are semantically
ungrounded (Kripke 1975). Ungroundedness may be considered as a defect
in the truth conditions for a sentence, but it is not a defect of a state
intended to analyze freedom of preference.
Second, the preference ends a possible regress of preferences with a loop.
Theories of higher order preference that appear to require a complete hier-
archy of preferences are often criticized on the ground that the require-
ment is psychologically unrealistic. We have noted that if preferences over
preferences are considered computationally or functionally, it is not clear
that any difficulty in principle results from such requirements. However,
the addition of preferences past the second or third level, though it can be
processed computationally, might lack efficacy. So, there is an advantage in
not requiring a hierarchy of preferences, and use of the power preference
allows us to dispense with that requirement. The power preference ensures
that any preference that one has at any level is one that one prefers to have,
for that is what the power preference is a preference for. It allows, in prin-
ciple, that freedom of preference could result from a first-order preference
for X and the unstratified power preference for X that would, in that case,
be a preference for the first-order preference and for itself.
Moreover, the present account allows for the possibility that preferences
that insure freedom might contain conflict between levels of preference.
Frankfurt’s account and my own earlier account indebted to him require
that the preferences insuring freedom be homogeneous in the sense that
every preference in the structure be a preference for the preference of the
52 K. Lehrer
preceding level. Such accounts require a conflict-free preference structure

concerning X, while the present proposal allows for the possibility of free-
dom in spite of conflict. I might prefer at the second level not to have the
preference at the first level that I have. As I consider the resulting prefer-
ence structure with the preference for X and the preference for not having
that preference, however, I might find that I prefer to have that preference
structure, including the conflict, and thus have a power preference for
having that preference structure for X. The conflict might strike me as rep-
resenting moral or just prudential complexity. I might, for example, prefer
to spend my resources on a friend who is hardly in need rather than on a
stranger who is in acute need. After some struggle within, I might still pre-
fer to spend my money on my friend, but admiring the nobility of people
having broader humanitarian concerns prefer not to have that preference.
As I consider that preference structure, I might prefer to have it, both
because I prefer to spend money on my friend and also prefer to have
different preferences expressing a nobler perspective. I prefer to have the
preference structure that I have including the conflict at the second level
rather than repudiating the conflict for the felicity of a homogeneous pref-
erence structure. Freedom of preference is, on the account I am proposing,
compatible with conflict within the structure.
Of course, it is essential that I not only have the power preference for X
but that I have it because I prefer to have it. Otherwise, it may be that I have
the preference structure for X because I am manipulated by the braino
rather than because I prefer to have it. This requirement, which I shall
call the primacy condition, should be understood to insure that the primary
cause or explanation of the preference structure is the preference for the
preference structure. There may, of course, be secondary causes or explana-
tions. For example, suppose that you are the friend and prefer that I spend
my money on you. If that is the primary cause or explanation of why I pre-
fer to spend money on you, there is no reason to conclude that I have free-
dom of preference. I may be in your power, and my preferences as well, and
so it is in your power what I prefer, it is up to you what I prefer, and I lack
freedom. I am in your control.
However, there is another possible manner in which your preference
might influence my preferences which is compatible with my freedom of
preference. I might prefer to do what you prefer in this instance because I
prefer to have that preference structure. The primary cause or explanation
of why I prefer what I do is that I prefer to have the preference structure I

have, in this case, including a preference for what you prefer. It is not in
your power or up to you what I prefer. It is in my power or up to me what
I prefer, though, in fact, my preferences include a preference for what you
prefer. Who has the power in the matter depends on whose preferences
are the primary cause or explanation of my preferences. I am free in my
preference for what you prefer if I prefer what you prefer because I prefer
to have that preference. I am in bondage to you if I prefer what you prefer
just because you prefer what you do. In short, if your preference is effective
because I prefer that I have the preference structure I do and prefer what
you prefer, then I am free and my preferences are in my power, while if
your preference is effective just because you prefer what you do, then I
am bound by your preferences and in your power.
Once we have in mind the distinction between the case in which the
power preference is the primary cause or explanation of the preference and
the case in which it is not, we are in a position to deal with a central
issue, that of the compatibility of causation and freedom. Moore noted a
long time ago that, though there is a sense of ‘could’ that yields the result
that under the assumption of determinism nothing ever could happen
other than it did, it remains an open question whether this sense of ‘could’
is the one that is involved when, claiming that we are free, we avow that we
could have done or preferred otherwise.3 The question of the compatibility
of freedom and determinism remains open, and a sufficient condition for
freedom that is compatible with determinism can be based on what has
been said above.
To clarify the issue of compatibility, it is important to be clear about the
logic of compatibility. It will suffice for the purposes of the compatibilist to
provide a set of sufficient conditions for freedom that are compatible with
the thesis of determinism. The principle is that if c entails f and c is consis-
tent with d, then f is consistent with d. This shows that we do not need to
concern ourselves with the question of whether a condition is necessary for
freedom in order to employ the condition in an argument for the compat-
ibility of freedom and determinism. That is important for two reasons.
First, the necessary condition for freedom of action or preference might
be some rather heterogeneous disjunction of conditions that is not very
perspicuous. Second, and more important, it appears from the discussion
above that we can formulate a sufficient condition for freedom of action
54 K. Lehrer
and preference in terms of the power preference and the primacy condi-
tion, even if such a condition should prove not to be necessary.
Here, then, is a sufficient condition. Let us begin with freedom of prefer-
ence. Suppose I prefer to do something, to write this paper, for example. We
are seeking a sufficient condition for this freedom without worrying about
whether it is necessary. If we add some bells and whistles that are not really
needed for freedom of preference, that will not matter. So let us help our-
selves to a set of conditions that are clearly sufficient, that clearly insure
freedom of preference. Here are the conditions I propose for our considera-
tion. My preference for doing X is free if
1. I have a power preference for X, that is, I would prefer to have the
preference structure for X that I have.
2. The power preference for X satisfies the primacy condition, that is,
I would have the preference structure for X because I would prefer to
have it.
These conditions, though they insure that my preference is not manipu-
lated by another might not provide adequate insurance that I could have
preferred otherwise. So let us add a third condition to ensure that.
3. If I had preferred to have a preference structure with the preference to do

otherwise, I would have had that preference structure with the prefer-
ence structure to do otherwise because I would have preferred to have it.
Conditions 1 and 2 ensure that I have the preference structure that I do

because I prefer to have it, while condition 3 ensures that if I had had the
preference structure to do otherwise, I would have had that preference
structure because I preferred to have it instead. I claim that this is sufficient
for freedom of preference. I prefer to do X because I prefer to have the pref-
erence structure for X including that preference for doing X, and if I had
preferred to have a preference structure with a preference to do otherwise,
I would have had that preference because I preferred to have it. It is up to
me what preference structure I have, both actually and counterfactually,
and that suffices for my freedom of preference. Whatever preference struc-
ture I have or would have concerning doing X, I would have it because
I prefer to have it.
It should be noted that the subjunctive formulation of conditions 1 and
2 in terms of ‘would’ requires not only that the person has the preference
to do X because he prefers to have the preference structure containing this,
but also that he would have it. This treats the very influential Frankfurt
(1969) counterexamples to the principle of alternative possibilities by re-
quiring alternative possibilities for freedom of preference. In these coun-
terexamples, though a person does what he prefers, namely, to do X, and
prefers to prefer to do X and so forth, there is an inactive counterfactual
intervener with knowledge of what the person is about to prefer and the
power, which he would have exercised if necessary, to make the person pre-
fer to do X if the person were about to not prefer to do X. Such a person as
described in the Frankfurt examples who prefers to do X would not have
the preference to do X because he preferred to have that preference when
he was about to not prefer to do X and the intervener caused him to have
the preference to do X. The person would have the preference because the
counterfactual intervener would intervene.
So we have a sufficient condition for freedom of preference when what-
ever preference structure I have or would have concerning doing X is one
I would have because I prefer to have it. Suppose that I do something
because I freely prefer to do it and I would have done otherwise if I had
preferred otherwise which I would also have preferred freely. That is a
sufficient condition for freedom of action. I could have done otherwise. So
from the three conditions formulated above we obtain an account of
freedom of preference and freedom of action.
Are the three conditions offered above compatible with determinism?
They are. The reason is related to the remarks about the primacy of causation
and explanation given above. You may cause me to prefer something with-
out that destroying my freedom of preference if I prefer what you prefer
because I prefer to have that preference structure and my preference for that
preference structure is the primary cause or explanation. Similarly, some
other reason or motive may cause me to prefer something without that
destroying my freedom of preference. Suppose I prefer to respond to that
reason or motive because I prefer to have a preference structure responding
to such reasons or motives. Moreover, suppose my preference for that
preference structure is the primary cause or explanation of it. That ensures
my freedom. Furthermore, the assumption of the primacy of preference in
causation and explanation is compatible with causation and explanation.
These remarks introduce a further complication, and a necessary one in
our theory of what is sufficient for freedom. The complication concerns the
influence of reasons or motives. Any compatibilist account must include a
56 K. Lehrer
theory of how reasons can influence our preferences, indeed, cause us to

prefer one thing rather than another, without undermining our freedom of
preference. Moreover, one advantage of a compatibilist account of freedom
of preference is that it permits us to explain the function and efficacy of
reasons and motives causally rather than leaving this unexplained. The
problem is to explain the difference between causation that insures free-
dom and causation that creates bondage. The solution consists in part of a
difference in how reasons or motives cause us to prefer what we do when
we are free as opposed to when we are not.
There is a system of reasons or motives to which I am responsive in the
formation of preference. What those reasons and motives are may not be
up to me. Recall that what we desire is often not up to us, though we may
freely prefer to satisfy the desire or freely prefer not to satisfy it. In other
cases, desires may overwhelm us, and we may seek the satisfaction of
them obsessively or compulsively without doing so freely. What is the
difference? The answer should be clear from the preceding remarks. The
difference is whether we respond to the reasons we have because we prefer
to do so. We are here going beyond the power preference, a preference
for preferences for X, to a more complicated structure of preference for
responding with preference to reasons. There is a special preference, which
I shall call an ultrapreference, for the more complicated structure of respond-
ing to reasons with preference. Again, however, I may have the ultraprefer-
ence I do have for X, including a preference for responding to reasons,
because I prefer to have it.
Here, then, is the theory I am offering as sufficient to enable us to distin-
guish between causation by reasons that insures our freedom of preference
and causation that does not. Suppose I have a system of reasons SR for X.
This system may include reasons for doing X and replies to objections to
that course of action. Now suppose I have a system of reasons SR and re-
spond to that system by forming a preference for X. Notice, first of all, that
I may prefer X under the satisfaction of the three conditions of freedom
and, therefore, freely prefer X even though I form that preference in re-
sponse to SR. That is so when I respond to the reasons because I prefer to
respond to SR in that way, and I would respond otherwise if I preferred
otherwise because I preferred to respond otherwise. I have an ultrapreference
to respond to certain reasons for X by preferring X, and I have that
ultrapreference because I prefer to have it. The primacy of causation and
explanation obtained from my having the power preference and ultrapref-

erence because I prefer to have them is what insures my freedom within the
nexus of causation.
We can now say that I freely prefer X because I prefer to have the prefer-
ence structure that I have for X and because I have the reasons that I do for
preferring X. The explanation of how both can be true is given by the
ultrapreference that I have for responding to certain reasons that I have be-
cause I prefer to have it. We do not need to appeal to unexplained or mys-
terious factors to explain this kind of freedom. We do not need to suppose
that the influence of motives is something other than causal. It is causal. It
is just that the causal efficacy of the reasons depends on my preference for
responding in that way. That is what permits me to be rationally free in my
preference at the same time that my preference is caused and, indeed,
caused by the reasons that are part of the system of reasons. I respond to
system of reasons because I prefer to do so.
At this point, a question naturally arises concerning the relationship be-
tween the ultrapreference and the responsiveness to the system of reasons.
One might ask whether I have my ultrapreference because I respond to the
system of reasons I have or whether I respond to the system of reasons
I have because of my ultrapreference? The answer is that both are true. I
have my ultrapreference because I respond to the system of reasons,
and I respond to the system of reasons because of my ultrapreference. The
relationship of the system of reasons to the ultrapreference is one of mutual
causal support and dependency.
An analogy I have used before to explain the causal dependency is that
of a keystone in an arch. The arch requires the keystone to stand, for oth-
erwise it would collapse, but the other stones, constituting the system of
reasons, are also essential to the arch. Moreover, the keystone, which is the
ultrapreference, contributes to its own support by supporting the sides of
the arch that support it. It is, in a way, self-supporting but only in relation-
ship to the other stones. Finally, if there is a system of arches, correspond-
ing to structures of reasoning for preferences, they may constitute a dome
with a common circular keystone at the top. In that case, the circular char-
acter of the keystone is part of what makes it effective in supporting the
dome of rational preference and the system of reasons, the arches, which,
in turn, support it. All of this is only an analogy and a metaphor. The point
of it is to illustrate that the relation of mutual support of the system of
58 K. Lehrer
reasons and the ultrapreference can be interpreted causally without para-

dox. The notion of mutual causal support is a commonplace one.
Nevertheless, the feeling might remain that something must come first.
Either we first have a system of reasons that causes us to form the ultra-
preference or we first form the ultrapreference that creates the system of
reasons. Sartre, who speaks of choice rather than preference, thinks there
is a kind of ultrachoice that we must make without justification and
without excuse and that choice is both free and fills us with dread as we
realize there is nothing to support it.4 Though I am indebted to Sartre for
my conception of an ultrapreference, I do not agree with him that it is
without justification and without excuse. Preference, unlike mere desire,
involves evaluation in terms of reasons we have for satisfying or not satis-
fying what we desire or consider. Thus, our preferences are formed with at
least personal justification, however fallible or fallacious the justification
might be.
On the other hand, a hard determinist affirming the incompatibility of
freedom of preference and causation might argue that what reasons we
have and what preferences we form based on those, including the ultra-
preference itself, are caused by our having those reasons, or in some other
way. So the reasons, the justifiers and excusers, cause our preferences,
including our ultrapreference, and there is no room for freedom in causal
sequences that lead to the formation of the reasons and from the formation
of reasons to the preferences. Thus, the traditional dilemma. Either the rea-
sons to which we respond cause the preferences including the ultraprefer-
ences, and, therefore, are not free, or we respond to the reasons we do
because of an uncaused free preference to respond to such reasons, an un-
caused version of the ultrapreference, and our preferences are not caused.
We can, as I have suggested, exit from this dilemma.
The exit is supported by the keystone ultrapreference. We respond to the
system of reasons we have because we prefer to respond in that way, that is,
because of the ultrapreference for responding in that way. If the ultrapref-
erence contains a power preference satisfying the three conditions given
above, the ultrapreference is free. I may have it because I prefer to have it
and, if I preferred otherwise, I would prefer otherwise because I would pre-
fer to have that preference. Does that entail that my ultrapreference is itself
uncaused and also without justification? Does inserting causation and
justification destroy the freedom of the ultrapreference? Not at all. The
ultrapreference, like other preferences, is justified by my evaluation in

terms of my reasons to which I respond with preference.
But which comes first, the system of reasons supporting the ultraprefer-
ence or the ultrapreference for preferring the system of reasons? Neither.
The system of reasons supporting the ultrapreference and the ultraprefer-
ence for the system of reasons stand or fall together composing an arch of
reason capped with the keystone ultrapreference. The causal structure
of the arch explains the possibility of compatibilism. The ultrapreference is
a free preference causally supported by the reasons that justify it.
There remains a question about the temporal order of the causal process,
about the sequence of causality, and what occurs within it. Any answer to
such a question is highly speculative. My proposal is that when we choose
and decide we often simultaneously choose to make the reasons that sup-
port the choice effective as we make the choice. Of course, we may reflect
on the reasons for and against performing an action. But when we choose,
that choice may constitute, at the same time, a choice to respond to the
reasons for it. The idea that we first choose to make some reasons the
effective ones in choice and then, after that, respond to the reasons by
making the choice, though it may sometimes occur, does not seem the typ-
ical structure in choice. In effect, choosing a course of action, to write this
paper, for example, and choosing to make the reasons in favor of the choice
effective, that I committed myself to do so to a friend and former student,
for example, seem to me to occur together and to depend on each other.
I choose at the same time to make the reasons effective and to write the
paper.
Since preference to do something is a disposition to immediately choose
to do it when the situation is propitious, the preference to write the paper
and the preference to make the reasons for doing that effective are made at
the same time. This may be obscured by earlier intentions and preferences
that influence me. It remains up to me, however, to prefer to fulfill those
intentions and to sustain the preferences. So the preference to act and the
preference to make the reasons for the preference effective are simultaneous.
The two preferences may, in fact, be included in one, the one may be the
same, in a sense, as the other, but that claim takes me beyond what is
required for my purposes. I leave it before us as a conjecture for evaluation.
There remains a synchronic question. How does this structure of a system
of reasons and an ultrapreference come into existence in the first place?
60 K. Lehrer
The simple answer is that it is caused. I am not sure how. However, I think
that the rough picture is that we at a certain stage have no preferences.
Only impulses and desires, perhaps informed by unreflective beliefs and
impressions, that drive our behavior. We form patterns of action, patterns
of responding to desires and beliefs, at the level of the first-order mind un-
encumbered by the burden of ratiocination and higher-order evaluation.
Then we acquire the higher-order mind, the metamind, that allows us to
represent and think about our desires and beliefs, to prefer the satisfaction
of some desires and not others, to accept some beliefs but not others, as we
sort out the first-order states. At this point, at the point of the intervention
of the metamind, we go through a brief, somewhat confused period, of vac-
illation between mind and metamind, and, finally, we arrive at a state of
preference and acceptance, of systems of reasons supporting preferences
and preferences supporting reasons. That is my conjecture concerning the
transition from mental causation to metamental causation and the free-
dom of preference. It is not more than a conjecture.
The argument from what I have said to the conclusion of compatibilism
is brief. Combine the account of freedom of preference in terms of
the three conditions concerning the power preference stated earlier with
the account of the structure of a system of reasons and the ultrapreference
containing a power preference satisfying those conditions, and you have
sufficient conditions for freedom of preference. A sufficient condition of
freedom of action is easily obtained therefrom. Suppose I do something
because I prefer to do it when my preference is free. Suppose in addition, I
would have done otherwise if I had preferred to and that preference would
have been free. On those suppositions, my action was free. Suppose my ac-
tion and preference satisfy these sufficient conditions for freedom. Is there
anything in these conditions that entails that these structures of preference
that suffice for my freedom are uncaused? Nothing at all. Is there anything
in this account which entails that something must be undetermined, that
there must be some point at which there is no causal explanation for why
I prefer one thing and not another? There need be no such point.
If I were completely rational in my preferences, which I admit without
regret I am not, all my preferences might be responses to reasons that
causally support them. There might be causation all the way around, before
and after my preferences. But some of them would be free. When I prefer
what I do because I prefer to have the preference structure I do, including
my ultrapreference, I have all the freedom of preference that interests me.

Why should I wish, in addition, that the causal order should be undeter-
mined? It may be undetermined, as many aver, but I am indifferent in con-
sideration of my freedom, though curious for theoretical reasons, about
whether they are right.
Having the positive argument before us, let us consider the advantages and
objections to it. First let us consider the advantages, and then we shall turn
to the sticking points. The account accommodates many of the claims of
libertarians without the assumption of indeterminism. Some, Reid (1788),
Chisholm (1966), and Taylor (1966), for example, have said that freedom
requires agent causality. I must be the author or agent of my preferences if
they are free. But that claim leaves open the question of what is involved in
an agent causing her own preferences. My claim is that an agent is the cause
of her preferences when she has those preferences because she prefers to
have them.
Now, one might add, if one wishes, that the agent is a substance, and the
substance causes her preference. My suggestion is that this causation
amounts to nothing more than her having her preferences because she
prefers to have them. What does she do to cause the preferences other than
prefer to have them? Perhaps nothing. Perhaps she simply responds to the
reasons to which she prefers to respond. Moreover, if she has the prefer-
ences because she prefers to have them, then she, and not another, is the
agent or author of them. If another were the agent or author of them, then
she would not have them because she prefers to have them but because the
other prefers that she does.
Many have argued that agency is to be explained in terms of rationality,
like Kant (1993), or by identifying with the desire to act in terms of reasons,
like Velleman (1992). Though acting from reasons may contribute to my
sense of agency, it also may not, and so this identification of agency with
rationality and the desire to respond to reasons seems to me to reflect a
rationalistic bias. I may feel alienated from the life of reason when reasons
alone lead me to a course of action. Personally, I find this familiar enough.
The life of reason sometimes leaves me feeling that reasons are ruling my
life, and I am bereft of any sense of agency in responding to them. I go
where the reasons lead me feeling alienated from the process and with a di-
minished sense of agency. The problem in such cases is that my preference
for acting according to the reasons I do is not a preference that I have
62 K. Lehrer
because I prefer to have it. I have a sense of agency when I prefer what I
do because I prefer to have that preference and when the effective reasons
for the preference, if any, are effective because I prefer that they be effective.
I am in this view close to the view that Frankfurt (1971) advances when he
emphasized the importance of identifying with a preference for a sense of
agency, but I suggest that my identification with the preference is the result
of having that preference because I prefer to have it and responding to my
reasons for having the preference because I prefer to respond to them.
There is also the issue of control and power. If I am free in my prefer-
ences, then they must be in my control, in my power. Suppose, then, that
my preferences and the system of reasons that supports them are ones
I have because I prefer to have them, and, moreover, are such that if I pre-
ferred not to have them, I would have that preference because I preferred
to have it. Are my preferences in my control? I control them all right.
I have them because I prefer to have them. If I preferred not to have them,
I would have that preference because I preferred to have it. What I prefer is
in my control in the sense that my preferences, whatever they may be, are
ones that I have because I prefer to have them. I am in control. Do I have
power over my preferences? They are in my power. I have the preferences
I do because I prefer to have them. To avoid tedium, I proceed no further
in this line of argument.
Is the theory a theory of volitions? Preferences are like volitions. One prob-
lem with trying to analyze freedom in terms of volitions is that the theory
appears to lead to an infinite regress, when we try to analyze the freedom of
volitions themselves, or, instead, a kind of surd when a philosopher claims
that volitions make us free without need to explain what makes the voliti-
ons free. But the power preference avoids the regress and the surd. What
makes the preference free is that we have the preference structure we do
because we prefer to have it. What makes that preference, the power prefer-
ence, free, is that we have it because we prefer to have it as a preference
included in the preference structure. The power preference, since it is a pref-
erence we have because we prefer to have it, ties freedom in, down, and
together in an explanatory loop.
Finally, the theory avoids a kind of paradox of libertarianism. The para-
dox is that the libertarian insists on the importance of indeterminism and
then faces the objection that the mere lack of causal determination does
not seem to contribute in any positive way to the account of freedom.
Ekstrom (2000), who provides us with a profound and articulate theory of

libertarianism, tries to find a useful place for indeterminism at the point of
forming a preference in consideration of reasons in the process of delibera-
tion.5 That is a plausible point for a libertarian to insert indeterminism. But
what purpose does it serve? It is supposed to bolster the idea that it is up to
me at this point what I prefer. But it is up to me only if I have some posi-
tive power to resolve the matter. The mere lack of determinism does not
explain the positive power. The positive power, I propose, is captured by
the primacy of the power preference, by my having the preference I do at
that point, and responding as I do to the reasons I have, because I prefer to
have those preferences and to respond in that way. In short, I propose that
where the libertarian thinks she gains some advantage by introducing
indeterminism into the picture at some point, what is needed is some
account of positive empowerment. That is what the power preference gives
us without appeal to indeterminism provided we have it because we prefer
to have it.
The advantages before us, let us look at the problems that confront com-
patibilism and explain how the sufficient account for freedom solves them.
First of all, incompatibilists will ask how it can be up to us or in our power
what we prefer or do, if our actions and preferences are caused in a way that
extends backward in time to conditions that are beyond our control. As
Spinoza (1901) noted, if a projectile were conscious, it might reflect, “I could
swerve to the left or to the right, but, I prefer to continue on this agreeable
parabolic path.”6 Don’t we seem like the projectile if determinism is true?
No doubt, some will find that image so compelling that there is no con-
vincing them otherwise. But notice that we do not believe that the projec-
tile would swerve off the path if it so preferred. So, there is a difference at
the level of action often noted by compatibilists from Moore on. Moreover,
even if the projectile were conscious and felt inclined to continue on its
parabolic path, it would not have a preference to move along that path be-
cause it preferred to have that preference any more than it would continue
on that path because it preferred to do so. The causal role of preference in
action and in the empowerment of preference is missing. That is why the
example is both convincing and misleading.
Let us, however, look at the harder arguments for incompatibilism. Both
Ginet (1966) and van Inwagen7 have argued that if we have no choice
about the past, which we do not, and if we have no choice about the laws
64 K. Lehrer
of nature, which are part of their definition of laws, then we have no choice
about the outcome of the past conditions and the laws of nature. Or to put
it another way, there is nothing that we can do to render true claims about
the past false or render true claims about the laws false, and, yet, if we can
do otherwise than we do, then either the true claims about the past would
be false or true claims about the laws would be false. Therefore, if we can do
otherwise, we can render either the claims about the past false or the claims
about the laws false. But we cannot do these things. So, if determinism is
true, then we cannot do otherwise than what we do. Such arguments are
very persuasive. How should we reply?
My first reply is that the word ‘can’ is highly ambiguous in a way that
makes it very difficult to argue in a cogent manner about the issue of com-
patibilism formulated in terms of ‘can’. As Moore (1912) noted, there is
clearly a sense of ‘can’ that makes the argument offered sound. I have no
doubt that there is a version of the argument that van Inwagen offers using
‘can’ as he intends so that he can meet the objections of his critics and
argue that if determinism is true, then no one could have done otherwise—
in his use of ‘can’. Similarly, there are other uses, meanings, of ‘can’, some
of which I have proposed, from which compatibilism follows. Examples
include conditional analyses and variations thereof in terms of accessibility
relations to possible worlds, analyzing accessibility in ways that favor com-
patibilism. Formulating the matter in terms of ‘can’ is, I ruefully conclude,
hazardous, and the question is easily begged.
Nevertheless, I do think that the conditions that I have offered are suffi-
cient conditions, not only for freedom, but also for saying that a person
could have done otherwise. So how do we avoid the argument that if de-
terminism is true, then no one could ever have done or preferred to do any-
thing other than what they did or preferred to do? There is a principle that
figures into the argument in one form or another which says that if a
person can do something, and something is a necessary condition for his
doing it (like the falsity of either the past or the laws of nature), then, if the
person can do the thing, then the person can bring about the necessary
conditions for doing it as well. This is plausible. It says that if something is
necessary to doing something, then if you can do the thing in question,
you can bring about the necessary condition. Plausible, but false. One
necessary condition of my writing this paper is the birth of my parents, but
there is nothing I can do to bring about the birth of my parents. Similar
criticisms hold for the general principle that says that if some condition is
necessary for a person doing something, then if the person can do the
thing, the person can bring about the condition.
Now any incompatibilist worth his or her arguments will jump up and
down and reply that it is only conditions that do not exist and which are
necessary for doing something that a person must be able to bring about if
they can perform the action. Again, however, this is false. If the conditions
simply will occur if the person performs the action, that is all that is required.
Now when we turn to what could have been, what a person could have
done, we must say that any condition that is necessary for the person to
have done the action is one that would have occurred. So, as Lewis (1981)
noted, when we say that a person could have done something that she did
not do, then, if determinism is true, we must conclude that if the person
had done what she did not do, then the necessary condition of her doing
it, namely, the falsification of the statements about the past or statements
of laws would have occurred. But we cannot conclude that the person
doing otherwise would have brought about the occurrence of that condi-
tion. That would be a mistake. All we can conclude is that one of those
statements would have been false.
Now let us look at the power preference. Suppose I had preferred other-
wise. Moreover, suppose that I had preferred otherwise because I preferred
to have a preference structure containing that preference. Then a necessary
condition for my preferring otherwise is that I not have preferred what I
actually did and that all necessary conditions for that were satisfied. That
includes that either the laws concerning my preference or the past would
have been otherwise. I see no problem with this. It does not follow from
the fact that I could have preferred otherwise that I could have thereby
brought about the necessary conditions for so preferring, and, more specif-
ically, it does not follow that I could have brought it about that the past or
laws would have been otherwise. All that follows is that they would have
been otherwise if I had preferred otherwise.
You might inquire further what would have been otherwise—the past or
the laws? Suppose it is the laws of nature that would have been otherwise.
What that means, of course, is not that something would have been both a
law and false but, rather, that something that was law would not have been
law. Is that strange? Some, van Inwagen (1975) and Ginet (1966), have
suggested that whether something is a law of nature does not depend on
66 K. Lehrer
our choices or preferences, though van Inwagen (1983) has noted the pos-
sibility of voluntaristic laws. Most laws do not concern human choices and
preferences. Such laws, the laws of chemistry, for example, do not depend
on our choices and preferences. However, they are not laws concerning our
choices and preferences, laws of the form, Lc, to the effect that under con-
ditions C a person S will choose or prefer X, for example. Now consider laws
like Lc, supposing there are some. What is the point of saying that they do
not depend on human choices and preferences? They concern our choices
and preferences, after all, and say that under conditions C people choose or
prefer X. Such laws of the form of Lc clearly do depend on our choices and
preferences. If we do not choose or prefer X under conditions C, then Lc
would not be a law. So, there is nothing strange in the idea that if I had
chosen or preferred other than X under C, then Lc would not have been a
law, though, in fact, it is a law.
Now we come to the critical objection, however. If Lc is a law, shouldn’t
we conclude from this that no one could have not chosen or preferred X
when they were in conditions C? No doubt that there is a sense of ‘could
have’ that would warrant this conclusion, but, as I have noted before, we
need the additional premise that when a person could not have chosen
otherwise in this sense, then the person could not have chosen otherwise
in the sense required for the person to be free to have chosen otherwise.
And it is precisely this step I wish to question. Notice, that a law statement
must be true, and there is a sense of ‘could have’ in which it seems obvious
that no one can make what is true false and, thus, that no one could have
made what is a true law statement false. But that sense of ‘could have’ does
not deprive us of freedom because it rests on the assumption that no one
can make something be both true and false, which is a tautology. That does
not preclude us from saying what is required for freedom, namely, that
somebody could have done something that would have made something
false instead of being true, though, in fact, she did not, and it is true that
she did not.
Philosophers have thought, however, that there is a more important
modal notion of ‘can’ connected with laws, a more significant sense of ‘can’
and ‘could have’, that enables us to distinguish laws from mere accidental
general truths. It is, for example, an accidental general truth that everyone
in this room is older than four years old. The accidental character of the gen-
eral truth might be captured by a modality, by saying that someone could
have falsified this general truth by bringing her four-year-old to hear about
freedom and the power of preference, but no one could have done anything
to falsify any law of nature. I have two replies with which I shall conclude.
First, it may be true of some but not all laws of nature, namely those con-
cerned with matters other than human choice and preference, that they
cannot be falsified by human choice or preference, but it does not follow
from this that no law about human choice or preference can be falsified by
human choice or preference.
Second, and more crucially, there is another and, I believe, better way of
distinguishing between laws and accidentally true generalizations than by
appeal to what can and could not be. Laws warrant counterfactual infer-
ence of a kind that is not warranted by accidental generalizations. The
specification of the exact kind of counterfactual inference is a delicate
matter requiring some subtle distinctions, but the general point is clear
enough. The accidentally true generalization that all people in this room
are older than four years does not warrant the counterfactual inference to
the conclusion that if a four-year-old had been brought in to hear about
freedom and the power preference she would have been older than four
years old. Laws by contrast do warrant counterfactual conclusions about
what would have been. So, if Lc is a law and I, in fact, do not choose or pre-
fer X, someone would be warranted in inferring that if I had been in con-
ditions C, which I was not, then I would have chosen or preferred X. The
person would not, however, be warranted in inferring that I could not have
chosen or preferred otherwise than X when I am in C. They could only
infer that I would not.
I conclude with the remark that this conception of a law is one that dif-
fers from what incompatibilists such as Ginet and van Inwagen have as-
sumed about laws in their arguments for incompatibilism. Am I, therefore,
just missing the point of their argument? One motivation for concern
about the problem of freedom and determinism is that it appears that a sci-
entific conception of the world based on the laws of nature and explana-
tion is incompatible with the sort of freedom we believe ourselves to have.
I have intended to allay this concern. The scientific conception of the
world is, I suggest, one including statements of laws that warrant inference
and counterfactual inference about the world. Some philosophers may go
beyond this conception to one affirming a thesis of laws that contain a
kind of necessity that is incompatible with human freedom. If one builds
68 K. Lehrer
incompatibility with freedom into the definition of laws by defining the

latter in such terms, then incompatibility will be the result. It would be a
mistake, however, to conclude that the scientific conception of the world
based on laws that warrant inference and counterfactual inference are in-
compatible with our conception of human freedom. One important prob-
lem is to explain how human freedom can fit into the causal nexus of
inference and counterfactual inference warranted by laws. My explanation
is that it is contained in the keystone loop of preference and ultrapref-
erence that ties together our conception of ourselves as free with our con-
ception of a causal order explaining our freedom. Causally explained
freedom accounts for our place in nature.
Notes
1. The classic statement of the conditional analysis was Moore (1912).
2. Peter van Inwagen and Krister Segerberg both communicated the objection to me
in correspondence. Cf. Peter van Inwagen (1983), chap. 4.
3. Moore (1912), chap. 6.
4. I was not able to pin down this interpretation to a specific passage, for his prose
has a charmingly diffuse character, but I am indebted to Sartre (1956).
5. See chap. 4.
6. See props. xlvii–xlix.
7. Originally argued in van Inwagen (1975), and more fully in van Inwagen (1983).
References
Chisholm, R. 1966. “Freedom and Action.” In K. Lehrer, ed., Freedom and Determin-
ism. New York: Random House.
Ekstrom, L. 2000. Free Will: A Philosophical Study. Boulder, Colo.: Westview Press.
Frankfurt, H. 1969. “Alternate Possibilities and Moral Responsibility.” Journal of

Philosophy 68: 388–404.
. 1971. “Freedom of the Will and Concept of a Person.” Journal of Philosophy

68: 5–20.
Ginet, C. 1966. “Might We Have No Choice?” In K. Lehrer, ed., Freedom and Deter-
minism. New York: Random House.
Kant, I. 1993. Critique of Practical Reason. L. W. Beck, trans. New York: Macmillan.
Kripke, S. 1975. “Outline of a Theory of Truth.” Journal of Philosophy 72: 53–81.
Lehrer, K. 1990. Metamind. Oxford: Clarendon Press.
. 1992. “Preferences, Conditionals and Freedom.” In K. Lehrer, Metamind.

Oxford: Clarendon Press. Reprinted from P. van Inwagen, ed. 1980. Time and Cause.
Dordrecht: Reidel.
. 1997. Self-Trust: A Study of Reason, Knowledge and Autonomy. Oxford:

Clarendon Press.
Lewis, D. 1981. “Are We Free to Break the Laws?” Theoria 47: 113–121.
Moore, G. E. 1912. Ethics. New York: Oxford University Press.
Reid, T. 1788. Essays on the Active Powers of Man. Edinburgh: J. Bell.
Sartre, J. 1956. Being and Nothingness. H. Barnes, trans. New York: Philosophical
Library.
Spinoza, B. 1901. Ethics. R. H. M. Elwes, trans. London: M. W. Dunne.
Taylor, R. 1966. Action and Purpose. Englewood Cliffs, N.J.: Prentice-Hall.
van Inwagen, P. 1975. “The Incompatibility of Free Will and Determinism.”

Philosophical Studies 27: 185–199.
Velleman, J. D. 1992. “What Happens When Someone Acts.” Mind 101: 461–481.
3 Agency, Responsibility, and Indeterminism: Reflections
on Libertarian Theories of Free Will
Robert Kane
1 The Compatibility Question: AP and UR
Free will, as I understand it, is “the power of agents to be the ultimate

creators and sustainers of their own ends or purposes.” Every word here is
significant. We create purposes by making choices or decisions. Hence the
importance of free choice or decision for free will. We sustain purposes by
making efforts to do so. Hence the importance of efforts of will (or striving
will, as Brian O’Shaughnessy calls it) in my account of free will. More will
be said about the terms ‘power’, ‘ultimate’, and ‘agency’ later.
I think the notion captured by this definition is the traditional idea of
free will that has been in dispute for centuries. I’ve tried to make the his-
torical case for this in many writings and won’t repeat it here. In writings
over the past two decades, I have sought to answer four questions about
free will: (1) Is it compatible (or incompatible) with determinism? (2) Why
do or should we want it? (3) Can we make sense of a free will that is in-
compatible with determinism? (4) Where and how might such a free will
exist in the natural order and can it be reconciled with modern images
of human beings in the natural and social sciences?1 These are the Com-
patibility, Significance, Intelligibility, and Existence questions respectively.
Throughout this century and notably in recent decades, the lion’s share of
debate has been on the Compatibility question, as we all know, with some
attention, but considerably less given to the other three questions.
In writings over these two decades, including two books, I have tried to
right this imbalance by giving an in-depth treatment to all four questions.
In addition, on all four questions, I have tried to point current debates
about free will in new directions. I focus on a few of these new directions in
the present paper. Consider the Compatibility question. Most recent and
Agency, Responsibility, and Indeterminism 71
past philosophical debate about this question has focused on the question
of whether determinism is compatible with ‘the condition of alternative
possibilities’ (which I call AP)—the requirement that the free agent ‘could
have done otherwise’. As we all know, most arguments for and against in-
compatibilism revolve around AP in one way or another. I don’t think these
debates are unimportant. Far from it. I have engaged in them myself and
they continue. But I think that exclusive focus on AP and ‘could have done
otherwise’ has led philosophers to ignore other important facets of the
debate about Compatibility and Significance questions. AP alone, I argue,
provides too thin a basis on which to rest the case for incompatibilism.
Debates over different interpretations of ‘can’, ‘power’, ‘ability’, and ‘could
have done otherwise’ are symptoms of this fact.
To see what I mean by saying AP provides too thin a basis for resolving
the Compatibility question, consider certain kinds of acts I will call Austin-
style examples (after J. L. Austin who, among others, introduced examples
of these kinds into free will debates in Austin 1956). Austin’s own example
of this type concerned a three foot putt he was trying to make in a golf
match, but missed by chance. Another example of this kind I have used is
this: suppose I approach a coffee machine intending to press the button for
black coffee, but have a brain cross as I stand before the machine and press
the button for coffee with cream instead by mistake. Imagine in both these
cases, Austin’s missing the putt and my pressing the wrong button, some
genuine indeterminacy was involved in the efferent neural pathways of his
arm and the impulses of my brain such that they were genuinely undeter-
mined. They were still actions we performed, as Austin notes of examples
of this type, but unintentionally, nonvoluntarily, by accident or mistake. In
each case, we could have done otherwise (made the putt and pressed the
right button since the outcomes were undetermined). So in each case we
have actions that are (i) such that we could have done otherwise (AP) and
(ii) that are undetermined. Yet neither is a free action in any normal sense
of the word since neither was under our control.
Now imagine a world full of Austin-style examples of this type; a world
in which indeterminism is prevalent and agents can often do otherwise in
an undetermined way in the Austin style—by accident, mistake, uninten-
tionally or inadvertently. But imagine also that in this undetermined
world, God has set all agents’ wills (their reasons, motives, intentions
and purposes) so that they desire and intend to do only what God has
72 R. Kane
predetermined. Austin might miss his putt, but his wanting and trying to
make it rather than miss was preset by God. I might press the wrong button
on the coffee machine but my wanting and intending to press the right
button was predetermined by God. And so it is with all agents for all their
acts in this world. I would argue that this is a world in which agents lack free-
dom of will, even though they are often capable of doing otherwise (have
AP) and many of their acts are undetermined. They are not the sources
of their own wills. All the ‘will-setting’ in this world is done by God and
none by the agents themselves.
This shows that being able to do otherwise (AP) even if the doing other-
wise is undetermined is not sufficient for free will (even if these two condi-
tions of AP plus indeterminism should be necessary). This is one way of
showing why AP is too thin a basis on which to rest the case for incompat-
ibilism. The agents must be able to will otherwise as well as do otherwise.
What we want for free will is not merely to be able to do otherwise but to
be able to do otherwise (or more generally, to act either way) voluntarily,
intentionally, and rationally, rather than one way voluntarily, intention-
ally, and rationally and the other only unintentionally, involuntarily, and
by mistake or accident, as in the Austin-style examples. I call these require-
ments of more-than-one-way voluntariness, intentionality, and rationality
the ‘plurality conditions’ for free will. By satisfying these conditions in a
manner that was undetermined, agents would be engaged in their own
‘will-setting’; setting their wills (intentions and purposes) would not all be
done by God or nature, society, or something else over which they had no
control. AP is not unimportant for free will but it is important because of
these strong requirements of ‘plurality’ and ‘will-setting’ for full-fledged
free will. I have therefore argued in my writings that more attention should
be given to plurality conditions and will-setting in discussions of Compat-
ibility and Significance questions rather than to AP alone.
But plurality conditions and will-setting are not enough. There is yet a
third condition that has to be considered besides AP to get a fully adequate
picture of the Compatibility debate. I have also argued that in the long
history of the free will debate one can find another criterion fueling incom-
patibilist intuitions that is even more important than AP, though compar-
atively neglected. I call it the condition of ultimate responsibility or UR. The
basic idea is this: to be ultimately responsible for an action, an agent must be
responsible for anything that is a sufficient reason (condition, cause, or
motive) for the action’s occurring.2 If, for example, a choice issues from, and
can be sufficiently explained by, an agent’s character and motives (together
with background conditions), then to be ultimately responsible for the
choice, the agent must be at least in part responsible by virtue of choices or
actions voluntarily performed in the past for having the character and mo-
tives he or she has now. Compare Aristotle’s claim that if a man is responsi-
ble for wicked acts that flow from his character, he must at some time in the
past have been responsible for forming the wicked character from which
these acts flow (Aristotle 1985).
This UR condition accounts for the ‘ultimate’ in the original definition of
free will: “the power of agents to be the ultimate creators and sustainers of
their own ends or purposes.” Now UR does not require that we could have
done otherwise (AP) for every act done of our own free wills—thus vindicat-
ing philosophers such as Frankfurt, Fischer, and Dennett, and others who
insist that we can be held morally responsible for many acts even when we
could not have done otherwise.3 But the vindication is only partial. For UR
does require that we could have done otherwise with respect to some acts in
our past life histories by which we formed our present characters. I call these
‘self-forming actions’, or SFAs. Consider Dennett’s comments on Martin
Luther (1984, 133). When Luther finally broke with the Church in Rome, he
said “Here I stand, I can do no other.” Suppose, says Dennett, at that mo-
ment Luther was literally right. Given his character and motives, he could
not then have done otherwise. Does this mean he was not morally respon-
sible for his act, or not acting of his own free will? Dennett says “Not at all.”
In saying “I can do not other,” Luther was not disowning responsibility, but
taking full responsibility for acting of his own free will. So ‘could have done
otherwise’, or AP, is not required for moral responsibility or free will.
My response is to grant that Luther could have been responsible for this
act, even ultimately responsible in the sense of UR, though he could not
then have done otherwise and even if his act was determined. But this
would be so to the extent that he was responsible for his present motives
and character by virtue of earlier struggles and self-forming choices (SFAs)
that brought him to this point in his life where he could do no other. Often
we act from a will already formed, but it is ‘our own free will’ by virtue of
the fact that we formed it by other choices or actions in the past (SFAs) for
which we could have done otherwise. If this were not so, there is nothing
we could have ever done to make ourselves different than we are—a
74 R. Kane
consequence, I believe, that is incompatible with our being ultimately re-

sponsible for what we are. So self-forming actions that are undetermined and
such that we could have done otherwise are only a subset of all actions done
‘of our own free wills’ that satisfy UR. But if none of our acts were self-forming
in this way, we would not be ultimately responsible for anything we did.
If the case for incompatibility cannot be made on AP alone, it can be
made if UR is added; and so, I suggest, the too-often neglected UR should
be moved to center stage in free will debates. If agents must be responsible
to some degree for anything that is a sufficient cause or motive for their
actions, an impossible infinite regress of past actions would be required un-
less some actions in the agent’s life history did not have either sufficient
causes or motives (and hence were undetermined). These are the SFAs
required by UR. Note that this argument makes no mention of AP or the
power to do otherwise. When we focus on ‘could have done otherwise’ we
are led to modal issues about ‘can’ and ‘power’, ‘possibility’ and ‘necessity’;
when we focus on UR, by contrast, we are interested in a different set of
issues about the ‘sources’ or ‘grounds’ or ‘explanations’ of our actions.
Where did they come from? Who or what was their ultimate source?
But surprisingly, this does not mean AP is out of the picture. For it can
be shown that the self-forming actions (SFAs) required by UR must also be
such that the agents could have done otherwise. (Thus the Consequence
argument of van Inwagen, Ginet, and others is vindicated after all, but only
for SFAs.) In other words, UR not only implies indeterminism, but by a dif-
ferent argumentative route, it also implies AP for some free actions. For, UR
implies the requirements of will-setting and plurality conditions men-
tioned earlier for some free actions in our life histories (i.e., SFAs) and these
conditions in turn imply AP. I forego the full argument for these entail-
ments here. But let me say that they follow surprisingly from the require-
ment that to have free will one must not only be ultimately responsible for
one’s actions but also for one’s will to perform those actions. The upshot is
that UR is the root which leads by one pathway to indeterminism and by
another to AP. I call this the dual regress of free will.
A few other things to note about the formal definition of UR before
I move on. The formal definition goes like this:
(UR) An agent is ultimately responsible for some (event or state) E’s
occurring only if (R) the agent is personally responsible for E’s occurring
in a sense which entails that something the agent voluntarily did or
omitted either was, or causally contributed to, E’s occurrence and made a
difference to whether or not E occurred; and (U) for every X and Y (where
X and Y represent occurrences of events and/or states) if the agent is
personally responsible for X, and if Y is a sufficient reason (condition,
cause, or motive) for X, then the agent must also be personally
responsible for Y.
This definition has two parts, a Responsibility condition R and an

Ultimacy condition U. U makes reference to the notion of a sufficient reason
for something’s being the case, in the traditionally broad sense of a suffi-
cient condition, cause, or motive. This traditional notion of a sufficient
reason is of course an ingredient of the notorious principle of sufficient
reason, regarded by Leibniz (1991) as the fundamental principle governing
all reasoning about matters of fact. This same notion of a sufficient reason,
I argue, is an ingredient in the traditional idea of free will, if that traditional
idea involves UR. Of course, the principle of sufficient reason (as opposed
to the idea of a sufficient reason) is another matter. If everything in the uni-
verse had a sufficient reason for being, as the principle insists, there could
be no free will in the sense of UR. But the idea of sufficient reason is quite
another matter. Similar things may be said about another traditional prin-
ciple of rationality in the history of philosophy, the so-called principle of
plenitude: every genuine possibility must be realized. If this principle were
universally true, free will would also not be possible. For it is essential to
free will that some things that do not get realized were nonetheless gen-
uinely possible. It should not surprise us that the principles of sufficient
reason and plenitude are in the same boat here with regard to free will. For
it can be shown that these pivotal rational principles are equivalent to each
other.
2 The Intelligibility Question
This different approach to the Compatibility question through UR raises a

host of further questions about free will, including how actions lacking both
sufficient causes and motives could themselves be free and responsible ac-
tions, and how, if at all, such actions could exist in the natural order where
we humans live and have our being. These are versions of the Intelligibility
and Existence questions mentioned earlier, and to which I now turn. The
Intelligibility question is connected to an ancient dilemma: if free will is
76 R. Kane
not compatible with determinism, it does not seem to be compatible with

indeterminism either. You might say the Compatibility question concerns
the first half of this dilemma, the Intelligibility question the second. I’ll not
trouble you with a detailed account of the problems posed by indetermin-
ism for free will which are familiar to anyone who has delved into the free
will debates. Suffice it to say that by various routes it has been argued and
has seemed to many persons that undetermined choices or actions would
occur as a matter of chance and hence would be ‘arbitrary’, ‘capricious’,
‘random’, ‘irrational’, ‘inexplicable’, ‘mere matters of luck’, and not under
the ‘control’ of the agents, hence not free and responsible actions at all.
What prompted me to begin thinking about free will thirty years ago was
disenchantment with the standard incompatibilist or libertarian strategy
for answering these familiar charges. The strategy behind all standard lib-
ertarian answers was this: indeterminism was required for free will all right,
but all it provided was the causal gaps in nature needed for free will. Some-
thing else was needed to fill those gaps, some additional form of agency or
causation that went beyond causation in the natural order, whether deter-
ministic or indeterministic. This general strategy led to numerous histori-
cal appeals to extra factors such as noumenal selves, immaterial minds,
transempirical power centers, immanent causes, and the like, to account
for libertarian free will with which we are all familiar. The other half of the
standard strategy—equally important—was to say that this extra factor,
whatever it might be, noumenal self, immaterial mind, or special agent
cause, could not be determined because it was not subject to deterministic
or causal laws of the natural order by its very nature. Perhaps it was outside
space or time, or it was an agent, not an event, and hence not capable of
being the effect of another cause.
So standard libertarian strategies postulated an extra factor to explain free
will; and they were aprioristic; they argued that this extra factor precluded
determinism in principle. I was suspicious of both prongs of the standard
strategy—the extra factors, whose operation often turned out to be mysteri-
ous or obscure, and the appeal to an a priori metaphysical resolution of the
determinism problem. It seems to me that the question of whether or not we
have free will is an empirical question which cannot be settled by postulat-
ing novel metaphysical entities that preclude determinism in principle.
Where else to go then if one rejects both prongs of this standard libertar-
ian strategy? In Kane (1989) I distinguished two radically different kinds of
libertarian theories, which I called agent-causal or AC theories, and teleolog-

ical intelligibility or TI theories. The agent-causal or AC theories are the best
modern examples of the standard libertarian strategy. They have their
origins in Berkeley and Reid, and able modern defenders including Randy
Clarke (1993) and Tim O’Connor (2000). They postulate, in the words of
O’Connor, “a sui generis form of causation by an agent that is irreducible . . .
to event-causal processes within the agent,” a kind of nonevent or nonoc-
current causation. Teleological intelligibililty or TI theories, by contrast,
try to make undetermined free actions intelligible in terms of reasons or
motives, intentions or purposes (hence teleologically intelligible) without
postulating sui generis kinds of agency or causation that cannot be spelled
out in terms of events or states of affairs involving the agents.
My theory is of this TI kind. I believe my 1985 book was the first attempt
in recent philosophy to work out a TI theory in detail, though the idea had
been suggested earlier by some philosophers such as David Wiggins and
Robert Nozick. Other TI theories have come along since then different from
mine, by other philosophers including two appearing in this volume, Carl
Ginet and Alfred Mele. (Others who have looked in a TI direction are Storrs
McCall, Hugh McCann, and Laura Ekstrom.) But I won’t get into the nu-
ances of different TI theories. I’ll merely note that they constitute another
option to standard libertarian strategies exemplified by AC theories, an
option that has been growing in influence since I wrote in 1985. From this
point onward, I’ll focus on my own TI approach.
3 Indeterminism and Responsibility
If you avoid standard libertarian strategies of postulating extra factors, then

I believe you have to take a whole new look at the indeterminist problem
from the ground up. The first step is to note that indeterminism does not
have to be involved in all acts done ‘of our own free wills’ for which we are
ultimately responsible, as argued earlier. Not all such acts have to be unde-
termined, but only those by which we made ourselves into the kinds of per-
sons we are, namely ‘self-forming actions’ or SFAs. Now I believe these
undetermined self-forming actions or SFAs occur at those difficult times of
life when we are torn between competing visions of what we should do or
become. Perhaps we are torn between doing the moral thing or acting from
ambition, or between powerful present desires and long-term goals, or we
78 R. Kane
are faced with difficult tasks to which we have aversions. In all such cases,
we are faced with competing motivations and have to make an effort to
overcome temptation to do something else we also strongly want. There is
tension and uncertainty in our minds about what to do at such times, I sug-
gest, that is reflected in appropriate regions of our brains by movement
away from thermodynamic equilibrium—in short, a kind of ‘stirring up of
chaos’ in the brain that makes it sensitive to microindeterminacies at the
neuronal level. The uncertainty and inner tension we feel at such soul-
searching moments of self-formation is thus reflected in the indeterminacy
of our neural processes themselves. What is experienced internally as
uncertainty then corresponds physically to the opening of a window of
opportunity that temporarily screens off complete determination by influ-
ences of the past. (By contrast, when we act from predominant motives or
settled dispositions, the uncertainty or indeterminacy is muted. If it did
play a role in such cases, it would be a mere nuisance or fluke, as critics of
indeterminism contend.)
When we do decide under such conditions of uncertainty, the outcome
is not determined because of the preceding indeterminacy—and yet it can
be willed (and hence rational and voluntary) either way owing to the fact
that in such self-formation, the agents’ prior wills are divided by conflict-
ing motives. Consider a businesswoman who faces such a conflict. She is
on her way to an important meeting when she observes an assault taking
place in an alley. An inner struggle ensues between her conscience, to stop
and call for help, and her career ambitions that tell her she cannot miss this
meeting. She has to make an effort of will to overcome the temptation to
go on. If she overcomes this temptation, it will be the result of her effort,
but if she fails, it will be because she did not allow her effort to succeed.
And this is due to the fact that, while she willed to overcome temptation,
she also willed to fail, for quite different and incommensurable reasons.
When we, like the woman, decide in such circumstances, and the indeter-
minate efforts we are making become determinate choices, we make one
set of competing reasons or motives prevail over the others then and there
by deciding.
Now let us add a further piece to the puzzle. Just as indeterminism need
not undermine rationality and voluntariness, so indeterminism in and of
itself need not undermine control and responsibility. Suppose you are try-
ing to think through a difficult problem, say a mathematical problem, and
there is some indeterminacy in your neural processes complicating the

task—a kind of chaotic background. It would be like trying to concentrate
and solve a problem, say a mathematical problem, with background noise
or distraction. Whether you are going to succeed in solving the problem is
uncertain and undetermined because of the distracting neural noise. Yet, if
you concentrate and solve the problem nonetheless, there is reason to say
you did it and are responsible for it even though it was undetermined
whether you would succeed. The indeterministic noise would have been an
obstacle that you overcame by your effort.
There are numerous examples supporting this point, first suggested by
J. L. Austin, Philippa Foot, Elizabeth Anscombe, and others in the 1950s,
where indeterminism functions as an obstacle to success without preclud-
ing responsibility. (They are variations on the Austin-style examples men-
tioned earlier.) Consider an assassin who is trying to shoot the prime
minister, but might miss because of some undetermined events in his ner-
vous system that may lead to a jerking or wavering of his arm. If the assas-
sin does succeed in hitting his target, despite the indeterminism, can he be
held responsible? The answer is clearly yes because he intentionally and
voluntarily succeeded in doing what he was trying to do—kill the prime
minister. Yet his action, killing the prime minister, was undetermined. Or,
here is another example: a husband, while arguing with his wife, in a fit of
rage swings his arm down on her favorite glass-top table top intending to
break it. Again, we suppose that some indeterminism in his outgoing
neural pathways makes the momentum of his arm indeterminate so that it
is undetermined whether the table will break right up to the moment when
it is struck. Whether the husband breaks the table or not is undetermined
and yet he is clearly responsible if he does break it. (It would be a poor ex-
cuse for him to say to his wife “chance did it, not me.” Even though there
was a chance he wouldn’t break it, chance didn’t do it—he did.)
Now these examples—of the mathematical problem, the assassin, and the
husband—are not all we want since they do not amount to genuine exercises
of (self-forming) free will in SFAs, like the businesswoman’s, where the will
is divided between conflicting motives. The assassin’s will is not divided be-
tween conflicting motives as is the woman’s. He wants to kill the prime min-
ister, but does not also want to fail. (If he fails therefore, it will be merely by
chance.) Yet these examples of the assassin, the husband and the like, do
provide some clues. To go further, we have to add some new twists.
80 R. Kane
Imagine in cases of inner conflict characteristic of SFAs, like the busi-

nesswoman’s, that the indeterministic noise which is providing an obstacle
to her overcoming temptation is not coming from an external source, but
is coming from her own will, since she also deeply desires to do the oppo-
site. Imagine that two crossing (recurrent) neural networks are involved,
each influencing the other, and representing her conflicting motivations.
(These are complex networks of interconnected neurons in the brain circu-
lating impulses in feedback loops that are generally involved in higher-
level cognitive processing.)4 The input of one of these neural networks
consists in the woman’s reasons for acting morally and stopping to help
the victim; the input of the other, her ambitious motives for going on to
her meeting. The two networks are connected so that the indeterministic
noise which is an obstacle to her making one of the choices is coming from
her desire to make the other, and vice versa—the indeterminism thus aris-
ing from a tension-creating conflict in the will, as I said. In these circum-
stances, when either of the pathways ‘wins’ (i.e., reaches an activation
threshold, which amounts to choice), it will be like your solving the math-
ematical problem by overcoming the background noise produced by the
other. And just as when you solved the mathematical problem by over-
coming the distracting noise, one can say you did it and are responsible for
it, so one can say this as well, I argue, in the present case, whichever one is
chosen. The pathway through which the woman succeeds in reaching a
choice threshold will have overcome the obstacle in the form of indeter-
ministic noise generated by the other.
Note that, under such conditions, the choices either way will not be
‘inadvertent’, ‘accidental’, ‘capricious’, or ‘merely random’ (as critics of
indeterminism say) because they will be willed by the agents either way
when they are made, and done for reasons either way—reasons that the
agents then and there endorse. But these are the conditions usually required
to say something is done ‘on purpose’, rather than accidentally, capri-
ciously, or merely by chance. Moreover, these conditions taken together,
I argue, rule out each of the reasons we have for saying that agents act, but
do not have control over their actions (compulsion, coercion, constraint,
inadvertence, accident, control by others, etc.).5 I show this for each of
these conditions in turn in Kane (1996). Of course, for undetermined SFAs,
agents do not control or determine which choice outcome will occur before
it occurs; but it does not follow, because one does control or determine
which of a set of outcomes is going to occur before it occurs, that one does
not control or determine which of them occurs, when it occurs. When the
above conditions for SFAs are satisfied, agents exercise control over their
future lives then and there by deciding.
Indeed, they have what I call ‘plural voluntary control’ over the options
in the following sense: they are able to bring about whichever of the op-
tions they will, when they will to do so, for the reasons they will to do so,
on purpose rather than accidentally or by mistake, without being coerced
or compelled in doing so or willing to do so, or otherwise controlled in
doing or willing to do so by any other agents or mechanisms. I show again
in Kane (1996) that each of these conditions can be satisfied for SFAs as
conceived above.6 The conditions can be summed up by saying, as we
sometimes do, that the agents can choose either way, at will.
Note also that this account of self-forming choices amounts to a kind of
‘doubling’ of the mathematical problem. It is as if an agent faced with such
a choice is trying or making an effort to solve two cognitive problems at
once, or to complete two competing (deliberative) tasks at once—in our
example, to make a moral choice and to make a conflicting self-interested
choice (corresponding to the two competing neural networks involved).
Each task is being thwarted by the indeterminism coming from the other,
so it might fail. But if it succeeds, then the agents can be held responsible
because, as in the case of solving the mathematical problem, they will have
succeeded in doing what they were knowingly and willingly trying to do.
Recall the assassin and the husband. Owing to indeterminacies in their
neural pathways, the assassin might miss his target or the husband fail to
break the table. But if they succeed, despite the probability of failure, they
are responsible, because they will have succeeded in doing what they were
trying to do.
And so it is, I suggest, with self-forming choices, except that in the case of
self-forming choices, whichever way the agents choose they will have suc-
ceeded in doing what they were trying to do because they were simultane-
ously trying to make both choices, and one is going to succeed. Their failure
to do one thing is not a mere failure, but a voluntary succeeding in doing
the other. Does it make sense to talk about the agent’s trying to do two
competing things at once in this way, or to solve two cognitive problems at
once? Well, we now know that the brain is a parallel processor; it can simul-
taneously process different kinds of information relevant to tasks such as
82 R. Kane
perception or recognition through different neural pathways. Such a capac-

ity, I believe, is essential to the exercise of free will. In cases of self-formation
(SFAs), agents are simultaneously trying to resolve plural and competing
cognitive tasks. They are, as we say, of two minds. Yet they are not two
separate persons. They are not dissociated from either task. The business-
woman who wants to go back to help the victim is the same ambitious
woman who wants to go to her meeting. She is torn inside by different
visions of who she is and what she wants to be, as we all are from time to
time. But this is the kind of complexity needed for genuine self-formation
and free will. And when she succeeds in doing one of the things she is trying
to do, she will endorse that as her resolution of the conflict in her will,
voluntarily and intentionally, not by accident or mistake.
4 Responsibility, Luck, and Chance
You may find this interesting and yet still find it hard to shake the intuition
that if choices are undetermined, they must happen merely by chance—
and so must be ‘random’, ‘capricious’, ‘uncontrolled’, ‘irrational’, and all
the other things usually charged. Such intuitions are deeply ingrained. But
if we are ever going to understand free will, I think we will have to break old
habits of thought that support such intuitions and learn to think in new
ways. The first step is to question the intuitive connection in most people’s
minds between ‘indeterminism’s being involved in something’ and ‘its
happening merely as a matter of chance or luck’. ‘Chance’ and ‘luck’ are
terms of ordinary language that carry the connotation of ‘its being out of
my control’. So using them already begs certain questions, whereas ‘inde-
terminism’ is a technical term that merely precludes deterministic causa-
tion, though not causation altogether. Indeterminism is consistent with
nondeterministic or probabilistic causation, where the outcome is not in-
evitable. It is therefore a mistake (alas, one of the most common in debates
about free will) to assume that ‘undetermined’ means ‘uncaused’.
Here is another source of misunderstanding. Since the outcome of the
businesswoman’s effort (the choice) is undetermined up to the last minute,
we may have the image of her first making an effort to overcome the
temptation to go on to her meeting and then at the last instant ‘chance
takes over’ and decides the issue for her. But this is misleading. On the view
I proposed, one cannot separate the indeterminism and the effort of will,
so that first the effort occurs followed by chance or luck (or vice versa). One
must think of the effort and the indeterminism as fused; the effort is inde-
terminate and the indeterminism is a property of the effort, not something
separate that occurs after or before the effort. The fact that the effort has
this property of being indeterminate does not make it any less the woman’s
effort. The complex recurrent neural network that realizes the effort in the
brain is circulating impulses in feedback loops and there is some indeter-
minacy in these circulating impulses. But the whole process is her effort of
will and it persists right up to the moment when the choice is made. There
is no point at which the effort stops and chance ‘takes over’. She chooses as
a result of the effort, even though she might have failed. Similarly, the hus-
band breaks the table as a result of his effort, even though he might have
failed because of the indeterminacy. (That is why his excuse, “chance broke
the table, not me” is so lame.)
Just as expressions like ‘she chose by chance’ can mislead in such con-
texts, so can expressions like ‘she got lucky’. Recall that, with the assassin
and husband, one might say ‘they got lucky’ in killing the prime minister
and breaking the table because their actions were undetermined. Yet they
were responsible. So ask yourself this question: why does the inference ‘he
got lucky, so he was not responsible’ fail in the cases of the husband and
the assassin where it does fail? The first part of an answer has to do with the
point made earlier that ‘luck’, like ‘chance’, has question-begging implica-
tions in ordinary language that are not necessarily implications of ‘indeter-
minism’ (which implies only the absence of deterministic causation). The
core meaning of ‘he got lucky’ in the assassin and husband cases, which is
implied by indeterminism, I suggest, is that ‘he succeeded despite the prob-
ability or chance of failure’; and this core meaning does not imply lack of
responsibility, if he succeeds.
If ‘he got lucky’ had other meanings in these cases that are often associ-
ated with ‘luck’ and ‘chance’ in ordinary usage (for example, the outcome
was not his doing, or occurred by mere chance, or he was not responsible
for it), the inference would not fail for the husband and assassin, as it
clearly does. But the point is that these further meanings of ‘luck’ and
‘chance’ do not follow from the mere presence of indeterminism. The
second reason why the inference ‘he got lucky, so he was not responsible’
fails for the assassin and the husband is that what they succeeded in doing
was what they were trying and wanting to do all along (kill the minister
84 R. Kane
and break the table respectively). The third reason is that when they suc-
ceeded, their reaction was not ‘oh dear, that was a mistake, an accident—
something that happened to me, not something I did’. Rather they endorsed
the outcomes as something they were trying and wanting to do all along,
that is to say, knowingly and purposefully, not by mistake or accident.
But these conditions are satisfied in the businesswoman’s case as well,
either way she chooses. If she succeeds in choosing to return to help the
victim (or in choosing to go on to her meeting) (i) she will have ‘succeeded
despite the probability or chance of failure’, (ii) she will have succeeded in
doing what she was trying and wanting to do all along (she wanted both
outcomes very much, but for different reasons, and was trying to make
those reasons prevail in both cases), and (iii) when she succeeded (in choos-
ing to return to help) her reaction was not ‘oh dear, that was a mistake, an
accident—something that happened to me, not something I did’. Rather
she endorsed the outcome as something she was trying and wanting to do
all along; she recognized it as her resolution of the conflict in her will. And
if she had chosen to go on to her meeting she would have endorsed that
outcome, recognizing it as her resolution of the conflict in her will.
Perhaps the problem is that we are begging the question by assuming
the outcomes of the woman’s efforts are choices to begin with, if they are
undetermined. One might argue that ‘if an event is undetermined, it must
be something that merely happens and cannot be somebody’s choice or
action’. But this assumption is clearly question begging in the extreme, as
one can see by considering that it entails ‘all choices and actions must de-
termined’ simply on the grounds that they are choices or actions. Well,
perhaps indeterminism does not undermine the idea that something is a
choice or action simply, but rather that it is the agent’s choice. But again,
why must that be so? What makes the woman’s choice her own on the
preceding account is that it results from her efforts and deliberation
which in turn are causally influenced by her reasons and her intentions
(for example, her intention to resolve indecision in one way or another).
And what makes these efforts, deliberation, reasons and intentions hers is
that they are embedded in a larger motivational system realized in her
brain in terms of which she defines herself as a practical reasoner and
actor.7 A choice is the agent’s when it is produced intentionally by efforts,
deliberation and reasons that are part of this self-defining motivational
system.
Well, then, perhaps the issue is not whether the undetermined SFA is a
choice, or even whether it is the agent’s choice, but rather how much con-
trol she had over it. It may be true, as I argued earlier (in the discussion of
plural voluntary control), that the presence of indeterminism need not
eliminate control altogether. But would not the presence of indeterminism
at least diminish the control persons have over their choices and other ac-
tions? Is it not the case that the assassin’s control over whether the prime
minister is killed (his ability to realize his purposes or what he is trying to
do) is lessened by the undetermined impulses in his arm—and so also for
the husband and his breaking the table? And this limitation seems to be
connected with another problem often noted by critics of libertarian
freedom—the problem that indeterminism, wherever it occurs, seems to be
a hindrance or obstacle to our realizing our purposes and hence an obstacle
to (rather than an enhancement of) our freedom.
There is something to these claims, but I think what is true in them
reveals something important about free will. We should concede that inde-
terminism, wherever it occurs, does diminish control over what we are
trying to do and is a hindrance or obstacle to the realization of our pur-
poses. But recall that in the case of the businesswoman (and SFAs gener-
ally), the indeterminism that is admittedly diminishing her control over
one thing she is trying to do (the moral act of helping the victim) is com-
ing from her own will—from her desire and effort to do the opposite (go to
her business meeting). And the indeterminism that is diminishing her con-
trol over the other thing she is trying to do (act selfishly and go to her
meeting) is coming from her desire and effort to do the opposite (to be a
moral person and act on moral reasons). So, in each case, the indetermin-
ism is functioning as a hindrance or obstacle to her realizing one of her
purposes—a hindrance or obstacle in the form of resistance within her will
that has to be overcome by effort.
If there were no such hindrance—if there were no resistance in her will—
she would indeed in a sense have “complete control” over one of her op-
tions. There would no competing motives that would stand in the way of
her choosing it. But then also she would not be free to rationally and vol-
untarily choose the other purpose because she would have no good com-
peting reasons to do so. Thus, by being a hindrance to the realization of
some of our purposes, indeterminism paradoxically opens up the genuine
possibility of pursuing other purposes—of choosing or doing otherwise in
86 R. Kane
accordance with, rather than against, our wills (voluntarily) and reasons
(rationally). To be genuinely self-forming agents (creators of ourselves)—to
have free will—there must at times in life be obstacles and hindrances in
our wills of this sort that we must overcome.
I think libertarians have traditionally tried to ignore this aspect of inde-
terminism. They knew it was required, but assumed it could be entirely cir-
cumvented by special agencies and would have no real impact on free will.
But hindrances and obstacles and resistance in the will are precisely what is
needed for free will, which, like life itself, exists near the edge of chaos. This
is an aspect of what is called in religious terms the ‘problem of evil’. Recall
Kant’s image of the bird that is upset by the resistance of the air and the
wind to its flight and so imagines that it could fly better if there were no air
at all to resist it (Kant 2003). But of course it would not fly better, but cease
to fly at all. So it is with indeterminism and free will. It provides resistance
to our choices, but a resistance that is necessary if we are to have free will.
I conclude with one final objection. Even if one granted that persons,
such as the businesswoman, could make genuine self-forming choices that
were undetermined, isn’t there something to the charge that such choices
would be arbitrary? A residual arbitrariness seems to remain in all self-
forming choices since the agents cannot in principle have sufficient or
overriding prior reasons for making one option and one set of reasons pre-
vail over the other. This is again one of those truths that tell us something
important about free will. In this case it tells us that every undetermined
self-forming free choice is the initiation of what I have elsewhere called a
‘value experiment’ whose justification lies in the future and is not fully ex-
plained by past reasons. In making such a choice we say, in effect, “Let’s try
this. It is not required by my past, but it is consistent with my past and is
one branching pathway my life can now meaningfully take. Whether it is
the right choice, only time will tell. Meanwhile, I am willing to take
responsibility for it one way or the other” (see Kane 1996, 145–146).
The term ‘arbitrary’ as I have often noted comes from the Latin arbitrium,
which means ‘judgment’—as in liberum arbitrium voluntatis, ‘free judgment
of the will’ (the medieval philosophers’ designation for free will). Imagine
a writer in the middle of a novel. The novel’s heroine faces a crisis and the
writer has not yet developed her character in sufficient detail to say exactly
how she will act. The author makes a ‘judgment’ about that which is not
yet determined by the heroine’s already formed past that does not give
unique direction. In this sense, the judgment (arbitrium) of how she will
react is ‘arbitrary’, but not entirely so. It had input from the heroine’s fic-
tional past and in turn gave input to her projected future. In a similar way,
agents who exercise free will are both authors of and characters in their
own stories all at once. By virtue of ‘self-forming’ judgments of the will
(arbitria voluntatis) (SFAs), they are ‘arbiters’ of their own lives, ‘making
themselves’ out of a past that, if they are truly free, does not limit their future
pathways to one. In reply to the charge that they did not have sufficient or
conclusive prior reasons for choosing as they did, they may respond: “True
enough. But I did have good reasons for choosing as I did, which I’m willing
to stand by and take responsibility for. If they were not sufficient or conclu-
sive reasons, that’s because, like the heroine of the novel, I was not a fully
formed person before I chose (and still am not, for that matter). Like the
author of the novel, I am in the process of writing an unfinished story and
forming an unfinished character who, in my case, is myself.”
Notes
1. See especially Kane (1996), which provides an overview of philosophical debates

about all four questions over the past fifty years and further development of many of
the ideas of this paper. Also, see an earlier work (Kane 1985).
2. For a formal statement and defense of this condition, see Kane (1996, ch. 3).
3. For defenses of this claim by these authors see the readings in Kane (2002) by
Dennett, Fischer, and Pereboom.
4. Readable and accessible introductions to the role of neural networks (including

recurrent networks) in cognitive processing include Churchland (1995) and Spitzer
(1999).
5. We have to make further assumptions about the case to rule out some of these
conditions. For example, we have to assume, no one is holding a gun to the woman’s
head forcing her to go back, or that she is not paralyzed, etc. But the point is that
none of these conditions is inconsistent with the case of the woman as we have
imagined it. If these other conditions are satisfied, as they can be, and the business-
woman’s case is otherwise as I have described it, we have an SFA. I offer the complete
argument for this in Kane (1996, ch. 8).
6. I show in greater detail that each of these conditions can be satisfied by SFAs in
Kane (1996, ch. 8).
7. That some such motivational system is necessary to define personhood and

agency has been persuasively argued by Fred Dretske (1988), David Velleman (1992),
88 R. Kane
and Owen Flanagan (1992). In Kane (1996, 137–142) I call the realization of such a
system in the brain the ‘self-network’.
References
Aristotle. 1985. Nicomachean Ethics. Translated by Terence Irwin. Indianapolis, Ind.:

Hackett.
Austin, J. L. 1956. “Ifs and Cans.” Proceedings of the British Academy 42: 109–132.
Churchland, P. M. 1995. The Engine of Reason, the Seat of the Soul. Cambridge, Mass.:
MIT Press.
Clarke, R. 1993. “Toward a Credible Agent-causal Account of Free Will.” Noûs 27:
191–203.
Dennett, Daniel D. 1984. Elbow Room: Varieties of Free Will Worth Wanting.
Cambridge, Mass.: MIT Press.
Dretske, F. 1988. Explaining Behavior: Reasons in a World of Causes. Cambridge, Mass.:

MIT Press.
Flanagan, O. 1992. Consciousness Reconsidered. Cambridge, Mass.: MIT Press.
Kane, R. 1985. Free Will and Values. Albany, N.Y.: State University of New York Press.
. 1989. “Two Kinds of Incompatibilism.” Philosophy and Phenomenological

Research 50: 219–254.
. 1996. The Significance of Free Will. New York: Oxford University Press.
, ed. 2002. The Oxford Handbook of Free Will. Oxford and New York: Oxford
University Press.
Kant, I. 2003. Critique of Pure Reason. Tr. N. K. Smith. Basingstoke: Palgrave MacMillan.
Leibniz, G. W. 1991. The Monadology. In his Philosophical Essays, translated by

R. Ariew and D. Garber. Indianapolis, Ind.: Hackett.
O’Connor, T. 2000. Persons and Causes: The Metaphysics of Free Will. New York:
Oxford University Press.
Spitzer, M. 1999. The Mind within the Net. Cambridge, Mass.: MIT Press.
Velleman, D. 1992. “What Happens When Somebody Acts?” Mind 101: 461–481.
4 Trying to Act
Carl Ginet
My topic is the concept of trying to perform this or that specific action. By an

action I mean an event that begins with the agent’s volition to exert force
with his body in a specific way (or, in the limiting case, in such a volition).
Paradigmatic actions are, for example, raising a hand, rising from a chair,
throwing a ball, opening a door, turning on a faucet, and uttering a sentence.
Thus in being concerned with what it is to try to act in some particular way, I
am not concerned with, for example, what it is to try to see (or hear) some-
thing or to try to remember (or forget) something or to try to feel glad (or sad).
Nor am I concerned with what it is to try not to act in a certain way. No doubt
these other things all have interesting conceptual connections to trying to
act but exploring those connections is not on the agenda for this paper.
1 Preliminaries
First, let me state some preliminary points, before I get to the main ques-
tion about trying to act that I want to talk about. These preliminary points
are not all uncontroversial, and I’ll give a bit of an argument for some of
them, but my purpose in stating them is not to persuade you of them, but
to let you know what I am assuming in the rest of my discussion.
1. I individuate actions in such a way that, for example, my action of turn-
ing on my computer by pressing a button on the keyboard is a causally
complex event that has, as a part, a distinct and more basic action of my
pressing the button (as well as a nonaction part caused by that more basic
action, namely, the computer’s turning on); and that more basic action has
as a proper part a still more basic action of my exerting a certain force with
my hand and index finger; and the latter action has as a proper part a most
90 C. Ginet
basic action of my willing to exert such a force. Thus my way of individu-

ating actions is not the minimizing extreme that says that the only actions
are the most basic ones. (Also, let me add, it is not the maximizing extreme
that says that any two nonequivalent action descriptions pick out distinct
actions, but I’ll not try to explain how I avoid that extreme.)1
2. The proposition expressed by a sentence of the form ‘S tried to A’ (where
‘A’ is a verb phrase signifying an action) does not entail that S A-ed. Nor
does that proposition entail that S did not A. When one tries to do some-
thing one can succeed.
3. Trying to do something is always itself constituted by acting in some way
or other. That is, the proposition ‘S tried to A’ entails that S performed some
action. And the action that S performed cannot have been merely S’s
deciding to A or any other act of mere deciding (mere intention forming).
To try to do something, it is not enough just to have or form an intention.
4. The kinds of actions one can try to perform are limited to those one can
try to perform without succeeding in performing them. One can try to A
only if it makes sense to suppose that a person could unsuccessfully try to
A. This means that, although one can try to open the door or try to exert
force with one’s hand, one cannot try to will to exert force with one’s hand
and one cannot try to decide to open the door (where ‘decide’ refers to the
simple mental act that concludes deliberation rather than to the process of
deliberation itself: one can, of course sensibly say, ‘I am trying to decide
whether to take that job’, if one means that one is trying, by deliberating
about the matter, to reach a decision).2
5. Given that A-ing is a kind of action one can try to perform, whenever it
is true that S A-ed intentionally and with the intention of A-ing, it is also
true that S tried to A. This is true despite the fact that in many contexts
where a sentence of the form ‘S tried to A’ is uttered, the speaker intends her
utterance to convey the information either that S failed to A or that S
thought there was some chance of failing to A. Even when something of this
sort is implicated by the utterance of ‘S tried to A’, it is not asserted and not
entailed by the proposition expressed by the utterance. This is shown by the
fact that a speaker can make true assertions with a two-sentence discourse
like the following: ‘I’ve never had a paralyzed arm. Whenever I’ve tried vol-
untarily to move my arm, I’ve succeeded in doing so’. Such a speaker need
not intend their quantifier ‘whenever’ to quantify only over those past cases
Trying to Act 91
of their voluntarily moving their arm where they thought they might not
succeed; it’s perfectly natural for it to be intended and interpreted as quan-
tifying over all past cases of their voluntarily moving their arm. And when
it is so interpreted, we would have no difficulty in taking what the speaker
asserts to be true, even if we know that in a great many of the cases quanti-
fied over the speaker had no thought of the possibility of failing in their
attempt to move their arm.
6. The verb ‘try’ does not signify a special kind of action.3 In particular,
contrary to what some philosophers have claimed,4 it is not the case that
trying is always a most basic action. When we have the thought that S tried
to A, the action we think of as constituting her attempt to A—what she did
that was her trying to A—need not be the most basic action she performed.
This is shown by the acceptability of such an utterance as ‘The same
attempt to turn on the computer I made earlier—namely, my pushing
the “on” button on the keyboard—succeeded this time in turning it on’.
Here the speaker implies that what the attempt consisted in was an action
of pressing a certain button; and such an action is not most basic (because
it includes the more basic action of the agent’s moving his hand and finger
in a certain way).
2 A Sufficient Condition: Intending the Action
Let me now turn to the question which it is my main purpose in this paper
to address. What is the mental component of trying to act? It seems evident
that an action can count as the agent’s trying to do a certain thing only if
the agent has an appropriate intention concerning that action. What
counts as an appropriate intention?
It’s fairly obvious that if S intended her action to be an attempt to do a
certain thing, then in that action S was trying to do that thing. That is, a
sufficient condition for the truth of ‘By Z-ing S tried to A’ is the following:
(S0) S intended of her Z-ing that in so acting she would be trying to A.

An obvious problem with S0 as an explication of the concept of trying to
do something is that it employs the concept being explicated. To say that in
Z-ing S tried to A if S intended in Z-ing to try to A is, to be sure, not absolutely
trivial and uninformative; but we’d like to find more informative sufficient
conditions, ones that do not employ the concept of trying to do something.
92 C. Ginet
One informative sufficient condition is fairly obvious and simple: an

action will count as an attempt to do a certain thing if the agent intends of
the action that by it she will do that thing. That is, the following is a sufficient
condition for the truth of a proposition of the form ‘By Z-ing S tried to A’:
(S1) S intended of her Z-ing that by so acting she would A.

This is, I hazard to say, a minimally sufficient informative condition. That
is, no weaker condition, one entailed by but not entailing it that also
eschews use of the concept of trying, is sufficient. One thing entailed by
but not entailing S1, is that S has a certain belief about her chances of A-ing
by Z-ing. S1 entails the following:
(B1) S believed of her Z-ing that by so acting she might A (that is, there
was a serious chance she would A).
This does not entail S1. This is shown by any case where S knowingly takes
a chance that her Z-ing might bring about certain consequences, but does
not intend by Z-ing to bring them about. For example, Hector might plug in
the new electric heater, knowing that there is a chance that by so doing he
might blow a fuse, but not with the intention of blowing a fuse. Such
examples show also that B1 does not entail that by Z-ing S tried to A. In plug-
ging in the new heater, Hector was not trying to blow a fuse. This nonen-
tailment supports the claim that S1 is a minimally sufficient condition for
the truth of ‘By Z-ing S tried to A’. At any rate, the weaker B1 is not sufficient.
But S1, though minimally sufficient, is not also necessary.5 Indeed, not
even the weaker B1 is necessary. Suppose Sam is confronted with a large log
that he would like to move. Suppose that Sam has no idea whether he can
move the log by just pushing on it but wants to find out. So he pushes on
the log as hard as he can. That is, he tries to move the log by pushing on it,
in order to find out whether he can move it by that means. He cannot
intend that his pushing will move the log, because he does not (yet) believe
that he might move it by pushing on it, that there is a chance that his push-
ing on it would succeed in moving it. For all he yet knows, it may be far too
heavy for him, or any human being, to move it just by pushing on it; and
it would be semantically unacceptable to say of him that, although he does
not know (and does not believe) that it is possible for him to move it by
pushing on it, his intention/aim/purpose in pushing on it is to move it.
So, for S’s Z-ing to count as S’s trying to A, though it is sufficient that S
intended of her Z-ing that by it she would A, it is not necessary. Some other
Trying to Act 93
sort of intention, with a different content, must be sufficient to make it the

case that by Z-ing S was trying to A. What other sort? The answer to this
turns out to be less simple than one might have hoped.
3 A Second Sufficient Condition: Intending to Find Out
One can start with the following fairly obvious thought: even if S did not
intend that by her Z-ing she would A, S must have intended that her Z-ing
have some sort of connection with the possibility of her A-ing. S must have
an intention whose content relates her Z-ing to the possibility of her
thereby A-ing in some way or other. The complications come when one asks
what specific content or contents for this intention will make an intention
such that, if S had that intention when Z-ing, then in Z-ing tried to A. There
are, it turns out, several different contents that will do the trick.
One we have already seen is our previous example where Sam’s pushing
on the log counted as his trying to move it, despite his not intending that
his pushing would move it. Though he did not intend to move it, he did in-
tend that his pushing on the log would show him whether he was able to
move it by that act of pushing on it.6 And, it seems clear, it is the presence
of this intention that makes it right to count his pushing as his trying to
move the log. After all, if he had not had any such intention having to do
with the possibility of moving the log, but had instead intended only that
by pushing on the log he would exercise his muscles, then it would clearly
not be right to count his pushing on the log as his trying to move it.
This example shows us a second informative sufficient condition for the
truth of ‘In Z-ing S tried to A’:
(S2) S intended of her Z-ing that by so acting she would find out
whether she could A by that Z-ing.
To have the intention ascribed to her in S2, S must believe that by Z-ing she
might find out whether she can A by that Z-ing. That is, S2 entails
(B2) S believed of her Z-ing that by so acting she would find out
whether she could A by that Z-ing.
But B2 does not entail S2. Suppose that Sam doesn’t know whether he can
move the log by pushing on it and, moreover, he doesn’t care. He pushes
on the log merely to exercise his muscles, but it occurs to him as he begins
to do so that this action might show him whether he can move the log by
94 C. Ginet
pushing on it. So he believes of his pushing that by it he might find out

whether he can move the log by that pushing, but it is not among his
intentions in pushing on it to find this out. And therefore finding this out
is not something he is trying to do by pushing on it. This supports a claim
that S2 is minimally sufficient for the truth of ‘In Z-ing S tried to A’; at any
rate the weaker B2 is not sufficient.
Let me note that S2 covers an interesting class of cases of trying to A
while not intending to A that was first noted by Michael Bratman (1984).
These are cases where S both Zs and Xs and intends to make it the case
that either she As by Z-ing or Bs by X-ing, but does not intend to A by Z-
ing and does not intend to B by X-ing: she intends the disjunction but
does not intend either disjunct. This is because she does not know which
attempt, the Z-ing or the X-ing, will succeed and believes that not both
can. For example, S knows that the double doors before her open either to-
ward her or away from her but not both, but she does not remember
which way it is. So S pushes on the left door with her left hand and pulls
on the right door with her right hand, with the intention that she will
either open the left one by pushing on it or open the right one by pulling
on it. In such a case S neither intends in pushing on the left door to open
it nor intends in pulling on the right door to open it: for her to have
both intentions would be irrational, given her belief that she cannot do
both things, and there is no reason to attribute to her one intention rather
than the other. But she is both trying to open the left door by pushing on
it and trying to open the right door by pulling on it. And this is made true,
I suggest, by the fact that she intended of her pushing on the left door to
find out whether she can open it by those means, and likewise for her
pulling on the right door.
4 A Third Sufficient Condition: Intending to Show Inability
Is the disjunction ‘S1 or S2’ necessary for the truth of ‘By Z-ing S tried to A’?
No. Suppose Sam’s five-year-old daughter Suzy comes along and asks her
father to push on the log and move it. Sam, having ascertained that the log
is far too heavy for him to move, nevertheless pushes against it as hard as
he can in order to show Suzy that he cannot move it by pushing on it. It
would be correct to characterize Sam’s pushing on the log as his trying to
move the log, indeed trying as hard as he can to move it. He makes an
Trying to Act 95
attempt to move the log in order to show Suzy that he cannot move it; and, of
course, knowing that he cannot move it, he cannot intend of his pushing
on it that he will thereby move it. To intend that one’s attempt will suc-
ceed, one need not believe that it will succeed, but one must not believe
that it is impossible that it will succeed: one must believe that it has some
chance of success.
From this example we learn a third minimally sufficient informative con-
dition for the truth of ‘In Z-ing S tried to A’, namely, the following:
(S3) S intended of her Z-ing that by so acting she would show R (some
particular person or persons in the context) that she could not A by that
Z-ing.
S3, like S1 and S2, entails a belief of S’s about how her Z-ing is related to the
possibility of her thereby A-ing. S3 entails
(B3) S believed of her Z-ing that she would not A by so acting and
might persuade R that she could not A by that Z-ing.
But B3 does not entail S3. Suppose that while Sam is pushing hard on the
log in order to exercise his muscles it occurs to him that his so doing might
persuade Suzy, who is looking on, that he cannot move the log by pushing
on it. Still it need be no part of his plan that by pushing on the log he will
persuade Suzy of his inability. And if that is so, then, clearly, he is not try-
ing by pushing on the log to persuade Suzy of his inability. This supports a
claim that S3 is minimally sufficient for the truth of ‘By Z-ing S tried to A’.
At any rate, the weaker condition B3 is not sufficient.
5 A Fourth Sufficient Condition: Intending to Comply with a Command
Are we finished? Are S1–S3 all of the different minimally sufficient condi-
tions for the truth of ‘By Z-ing S tried to A’? I’m afraid not.
Suppose Sam knows that the log is far too heavy for him to move by push-
ing on it. But Sam is a soldier under the command of Rudy, and Rudy has or-
dered Sam to try to move the log by pushing on it. Sam would very much
prefer that Rudy not find out that he (Sam) cannot move the log by pushing
on it, but the penalty for refusing to obey a command would, in Sam’s view,
be a worse consequence than Rudy’s knowing that Sam cannot move the
log by pushing on it. So Sam pushes on the log as hard as he can in order to
comply with Rudy’s command to try to move it by pushing on it, but very
96 C. Ginet
much regretting—certainly not intending—that by so doing he will reveal

to Rudy that he cannot move it by pushing on it. And surely Sam’s pushing
on the log does comply with Rudy’s order. He does try to move the log by
pushing on it. He does so, despite not intending that he will thereby move
it (he knows that it’s quite impossible for him to move it by pushing on it),
despite not intending that he will thereby show someone that he cannot
move it (he regrets that this will be revealed), and despite not intending to
find out whether he can move it (he already knows that he cannot).
So this example reveals a fourth sufficient condition for the truth of ‘By
Z-ing S tried to A’:
(S4) S intended that her Z-ing would comply with a command

(or request) that S try to A by Z-ing.
What belief about her Z-ing does this intention require that S have? Pretty
clearly it requires that
(B4) S believed that her Z-ing would comply with a command

(or request) to try to A by Z-ing.
It’s not clear to me, however, that in this case the entailed belief is a weaker
condition than the intention that entails it.
For one thing, it is unclear to me how one could believe that one’s action
complies with a command without intending that it comply with it, given
that ‘complying with a command’ means acting with the intention that
one’s action be what was commanded. Of course, it is possible to believe
that one’s action will satisfy the description given in a command or request
without intending to be complying with that command and even without
intending to be acting in such a way as to satisfy that description. One
might be acting entirely out of other motives.
Consider, for example, the parent of a child who has a game in which she
“requests” the parent to do what she knows the parent is going to do any-
way: “Please eat breakfast this morning.” “Please go to work today.” The
parent knows that his or her actions will satisfy these requests, but to sat-
isfy them is not among the parent’s intentions in so acting.
In this case, though, the parent does intend that her action will satisfy the
description that is in fact given in the child’s request; for example, that it
will be her eating breakfast. But not even this intention need be present in
an agent who believes at the time of her action that in the circumstances
it will satisfy the specification given in a request to her. Suppose Suzy is a
Trying to Act 97
guest for dinner at a friend’s house. She approaches the only remaining
vacant chair at the table with the intention to sit in it. Just as she is about to
sit, Rudy, who is already seated and whom Suzy intensely dislikes, says
“Please sit where I can gaze at your beautiful profile.” As Suzy sits she knows
that her action will satisfy Rudy’s description—it will be sitting where he
can gaze at her profile—but she certainly does not intend of her sitting that
it satisfy that description.
Matters seem different, however, when the description in the command
or request has the form that ‘S try by Z-ing to A’. For example, suppose Suzy
regularly exercises her muscles by pushing against a certain large tree outside
her house. Her daughter knows this, and as Suzy approaches the tree, her
daughter says, “I order you to try to uproot that tree by pushing on it.” Now
at the time of her action Suzy might or might not have the belief she would
express by saying, “In pushing on this tree I am obeying my daughter’s order
to try to uproot it by pushing on it”; but if she does have that belief, then she
believes that in pushing on the tree she is trying to uproot it. And if she be-
lieves that, it is hard to see how she could at the same time lack the intention
in pushing on the tree to try to uproot it. How could she think she is trying
and not intend to be trying? And if she has that intention then she is trying
to uproot it by pushing on it.
There are, of course, other sorts of beliefs entailed by S4 that clearly do
not entail it. For example, S can have the intention attributed by S4 only if
S believes that she is Z-ing. But this belief by itself does not entail that S’s
Z-ing is an attempt to A. So there seems no reason to suspect that S4 is not
a minimally sufficient condition for the truth of ‘By Z-ing S tried to A’.
But, as you have no doubt long since been aware, there is a problem with
S4: it is not as informative a condition as we’d like to have. It employs the
concept we wish to use it to explicate, the concept of trying to do some-
thing. Is there any way we can refine S4 so as to rid it of this feature?
Let us approach this problem by considering the following question: if R
asked S to try to A by Z-ing, what must R have believed about the possibility
of S’s A-ing by Z-ing for this to be a serious request, one that R thought it was
possible for S to comply with? It seems to me, for starters, that R need not
have been joking if R had any one of the following beliefs:
R believed that by Z-ing S might A.
R believed that S believes that by Z-ing she might A.
98 C. Ginet
R believed that by Z-ing S would show someone, who needs convincing

of this, that S cannot A by Z-ing.
R believed that S believes that by Z-ing she would show someone, who
needs convincing of this, that she cannot A by Z-ing.
But these are not the only sorts of beliefs on R’s part that would show that R
thinks it possible for S’s Z-ing to be S’s trying to A. Consider again the situa-
tion where Sam is a soldier under Rudy’s command. Being sadistic and en-
joying for its own sake the exercise of his authority over Sam, Rudy seriously
commands Sam to try to move the log by pushing on it as hard as he can,
even though Rudy does not believe that there is any chance at all that Sam
could move the log by pushing on it or that Sam believes that there is; and,
moreover, Rudy does not believe, or believe that Sam believes, that there is
someone around who needs to be persuaded of the fact that Sam cannot
move the log by pushing on it. (For a less dark story, imagine that Rudy is not
Sam’s commander but instead an eccentric friend who has offered Sam a
large sum of money for just trying to move the log by pushing on it.)
On the other hand, not just any action that Sam could perform is such
that Rudy can seriously command Sam to try to move the log by perform-
ing that action, given that Rudy fails to believe (or believe that Sam be-
lieves) that there is any chance at all that Sam could move the log by that
action or to believe (or believe that Sam believes) that there is someone
around who needs to be persuaded that Sam cannot move the log by
that action. For example, Rudy cannot seriously command Sam to try to
move the log by whistling Dixie, if Rudy does not believe (or believe that
Sam believes) that there is any chance at all that Sam could move the log
by whistling Dixie or that there’s someone around who needs to be per-
suaded that Sam cannot move the log by whistling Dixie. What is the
difference between pushing on the log and whistling Dixie, vis-à-vis the
possibility of moving the log by such actions, that explains this?
The difference, I think, is that, although Rudy does not believe that Sam
can move the log by pushing on it, Rudy does believe (or believes that Sam
believes) that pushing on the log is a sort of action that might, if it were
performed with sufficient strength or skill or in the right manner, succeed in
moving the log; whereas Rudy does not believe (or believe that Sam believes)
that whistling Dixie is a sort of action that might, if performed with suffi-
cient strength or skill or in the right manner, succeed in moving the log.
Trying to Act 99
Note that the belief that pushing on the log is a sort of action that might,
if it were performed with sufficient strength or skill or in the right manner,
succeed in moving the log is entailed by the belief that Sam might move
the log by pushing on it: one who believes the latter believes the former.
The general lesson of these observations is that, when R orders S to try to
A by Z-ing, S can take this order seriously (can suppose that R thinks it is
possible for S to comply with the order) only if S thinks that R believes (or
believes that S believes) that Z-ing is a sort of action by which someone
might succeed in A-ing if they performed it with sufficient strength or skill
or in the right manner, or S thinks that R believes (or believes that S
believes) that by Z-ing S would show someone that S cannot A by Z-ing.
It is not necessary that the command (or request) have been that S try to
A by Z-ing. The command might, for instance, have been just that S try to A
or just that S Z. If the content of Rudy’s order were simply that Sam try to
move the log, and in response Sam pushes on the log as hard as he can with
the intention and belief that his so doing will comply with Rudy’s order
and that it will do so because Rudy’s reason for giving the order was his
belief (or belief that Sam believes) that pushing on the log, if performed in
the right manner with sufficient strength or skill, is a sort of action by
which someone might succeed in moving the log, then it follows that in
pushing on the log Sam was trying to move it.
It is likewise if the content of Rudy’s order were simply that Sam push on
the log. If in response Sam pushes on the log as hard as he can with the
intention and belief that his so doing will comply with Rudy’s order and
that it will do so because Rudy’s reason for giving the order was his belief (or
belief that Sam believes) that pushing on the log, if performed in the right
manner with sufficient strength or skill, is a sort of action by which some-
one might succeed in moving the log, then it follows that in pushing on
the log Sam was trying to move it.
Or suppose that in response to either of those orders Sam pushes on the
log with the intention and belief that his so doing will comply with Rudy’s
order and will do so because Rudy’s reason for the order was that he believes
(or believes that Sam believes) that by pushing on the log Sam would show
someone that Sam cannot move the log by pushing on it. Again it follows
that in pushing on the log Sam was trying to move it.
What if Rudy’s command were that Sam should move the log by pushing
on it and, as before, Sam believes that it is impossible for him to move it by
100 C. Ginet
pushing on it?7 Then Sam cannot intend that his pushing on it will com-
ply with the order. The most he can intend is that it will be as good as he can
do to comply with the order. But if he does intend this latter, then his push-
ing on it will be his trying to move it by pushing on it.
Generalizing, we can say the following: If S believes (and intends) that
her Z-ing will comply with the command, or will be as good as she can do
to comply with it, and it will be so because the giver of the command
believes (or believes that S believes) that Z-ing, if performed with sufficient
strength or skill or in the right manner, is a sort of action by which some-
one might succeed in A-ing then in Z-ing S is trying to A.
And if S believes (and intends) that her Z-ing will comply with the com-
mand, or will be as good as she can do to comply with it, and it will be so
because the giver of the command (or request) believes (or believes that S
believes) that by Z-ing S would show someone that S cannot A by Z-ing,
then in Z-ing S is trying to A.
These conditionals hold whether or not the agent S believes that there is
any chance she might A by her Z-ing, or believes that there is someone
around who needs showing that S cannot A by Z-ing, or that Z-ing, if per-
formed with sufficient strength or skill or in the right manner, is a sort of
action by which someone might succeed in A-ing. Thus, for example, if
Rudy gives Sam the order to try to move the log by whistling Dixie and Sam
thinks Rudy meant the order seriously because Rudy thinks that whistling
Dixie in the vicinity of a large heavy object, if performed in exactly the
right manner, is a sort of action by which someone might succeed in mov-
ing the object, then Sam can believe and intend that by whistling Dixie he
is complying with Rudy’s order, even if Sam does not share Rudy’s belief
about the object-moving power of whistling Dixie.
This result gives us a way to formulate an informative condition that
covers those new cases of trying to A by Z-ing that S4 brings in. The for-
mulation is the following:
(S4) S intended of her Z-ing that it would comply with a command (or
request) to her and S also either (i) intended that it would do so for the
reason that the giver of the command (or request) believes (or believes
that S believes) that Z-ing, if performed in the right manner with
sufficient strength or skill, is a sort of action by which someone might
succeed in A-ing or (ii) intended that it would do so for the reason that
Trying to Act 101
the giver of the command (or request) believes (or believes that S believes)
that by Z-ing S would show someone that S cannot A by Z-ing.
If in her Z-ing S has a complex intention satisfying the specification in
S4, then her Z-ing is her trying to A.
6 A Disjunctive Sufficient Condition That May Also Be Necessary
So we have uncovered four different minimally sufficient conditions for

the truth of ‘S tried to A’. Are there any others? I can’t think of any, but at
this point that does not strike me as compelling evidence that there aren’t
any others. So I will conclude only that each of these four conditions is suf-
ficient and leave it an open question whether their disjunction is necessary.
We can wrap it all up in one large statement as follows.
S tried to A if there was an action of S’s, S’s Z-ing, such that either
(S1) S intended of her Z-ing that by so acting she would A,
(S2) S intended of her Z-ing that by so acting she would find out
whether she could A by that Z-ing,
(S3) S intended of her Z-ing that by so acting she would show R

(some particular person or persons in the context) that she could not A
by Z-ing, or
(S4) S intended of her Z-ing that it would comply with a command (or
request) to her and S also either (i) intended that it would do so for the
reason that the giver of the command (or request) believes (or believes
that S believes) that Z-ing, if performed in the right manner with
sufficient strength or skill, is a sort of action by which someone might
succeed in A-ing or (ii) intended that it would do so for the reason that
the giver of the command (or request) believes (or believes that S believes)
that by Z-ing S would show someone that S cannot A by Z-ing.
S1–S4 specify four different intention contents each of which is such that,
if S’s Z-ing is accompanied by an intention with that content, then S’s Z-ing
is her trying to A. As far as I can discern, the only feature these intention con-
tents have in common is that each specifies some relation or other between
S’s Z-ing and the possibility of S’s A-ing. It would be nice in a way—in that it
would be more in accord with our preanalytic hopes—if there were some-
thing more specific tying together the sufficient conditions for trying to act.
102 C. Ginet
But it is nice in another way—in that it is more interesting—that this

concept is as complex and nonunified as this analysis makes it out to be.
Notes
1. For a fuller explanation of my view on how to individuate actions, see Ginet

(1990, ch. 3).
2. Here I agree with McCann (1975).
3. Here again I agree with McCann (1975).
4. For example, O’Shaugnessy (1973) and Hornsby (1980).
5. Here I disagree with McCann (1991).
6. I say ‘by that act of pushing on it’, which refers to the particular action he per-
formed, rather than ‘by pushing on it’, which refers to the type of action, because
there is the following possibility (brought to my attention by Richard Reilly): Sup-
pose Sam decides that he will move the log off the driveway later, after lunch, if he
can. Sam knows that, if he can move the log at all by pushing on it, he can move it
all the way off the driveway by (a sustained period of) pushing on it. So he pushes
hard on the log for a few seconds and it moves a little bit, whereby he finds out that
he can move it off the driveway by pushing on it. But he did not try to push it off the
driveway, because it was not his intention in pushing on it to find out whether he
could push it off the driveway by that act of pushing on it.
7. I’m grateful to Harry Silverstein for calling my attention to the import of this
possibility.
References
Bratman, M. 1984. “Two Faces of Intention.” Philosophical Review 93: 375–405.
Ginet, C. 1990. On Action. New York: Cambridge University Press.
Hornsby, J. 1980. Actions. London: Routledge and Kegan Paul.
McCann, H. 1975. “Trying, Paralysis, and Volition.” Review of Metaphysics 28:

423–442.
. 1991. “Settled Objectives and Rational Constraints.” American Philosophical

Quarterly 28: 24–36.
. 1998. The Works of Agency. Ithaca, N.Y.: Cornell University Press.
O’Shaugnessy, B. 1973. “Trying as the Mental ‘Pineal Gland’.” Journal of Philosophy

70: 365–386.
The Metaphysics of Moral Responsibility
5 The Sense of Freedom
Dana K. Nelkin
1 Introduction
We appear to have an inescapable sense that we are free, a sense that we

cannot abandon even in the face of powerful arguments that this sense
is illusory. It has often been suggested, perhaps most notably by Kant
(1781Ⲑ1965 and 1785Ⲑ1981), that we have such a sense of freedom in virtue
of being rational deliberators. More precisely,
(R) Rational deliberators, in virtue of their nature as rational

deliberators, necessarily have a sense that they are free.
This claim has a great deal of intuitive plausibility: it is when we engage in

rational deliberation about what to do that we are most likely to become
aware that we have a sense of ourselves as free. Even skeptics about free
will, who are convinced that they are not free, often admit to a sense that
they are free when they deliberate about what to do.1
(R) constitutes a rare point of agreement among philosophers who are
deeply divided on the question of whether we really are free. But while
there are skeptics who accept (R) and lament the fact that we are thereby
condemned to live under an illusion, there are others who have much
higher hopes for (R): they take (R) to be the first major step in arguments
that we are in fact free. For example, after claiming that we have a belief in
our own freedom “so necessary in most of our rational operations,”
Thomas Reid writes that
This natural conviction of our acting freely, which is acknowledged by many who
hold the doctrine of necessity, ought to throw the whole burden of proof upon that
side . . . . (Reid 1788/1983, 344)
106 D. K. Nelkin
and Kant writes,

Now I say that every being which cannot act in any way other than under the idea
of freedom is for this very reason free from a practical point of view. (Kant
1785/1981, 50 [GW 448])2
It is important to determine whether (R) is true, then, for at least two rea-
sons. First, if (R) is true and we are not free, then one of our most central
conceptions of ourselves is both false and inescapable. On the other hand,
if Reid and Kant are correct, and (R) can be shown to be true, then a major
premise can be established in an argument that we are in fact free.3
Is (R) true? In this paper, I will explore two lines of reasoning supporting
the conclusion that it is. I believe that one of these lines is promising, and
will suggest some different ways in which it might be developed. I begin by
offering a more detailed account of the sense of freedom attributed to all
rational deliberators. As might be expected, philosophers have understood
(R) in different ways. In section 2, I examine what I call the ‘indeterministic’
reading, according to which the sense of freedom in (R) is the belief that
one can choose from a set of undetermined alternative courses of action.
Although I argue that, on the indeterministic reading, (R) is false, I believe
the ways in which it has been defended contain important insights. In
section 3, I argue that we should adopt what I call the ‘belief-concept’ read-
ing of (R), according to which the sense of freedom in (R) is the belief that
one’s actions are up to one in such a way that one is accountable for them.
2 The Indeterministic Reading of (R)
According to the indeterministic reading of (R), all rational deliberators nec-

essarily believe that they can choose from multiple future courses of action,
none of which is determined. Perhaps the earliest adherent of the indeter-
ministic reading is Aristotle. In the Eudemian Ethics, Aristotle claims that
about other [things] not only existence and non-existence is possible, but also
human deliberation; these are things the doing and not doing of which is in our
power. (Aristotle 1984b, 1942 [EE 1226a, 26–27])4
As stated in this passage and elsewhere in Aristotle’s work, it initially

appears that Aristotle is making the strong claim that indeterminism is
required for deliberation rather than the weaker claim that a belief in inde-
terminism is required for deliberation. However, in several places, Aristotle
The Sense of Freedom 107
qualifies his claims about the proper objects of deliberation in a way that
suggests that deliberation requires the belief that the object of deliberation
possesses certain properties rather than that the object of deliberation
actually possesses those properties.5 What is essential to deliberation are the
subject’s epistemic attitudes about the object of deliberation, and, in par-
ticular, the belief that any object of one’s deliberation is possible but not
determined.
While the inspiration for the indeterministic reading can thus be traced
to Aristotle, the view has many contemporary adherents, and some have
thought Kant to be another advocate.6 Despite its long history and current
popularity, I claim that the indeterministic reading is false.
Let me begin by setting out one line of reasoning that might be used to
support the indeterministic reading of (R).7
(i) If one deliberates about an action, A, then one must believe that it is
in one’s power to do and to forgo A.
But,
(ii) If one believes that it is in one’s power to do/forgo A, then one

must believe that there exist no conditions sufficient to render inevitable
either A or not-A.
Therefore,
(iii) If one deliberates about A, then one must believe that there exist
no conditions sufficient to render inevitable either A or not-A.8
Premise (i) has seemed to many to be an attractive thesis, but I will present
several thought experiments designed to undermine it. Then I will go on to
challenge what are perhaps even more basic assumptions on which the in-
deterministic view must ultimately rest: assumptions about the very nature
and purpose of rational deliberation.
First, consider the following case: imagine that you know that a brilliant
scientist has the ability to fiddle with your brain in a way that causes you
to act as she wishes you to. You know that she wants you to vote for Gore
over Bush in the upcoming presidential race, and that if you do not decide
to vote as she wishes, she will cause you to vote that way. So, for instance,
you know that if you were to prepare to vote for Bush or otherwise fail to
decide to vote for Gore, the brilliant scientist would cause you to vote
for Gore. It seems to me that you could still evaluate the reasons for voting
108 D. K. Nelkin
for each candidate and decide to vote for Gore on the basis of those
reasons. In this case, contrary to (iii), you know that conditions exist which
are sufficient for your voting for Gore, while you nevertheless deliberate
and decide to vote for him. Further, contrary to (i), you do not believe you
could forgo the action upon which you decide.9
Now one might object that your voting for Gore as a result of deliberation
and decision differs in kind from the action of voting for Gore as a result of
the brilliant scientist’s interference with your brain. One might think that
the actions are of different types, especially if one is convinced that types of
actions are individuated in part by the types of causes they have.10 If this is
so, the objection goes, then you do deliberate about something that you
believe you could do or forgo: a particular type of voting for Gore.
Although this objection has some force, I think that it can be met. For if
you accept the initial description of the thought experiment, what you
evaluate with a view to acting are the reasons for voting for Gore. The
objection, by contrast, has you evaluating reasons for voting for Gore on
the basis of deliberation and decision, where this action is something that
you could do or forgo. Even if this sort of deliberation is possible, it remains
a coherent part of the thought experiment that you deliberate and decide
to vote for Gore. Thus, additional objects of deliberation that might be
regarded as actions to be done or forgone within the thought experiment
do not alter the fact that there is something about which one deliberates
and decides, despite not believing that one could either do or forgo it.
The objection might be pressed in the following way, however: consider
a modified version of the original voting case (the ‘inevitability’ case) in
which you believe it to be inevitable that, no matter what, the scientist will
fiddle with your brain in such a way that you vote for Gore. To ensure that
the case is one in which it is inevitable that your action will have the same
cause in counterfactual circumstances, let it be built into the case that there
is no possibility of overdetermination: you believe that, no matter what,
the changes in your brain induced by the scientist will be the sole cause of
your voting for Gore. In this case, the objection proceeds, surely you can-
not deliberate, and the explanation is that you fail to believe that the
action in question—voting for Gore in a particular way—is one that you
could either do or forgo. If the lack of such a belief is the correct explana-
tion of the failure of deliberation in this case, then such a belief must be a
necessary condition for deliberation. Further, this conclusion shows that
there must be a mistake in the description of the original voting case. In

particular, the objection continues, the conclusion of the inevitability case
reveals that the supposed lesson of the original case rests on a mistaken as-
sumption about action individuation, a mistake that is successfully avoided
in the inevitability case.
The conclusion drawn about the inevitability case is questionable, how-
ever. It is clearly a case in which you believe that you will perform a particu-
lar action in a particular way and so lack the belief that you can either do or
forgo an action. It may also be that you cannot deliberate in this case. But it
is not the former point that explains the latter. Rather, the explanation for
your inability to deliberate in such a case is that you do not believe that the
deliberation and the resulting decision on your part could be causally effica-
cious. Possession of the belief that one’s deliberation and decision could be
causally efficacious is a necessary condition for deliberation, and the lack of
such a belief in this case is the explanation for any inability to deliberate.11
These two beliefs—that one’s deliberation could be causally efficacious
and that one’s future actions are undetermined—might easily be conflated.
The idea that one’s deliberation and decision caused one to perform an
action might be confused with the idea that one’s deliberation and decision
resulted in one’s making an undetermined possibility actual. But if we are
careful to distinguish between these two ideas, then we can resist the
present challenge to the original voting case.
At this juncture, the proponent of the indeterministic reading of (R)
might point out that even if the original voting example is effective in rais-
ing questions about (i) and so about (iii), it remains the case that it is not
an example in which the agent is said to believe that determinism is true.
While it is perhaps true that, in one sense, the example is a case in which
the agent believes that conditions exist which are sufficient for his or her
future actions, it is not true if by ‘conditions’ we mean ‘a previous state of
the world together with natural laws’.12
Yet there are cases in which individuals can rationally deliberate despite
lacking a belief in indeterminism. For example, consider the fact that there
are determinists who deliberate. More particularly, suppose that a person
believes that scientists have succeeded in creating a computer that predicts
her actions on the basis of past states of the world and natural laws. This
case is clearly one in which someone believes that there are conditions suf-
ficient for her performing the actual action she will perform and in which
110 D. K. Nelkin
she believes that her action is physically determined.13 The question is, can
such a person deliberate if she knows that such a computer is busy churn-
ing out its predictions? The answer seems to me to be “yes.” I can imagine
that such a computer exists right now somewhere in Siberia, printing out
predictions about my future actions. Even if I believed this to be true, and
so did not believe it possible for me to perform each of a number of alter-
native actions, I would go right on deliberating about all sorts of things.
To the case of the deliberating determinists, the following reply might be
offered. Such people do exist, and while they believe that determinism is
true, they also hold the contradictory belief that determinism is false.14
Thus, although there are those who believe that determinism is true, as long
as they continue to deliberate their behavior manifests the belief that they
can do or forgo the actions about which they deliberate, and thus manifests
the belief that determinism is false. In fact, it appears that the proponent of
the indeterministic reading could account for every purported counterexam-
ple to (i) or (iii) simply by attributing contradictory beliefs to every delibera-
tor who claims to have a belief contrary to that attributed in (i) or (iii).
However, the idea that deliberation manifests either of these beliefs can be
seen to rest on still more fundamental assumptions which can themselves
be questioned, assumptions about the nature and point of rational deliber-
ation. For we can ask: what are the conditions under which we take a piece
of behavior to manifest a belief? The answer is that, in general, we do so
when we think that the behavior is rationalized by a belief, that is, when we
think that the presence of the belief in question is part of the correct expla-
nation of the behavior that makes it intelligible in the context of the agent’s
beliefs, desires and actions.15 For example, consider the case of a store owner
who hires a private detective to watch and report on his cashier’s behavior.
In the absence of a belief that it is likely that the cashier is embezzling
money, the store owner’s action does not make sense in light of his other
actions and attitudes. Thus, this belief rationalizes his hiring of a detective.
Given this general understanding of what it would be for an activity to
manifest a belief, let’s consider why it might be thought that the beliefs
attributed to deliberators by the indeterministic view are manifested in the
activity of deliberation. I will focus directly on the belief attributed in (iii) for
the sake of simplicity, although the reasoning can easily be transposed to
account for the belief attributed in (i), as well. Here, then, is the question: in
what way can the belief attributed in (iii) be said to rationalize deliberation?
Either of two answers might be given, one having to do with the essen-
tial nature of deliberation, the other with its point, or purpose. Richard
Taylor provides material for the first sort of answer. He writes that if an
agent, Adam, believes that conditions already existed sufficient for his
performing the action he will perform in the future, then “he can no longer
deliberate about the matter because . . . he believes it is not up to him what
he does; the matter has already been ‘decided’, one way or the other, and
there is no decision for Adam to make” (Taylor 1964, 77). On one natural
reading of this passage, the purpose of deliberation is to decide upon an
action, where ‘decision’ is constituted by an agent’s closing all but one of
several previously ‘open’ possibilities. It follows that if conditions exist
sufficient for a particular action’s being performed, and hence that there
are no previously open possibilities to be closed or to be chosen to remain
open by the agent, then there can be no true decisions on the part of any-
one. It could then be argued that if an individual did not believe that his
deliberation could succeed in bringing about a decision, then his deliberat-
ing would not make sense. Therefore, the belief that he could make a
decision would rationalize his deliberation, and so that belief would be
manifested in the activity of deliberation. Finally, given the present under-
standing of ‘decision’, it follows that one can only deliberate if one believes
that there are open possibilities for one to close.
The problem with this suggestion is that the understanding of ‘decision’
on which it is based ought to be rejected. For on the view that a decision is
the closing of previously open possibilities, it is not strictly speaking true to
say that someone has reached a decision if, for some reason, it cannot be
carried out. Yet the naturalness of saying that someone made a decision
that could not be carried out suggests that we do not normally think of a
decision as being defined by its effects. Rather, it is more natural to say that
a decision is something like the formation of an intention which, depend-
ing on the circumstances, may or may not be carried out. According to this
latter understanding of ‘decision’, we can successfully decide to do some-
thing even if the decision does not succeed in effecting anything, let alone
in the closing of all but one of several previously open possibilities.
The second way in which the belief attributed in (iii) might be said to
rationalize deliberation involves a somewhat more modest view of the con-
nection between determinism on the one hand, and deliberation and deci-
sion on the other. The idea is that if deliberation and decision are to have a
112 D. K. Nelkin
point at all, then they must succeed in closing all but one of several previ-
ously open possibilities. This view is suggested by Aristotle, who, in arguing
that it would be absurd to think that everything that happens, happens of
necessity, claims that in that case “there would be no need to deliberate or
to take trouble” (Aristotle 1984d, 29 [DI 9, 18b31–2]).16
On this view, one could be said to make a decision even if determinism
were true, although in that case the decision would lack a point. It could
then be argued that if an agent failed to believe that there was a point to
making a decision, then it would not make sense for her to deliberate.
Therefore, deliberation manifests the belief that there is a point to making
a decision, and thus, that there are open possibilities to be closed by the
agent as a result of deliberation and decision.
But is it in fact true that rational deliberators can only take rational
deliberation to have a point if they take it that their decisions result in the
closing of all but one of several previously open possibilities? Of course, I
take it that the original voting case provides a negative answer here. For I find
it perfectly coherent that the agent in that case does not believe that her pur-
pose in deliberating is to change the course of events from what it would oth-
erwise be, or even to find out what she will do, and yet the possibility that
she deliberates is perfectly intelligible. Yet merely appealing to this sort of
case will not convince those who would attribute contradictory beliefs to the
voter. I must say what it is that the case reveals about our attitudes toward the
point of deliberation. And that is that reasons can motivate us to act in a cer-
tain way even if we know that we will act in that way no matter what. The
case shows us that the possibility of acting on the basis of good reasons can
itself be seen to have value, and so, one can take the activity of rational de-
liberation to have a point if one takes it that the activity can result in one’s
acting for good reasons. And the possibility that rational deliberators could
take deliberation to have a point in the absence of indeterminism under-
mines this kind of defense of the indeterministic reading. Thus, we have not
been given sufficient reason to conclude that the belief attributed in (iii) is
manifested by rational deliberation.
Each of the two attempts to provide an argument for the conclusion that
rational deliberation manifests the belief attributed in (iii) relies on a prob-
lematic premise: the first rests on a dubious assumption about the nature of
a decision, the second on a questionable assumption about the point of
deliberation. Of course, there may be reasons other than the two that I
have considered here for accepting the conclusion. However, the burden
rests with those who advocate the indeterministic reading; for the fact
that the purported counterexamples to it can only be resisted at the cost of
attributing contradictory beliefs to otherwise rational people provides
some prima facie reason to reject the view.17
Having made it clear which aspects of the reasoning to the indetermi-
nistic reading I wish to reject, let me emphasize an aspect of the reasoning
that I have not rejected. I believe that the general strategy of showing that
the sense of freedom rationalizes the very activity of rational deliberation
is a promising one, and I will pursue it in section 3. In particular, I will
return to the idea that for rational deliberation itself to be an intelligible
activity, deliberators must take such activity to have a point or purpose.
For even if deliberation can be seen to have a point in the absence of inde-
terminism, it does not follow that it has a point in the absence of freedom.
And if we refrain from prejudging the question of whether freedom
requires indeterminism, it is possible that, in an important sense of ‘free’,
we cannot take our deliberation to have a point if we do not have a sense
that we are free.
3 The Belief-concept Reading
The Sense of ‘Freedom’

To explain the sense of ‘freedom’ in the type of reading of (R) I favor, it will
be useful to borrow some terminology from John Rawls, and distinguish
between the concept of freedom and particular conceptions of freedom
(Rawls 1971, 5). First, there is a concept of freedom that all (or at least
many) people grasp, even if they disagree as to the necessary and sufficient
conditions for its instantiation. There must be such a concept if compati-
bilists, incompatibilists, and those who believe freedom impossible are
rightly said to disagree. Of course, it is possible to argue that the whole
problem with the free will debate is that the participants are indeed speak-
ing different languages, and that agreement could be reached immediately
if this fact were only acknowledged. I think that this view gives too little
credit to the participants in the debate, however, and that there is at least
one important concept about which there is disagreement.18
Second, there are also different analyses of freedom offered (such as the
ability to choose from among various undetermined actions or the ability
114 D. K. Nelkin
to act in the absence of constraint). These are conceptions of freedom,

attempts to spell out in detail the conditions under which actions are free.
It is easy to describe various conceptions of freedom, then. But what is
the concept of freedom? In my view, the concept of free agency is the
notion of one’s actions being up to one in such a way that one is, in a basic
sense, responsible or accountable for them.
This account must be clarified in various ways. First, let me emphasize
that in identifying the concept of freedom in this way, I mean to suggest
that the mutual implication between ‘free’ and ‘being up to one in such a
way that one is accountable’ should be recognized by anyone in possession
of the concept of freedom. Second, ‘responsibility’ and ‘accountability’,
like ‘freedom’, each have many different senses. Perhaps ‘accountability’
has fewer meanings, and so there is reason to favor this term over ‘respon-
sibility’. Third, it is also very important that ‘responsible’ and ‘accountable’
not be read as ‘morally responsible’ and ‘morally accountable’, despite the
fact that the former locutions are often used as shorthand for the latter. I
think that we have a more basic notion of responsibility (or accountability)
that underlies various kinds of responsibility: responsibility for one’s pro-
ductivity and creativity, moral responsibility, and so on. Responsibility in
this basic sense is a necessary, but not sufficient, condition for moral re-
sponsibility. The reason is that moral responsibility requires certain sorts of
knowledge, both general and particular. But this does not mean that the
concept of freedom I have characterized is a weak notion. It captures what
is traditionally known as the ‘freedom’ condition for moral responsibility,
while leaving open the question of how to characterize the ‘knowledge’
condition. And it provides the condition for the normativity required by
moral responsibility, while leaving open the possibility that one who is free
is not morally obligated.
I believe that this concept remains neutral as among various particular
conceptions of freedom. To see this, consider that an agent causationist
might argue that if one’s actions are to be up to one in this way, they must
be agent caused.19 Alternatively, one might argue that if one’s actions are to
be up to one in this way, they must be determined by one’s valuational
judgments.20 The first of these views requires the truth of indeterminism,
while the second does not.
Some might argue that the true concept of freedom is not in fact neutral,
and that the concept of freedom is just the notion that ‘one can choose
from among alternative courses of action’ or that ‘one could have done
otherwise’ or that ‘one is the undetermined source of one’s actions’. Yet,
there have been numerous attempts to provide such conditions that do not
allude to the choosing among alternatives, having the ability to do other-
wise, or being undetermined. I do not mean to suggest that any such at-
tempts have met with complete success, only that they have been genuine
attempts. If the concept of freedom were the notion that one can do oth-
erwise, then it would seem that agreement as to the entailment between
freedom and the ability to do otherwise would be readily forthcoming. The
fact that many who have attempted to provide necessary and sufficient
conditions for freedom have not granted the point suggests that the
concept of freedom is not simply the notion that one can do otherwise.
Further, even if we were to accept the point of the objection for the sake of
argument, we can still accept that the notion I have described is an impor-
tant condition for moral and other sorts of responsibility.21 Thus, if it can
be shown that we must represent ourselves as free in the sense that I have
described, that would be a significant result. Further, if Kant and Reid
are correct, and (R) is the first step in a sound argument that we are free
(in the sense specified in (R)), then that, too, would be a significant result.
Finally, the concept I have described satisfies widely accepted constraints
on the notion of freedom that is at the heart of the free will debate. For
example, philosophers often claim that the notion of freedom in which
they are interested is a notion that applies to persons, but not to many
nonhuman animals who are thought not to be persons.22 That is, there is a
sense in which the actions of humans can be up to them, or be their own,
in a way that the actions of a spider cannot. Since the notion of account-
ability might be thought to be limited in its applicability in such a way as
to exclude many nonhuman animals, the concept that I have described
matches this widely accepted feature of freedom. The readings I will con-
sider in what follows take ‘free’ in the sense specified in (R) to pick out the
concept of freedom. Hence, they are ‘conception-neutral’.23
The Sense of ‘Sense’

There are a number of different ways one might understand the ‘sense’ in
‘sense of freedom’, including ‘feeling’, ‘appearance’, ‘belief’, and ‘rational
commitment’. I begin the next subsection by arguing that the sense of
freedom in (R) is an epistemic commitment, or belief, that one is free. It
116 D. K. Nelkin
does not follow from this reading that all rational deliberators can articulate
their commitment without considerable probing and reflection; but it does
follow that they actually represent themselves as free. In this respect, it is
like the indeterministic reading of (R). In what follows, I call the conception-
neutral reading that understands ‘sense’ in this way, the ‘belief-concept’
reading. In the final subsection, I consider a weaker reading of (R) according
to which the sense of freedom is a rational commitment that one is free.
Roughly, one is rationally committed to a proposition when reflection and
recognition of the features of one’s own mental states and reasoning are suf-
ficient for one’s believing that proposition. But one need not actually be-
lieve that one is free in order to be rationally committed to the proposition.
I call the reading of (R) that incorporates both the conception-neutral reading
of ‘free’ and the rational commitment reading of ‘sense’, the ‘commitment-
concept’ reading. Although I believe that there is good reason to accept the
belief-concept reading, it is true that even weaker premises are needed to
support the commitment-concept reading, and thus, that the latter requires
less in the way of defense. At the same time, as I will explain, both readings
can do powerful explanatory and justificatory work.
An Argument for the Belief-concept Reading

We can begin by focusing once again on the nature of rational deliberation.
Recall that one rationally deliberates when one considers and evaluates
reasons with a view to deciding to act. The culmination of rational deliber-
ation is the adoption, on the basis of one’s evaluation, of certain reasons as
one’s reasons for performing an action, and the consequent decision so to
act. In other words, the deliberator decides to act on the basis of those
reasons.
Given the nature of rational deliberation just described, deliberators are
the sorts of beings who have a guiding conception of their purpose when
they perform intentional, goal-directed actions. Further, it is important
to note that they have a sense of their purpose whether or not they explic-
itly deliberate about what to do. For example, if a rational deliberator is
engaged in seeking the latest international news, then she has a conception
of the goal that defines her activity. If she does not conceive herself to be
seeking the latest news, then whatever activity in which she is engaged, she
is not seeking the news. This point applies to rational deliberation itself.
Here, too, one need not be able to articulate one’s purpose in deliberating
without considerable reflection, but one must have a conception of one’s

purpose when one deliberates, namely, that of finding, adopting, and
acting on good reasons.
Accepting this point depends crucially on an understanding of the nature
of rational deliberation as an activity directed by the agent toward a partic-
ular goal. The fact that a set of mental states constitutes a genuine piece of
rational deliberation is determined not merely by the content of those
states (e.g., ‘R is a reason to do action A’), and not merely by the fact that
they tend to issue in action. It is also essential that the deliberator be thereby
attempting to accomplish something, namely, the adoption of good rea-
sons as her own. This requires that the deliberator have a guiding sense that
her activity is aimed at this goal.24
If rational deliberators have a conception of themselves as seeking to
adopt and act on good reasons, then they must take themselves to be capa-
ble, in general, of finding such reasons. This claim is independently plausi-
ble, but I believe it derives additional support from its being explained.
Implicit in the sense of purpose that guides their intentional, goal-directed
actions is a commitment to a plan (however incomplete) to achieve it. Yet
if one does not view one’s purpose as attainable, then one cannot envision
a plan for its implementation. In other words, in order to have a guiding
conception about how one can achieve one’s purpose, one must believe that
one can.25 It is not necessary for deliberators to believe that they must suc-
ceed on each occasion of deliberation; but they must believe that they have
the ability to succeed in at least some typical situations. In other words,
they must believe that they are the sort of being who, by engaging in the
activity of rational deliberation, can succeed in finding and adopting good
reasons for acting. Otherwise, it would not be possible to conceive of their
activity as constituting a way of achieving their purpose. Thus, rational
deliberators must take themselves to be the sorts of beings who can and
do sometimes find, adopt, and act on good reasons.
If this is the case, then rational deliberators must believe that there are
good reasons to be found, adopted, and acted upon. In other words, they
must believe that there are reasons they ought to act upon. But if one is the
sort of being to whom such ‘ought statements’ apply, then one’s actions
can be rationally justified or unjustified. And that one’s actions are poten-
tial objects of this sort of justification is the way in which one can be
accountable for one’s actions.
118 D. K. Nelkin
At this point, the question arises whether rational deliberators must

believe that they are accountable simply because they believe that there are
reasons upon which they ought to act. Of course, it is not true in general
that an agent must believe the consequent of a true conditional when she
believes the antecedent. But this is a special case. There is a tight concep-
tual connection between the idea of there being reasons upon which one
ought to act and the idea of one’s being accountable; it seems that in order
to see oneself as the sort of being to whom such reasons apply, one must see
oneself as the sort of being who is accountable for his or her actions.26
Finally, then, we can see how one must believe that one’s actions are up
to one in such a way that one is accountable for them: one believes that
one’s actions can be performed as a result of one’s own adopted reasons
and, further, that they are potential objects of rational justification.27 Thus,
by identifying essential features of rational deliberation, we can see why
deliberators must have a sense of themselves as free (in the conception-
neutral sense).
Objections, Replies, and Elaboration

At this point, a number of questions might be raised about this line of rea-
soning. One natural question is whether rational deliberators can take their
activity to have a purpose different from the one I identify. In particular,
can rational deliberators who come to believe that they are not free con-
tinue to deliberate while taking themselves to have a different purpose in
doing so? For example, they might take the purpose of their deliberation to
be to achieve the maximal satisfaction of their desires, or to increase the
beauty in the world by being a passive receptor of reasons.
Although this objection has some force, I believe that its force can be de-
fused by focusing once again on the nature of deliberation. It is compatible
with the reasoning above that rational deliberators take their activity to
have more than one point or purpose. The claim that those engaged in
rational deliberation take their activity to have as its purpose their acting for
good reasons is perfectly consistent with the claim that they also take that
activity to have an additional, or further, point. For example, a rational
deliberator might be asked to deliberate and decide on a course of action as
part of her participation in a researcher’s psychological experiment. In
deliberating, she might have as her purpose to please the researcher, but if
she follows the researcher’s instructions to the letter, she must also take as
her purpose to act for good reasons. In order to do so, she must take it that
her deliberation can be efficacious and that there are reasons on the basis
of which she should act. Hence, as I argued earlier, she must have a sense
that she is free. The fact that one can imagine a rational deliberator taking
her activity to have any of a number of different points does not show that
rational deliberators can avoid taking their activity to have the point of
acting on the basis of good reasons.28
Further, in addition to consulting our intuitions about the nature of
rational deliberation in the abstract (as I have so far been urging us to do),
it is useful to consider in more concrete terms the question of whether
rational deliberators must take deliberation to have a particular purpose. To
this end, consider our own behavior when we believe that there are no
good reasons for acting in a certain way. For example, suppose you are at an
ice cream shop, having decided to buy an ice cream cone. You believe that
all the flavors are equally good. You don’t generally deliberate in such cir-
cumstances, but simply pick a flavor at random. On my view, the reason for
your lack of deliberation in such a case is that you see no point in deliber-
ating. We see no point in engaging in the evaluation of reasons for acting
when we think that there is no reason that we should act in a particular
way. We are not in a position in which we could act for good reasons in
choosing a flavor of ice cream, hence we see no point to deliberating, and
so do not engage in it.
This thought experiment—together with the investigation into the
nature of goal-directed intentional activity I sketched earlier—strongly sug-
gests that the very activity of rational deliberation manifests the sense that
there is a point or purpose to the activity. Further, these considerations
suggest that rational deliberation manifests the sense that the point of
one’s activity is to decide and ultimately to act on the basis of good reasons.
At this point, a second objection might naturally arise: even if one has a
conception of the purpose of one’s activity, one need not believe that one’s
purpose can be achieved in order to be rational in engaging in it. In what
follows, I defend the idea that one must believe that one has the general
ability to succeed in order to have a conception of one’s purpose in
performing goal-directed intentional actions. There is a long history of
controversy on this and related questions, and I cannot do full justice to it
here.29 But I will suggest one approach to it in the context of defending a
key premise in the reasoning for the belief-concept reading. Let me begin
120 D. K. Nelkin
by emphasizing that I am not relying on the claim that one must believe
one can succeed on every occasion. Rather the claim is that one must be-
lieve one can succeed in typical circumstances when one attempts the rele-
vant type of action. This point might be sufficient to defuse the present
worry about the belief-concept reading. For as I hope to show, those on
both sides of the debate over the connection between intentional action
and belief might be able to find agreement once this point is noted.
In thinking about the relation between intentional action and belief,
intuitions about examples play an important role. Here is an example that
provides support for the idea that one must believe one can achieve one’s
purpose, when one has a conception of it in acting intentionally: I cannot
intentionally engage in any activity that might be described as either flying
or trying to fly, no matter how much I desire to and how much I flap my
arms. The natural explanation seems to be that I do not believe it possible
for me to fly. In fact, I believe that I lack the general ability to fly. I can
imagine circumstances in which someone else tries to fly. But these
circumstances include delusions on the part of the person making the
attempt. If someone were to believe that his physical abilities were very
different from what they are, or that the laws of aerodynamics or gravity
were different from what they are, then he could try to fly. (Of course, this
would require a great number of changes in one’s belief system and proba-
bly a great deal of irrationality.) Here, it seems that what explains the
difference between this person and me is the difference between our beliefs
about what flapping our arms could possibly accomplish.
Those who argue against a necessary connection between belief and
intentional action offer examples on the other side. Kirk Ludwig offers this
one, among others30: A friend insists that I can make a basket from half-
court. I believe it impossible for me to make a basket from that distance,
and set out to show him that even if I try as hard as I can, I still can’t do it.
I try, and, amazingly, make the basket. This appears to be a case in which I
intentionally make a basket without believing that it is possible. According
to Ludwig, my intentionally making the basket in this case shows that I
also intended and tried to make it before surprising myself with my success.
Since I have argued that rational deliberators engaged in intentional goal-
directed activity have a sense of their purpose, the case seems to be one in
which the sense of purpose is not accompanied by a belief that it can be
achieved.
Reasons might be offered to resist this example and others like it, but
what is more important for our purposes here is the kind of example it is
(and is not). For even those who argue that intentional action does not re-
quire the belief that success is possible agree that there are some things one
cannot intentionally do (or even try to do), like flying or making a basket
from ten miles away. Some explanation for this fact is required.
Ludwig, for example, acknowledges that you could not try to hit a home
run by holding the bat in a “bunt” position.31 Why not? The reason, ac-
cording to Ludwig, is that this stance and the limited motion it permits are
not designed to bring about the end of hitting home runs. In cases like this
in which one intentionally performs an action without believing that one
can succeed, one might recognize that circumstances are special in such a
way that one’s activity cannot succeed. This is a cashing out of the idea that
in order to act intentionally, one must conceive one’s actions as of a type
designed to bring about a certain end.
If this is right, then we have a way to distinguish between the cases of mak-
ing a basket from half-court and swinging “all out” for a home run on the
one hand, and making a basket from Mars and a home run from a “bunt”
position on the other. The view also suggests that we assume that when we
act intentionally our actions are of a type that, under at least some typical
circumstances, can succeed. Further, there is a rationale for this: if one views
a certain kind of activity as one some of whose typical instances are success-
ful in achieving their purposes, then one can have a conception of how they
are successful. This makes it possible for us to envision our own activity of
this type as a way of implementing a plan to achieve its purpose.32
Admittedly, it is difficult to give criteria for ‘type’ of action here. But I
think it is possible to rely on an intuitive idea. I can’t try to fly, or make a
home run from the bunt position, because these are not the kinds of
actions at which I could succeed under anything like normal circum-
stances. I must believe that I have the general ability to succeed, if I have a
sense of my purpose in acting.33
This conclusion is all that is needed in order to defend the belief-concept
reading of (R) from the present objection. For even if one need not believe
that one can succeed on a particular occasion of deliberation, one must be-
lieve that one has the general ability to succeed. One must take it that one’s
activity is successful under at least some typical circumstances. Otherwise,
one won’t be able to view one’s engagement in the activity as counting as
122 D. K. Nelkin
a way of achieving the purpose of adopting good reasons for acting. Thus,
one must believe that one is free in the conception-neutral sense.
As I have argued, this reasoning should persuade those on a wide spec-
trum of views positing particular connections between intentional action
and belief. However, others might worry that, according to this reasoning,
rational deliberation requires too much in the way of conceptual develop-
ment and self-reflection. Young children and nonhuman animals would
appear to be counterexamples to the reasoning behind the belief-concept
reading of (R) since they include rational deliberators who lack mental-
state and related concepts such as ‘action’, ‘reasons’, and ‘responsibility’,
and those who have not reflected on the purpose of their activity. This sort
of worry deserves to be taken seriously.
However, I believe that the appeal of the alleged counterexamples is
undermined by reflection on the nature of the central concepts in the rea-
soning behind the belief-concept reading.34 First, let me emphasize that
rational deliberation as I understand it is itself a very sophisticated cogni-
tive activity: the consideration and evaluation of reasons with a view to
deciding to act, where one’s decision is based on one’s evaluation and
adoption of reasons as one’s reasons for performing an action. Once this is
understood, it becomes difficult to maintain that young children and
nonhuman animals provide counterexamples to the claim that all rational
deliberators must have the concept of reasons, for example. For the cogni-
tive sophistication required to engage in rational deliberation itself would
seem to rule out at least some members of these groups, and, in particular,
the very same members who are excluded from the possession of sophisti-
cated concepts such as reasons.
At the same time, it is important to emphasize once again that the concepts
in question do not include the concepts of moral reasons and moral respon-
sibility, but the concepts of reasons and responsibility in a basic sense. Thus,
the argument for the belief-concept reading of (R) does not presuppose the
possession of any moral concepts by rational deliberators. Recognition of the
robust understanding of rational deliberation articulated above together with
the relatively basic nature of the notions of reasons and responsibility should
dispel the worry raised by the alleged counterexamples that there must be a
faulty step somewhere in the reasoning behind the belief-concept reading.
Finally, even if lingering doubts remain about the strength of the belief-
concept reading of (R), it is possible to adopt a weaker conception-neutral
reading without giving up much of the explanatory and justificatory role

of (R). According to the commitment-concept reading, rational deliberators
are rationally committed, in virtue of being rational deliberators, to their
being free. They need not actually believe that they are free, but they are in
a state such that mere reflection and recognition of features of their own
mental states and activity would suffice for the beliefs in question. To reach
this weaker conclusion, each premise of the reasoning for the belief-concept
reading of (R) might be weakened in such a way as to incorporate rational
commitment in place of belief, and thus, to require less in the way of self-
reflection than the reasoning for the belief-concept reading. Equally impor-
tantly, the commitment-concept reading of (R) can do much of the work
that the belief-concept reading can do. For it can explain the centrality
of the belief that one is free for those (like us) who have reflected on the
matter. And it can also play a powerful role in arguments like Reid’s and
Kant’s that we are in fact free. To take the simplest example, Reid’s burden
of proof argument set out earlier does not seem to lose much of its force
when we substitute ‘rational commitment’ for ‘conviction’. Thus, even if
we adopt the commitment-concept reading of (R) over the belief-concept
reading, (R) remains an important thesis.
Before concluding, it remains to consider one final objection. Unlike the
others, this one does not question the reasoning for the belief-concept
reading of (R), but instead questions the meaning and significance of
the conclusion. The objection is that once we see that freedom in the
conception-neutral sense is closely tied to the ability to act for good reasons,
it is not clear that the notion of freedom in the conception-neutral readings
of (R) is really the concept of freedom after all. The quality of ‘being chosen’
seems to fade into the background while the notion of rational capacity can
be seen to undergird the conception-neutral notion of freedom described.35
In response, it is important to note that although the concept of freedom
can be seen to be closely associated with reasons and accountability, this fact
does not detract from its conception-neutral character. For example, noth-
ing that I have said rules out the possibility that satisfaction of the concept
of freedom as I have characterized it requires the falsity of determinism.
Further, the concept, ‘freedom’, is not simply to be equated with the con-
cept, ‘ability to respond to good reasons to act’. The concept of freedom, on
my view, is that one’s actions are up to one in a particular and special way,
namely, such that one is accountable for one’s actions. It is true that at least
124 D. K. Nelkin
one way of one’s actions being up to one such that one is accountable for
them depends partly on the agent’s ability to respond to good reasons for
acting. But I believe that this is a connection that we should find attractive,
particularly once the connection between obligation by reasons and ac-
countability is noted. For if one’s actions are up to one in such a way that one
is (in a nonmoral sense) accountable for them, then we would seem to have
just what is wanted in a ‘freedom’ condition for moral responsibility.
Finally, even if we set aside the question of whether the concept of free-
dom is the concept of one’s actions being up to one such that one is
accountable for them, the latter concept remains a significant one. For it is
one that provides an important condition for moral and other sorts of
responsibility, and one that we care deeply about.
In sum, the belief-concept reading of (R) faces important challenges. Yet
an intuitively plausible line of reasoning gives it considerable resources
with which to respond. As a result, the belief-concept reading (together
with the weaker commitment-concept reading) remains a promising
interpretation of the widely accepted idea that in virtue of being rational
deliberators we cannot escape the sense that we are free.
4 Conclusion
Both the indeterministic reading and the belief-concept reading of (R) rest
on reasoning that takes the sense of freedom to be a belief manifested by
the very activity of rational deliberation. They diverge in the content of the
beliefs each attributes to rational deliberators. Equally important is the
difference between the notions of rational deliberation employed by each
reading, for it is on this difference that the difference in beliefs ultimately
rests. Unlike the indeterministic reading, the belief-concept reading takes
the essence of rational deliberation to be the attempt to find and adopt
good reasons for acting. By building on the reason-seeking features of
rational deliberation, it is possible to explain why rational deliberators
must have a sense of their actions being up to them in such a way that they
are accountable for them. At the same time, this focus on the reason-seeking
features of rational deliberation makes the belief-concept reading of (R)
particularly well suited to an antiskeptical argument in the spirit of Kant
and Reid. For if rational deliberation is essentially a reason-seeking faculty,
it is tempting to conclude that the simple possession of such a rational
faculty could not be responsible for a false belief. Of course, turning this
thought from a tempting idea into the conclusion of a sound argument
that rational deliberators are free is a project of its own. Yet if the belief-
concept reading of (R) is correct, then we are at least entitled to a key
premise in such an argument.
Acknowledgments
This paper is a longer version of a paper presented at the Inland Northwest

Philosophy Conference 2001. I am very grateful to the organizers of the
conference and the editors of this volume, Joseph Keim Campbell, Michael
O’Rourke, and David Shier, for the opportunity to be part of such a stimu-
lating and constructive exchange of ideas. I would also like to thank my
commentator, Ron Wilburn, and the audience for their helpful comments
and for an excellent discussion. I presented an earlier version of the paper
at the University of Virginia in February 2001, and I thank the audience for
valuable discussion on that occasion. A number of people read versions of
this paper and offered comments that caused me to improve the paper. For
this, I am very grateful to Tal Brewer, Tyler Burge, Jim Cargile, Peter Hanow-
ell, Barbara Herman, Andrew Hsu, Tina Huggins, Pat Matthews, Peter
Murphy, Michael Otsuka, Derk Pereboom, Sam Rickless, George Thomas,
and two anonymous referees for this volume.
Notes
1. For example, Galen Strawson writes that a free will skeptic who concentrates on
abandoning his “ordinary conception of freedom” may temporarily experience “a
total paralysis of all purposive thought as it is ordinarily conceived and experienced”
(Strawson 1986, 102). It is when we are trying to make decisions (moral and
otherwise) that our belief in our own freedom is most acutely felt. (See p. viii, and
chapter 3.) At the same time, Strawson argues that freedom is impossible and that
our belief is false. It must be noted that Strawson does not accept (R), but rather the
weaker claim that humans who rationally deliberate necessarily believe themselves
free.
Castañeda accepts something like (R), and at the same time accepts the possibility
that the skeptic is right: if so, “the universe is ugly; given the biological and psycho-
logical primacy of practical over contemplative thinking, we are, thus, condemned
to presuppose a falsehood in order to do what we must think practically” (Castañeda
1975, 134).
126 D. K. Nelkin
2. Controversy abounds over just how Kant should be interpreted here. I will not
enter into that controversy here.
3. I explore different ways this argument can be developed in Nelkin (in preparation).
4. See also Aristotle (1984c, 1798 [NE 1139a, 13–14]) for a similar statement.
5. See, for example, Aristotle (1984b, 1941 [EE 1225b, 34–36]) : “. . . nor does he even
choose what is possible, generally, if he does not think it in his power to do or abstain
from doing it.” See also, Aristotle (1984b, 1942 [EE 1226a, 25–26]) : “. . . about these
[things the production of which is not in our power] none would attempt to deliberate
except in ignorance.” These qualifications have suggested to some that what Aristotle
meant in expressing the apparently stronger claim that deliberation requires indeter-
minism was actually the weaker claim that deliberation requires a belief in indeter-
minism. There is some dispute about this, however. See Gail Fine (1981, 572 and
note 10) for a statement of this view, as against Sorabji (1980, 228 and 245–246), who
credits Aristotle with the stronger claim and argues only that Aristotle should have
offered the weaker. See my note 16 for further textual evidence in support of the view
that Aristotle held the weaker thesis.
6. See, for example, Castañeda (1975) who accepts the indeterministic reading. Van
Inwagen accepts something like the indeterministic reading of (R). (See van Inwagen
1983). He argues that deliberators must believe that multiple alternative possibilities
are each within one’s power. However, van Inwagen stops short of claiming that de-
liberators must believe that these alternative possibilities are undetermined. Those
who have understood Kant as advocating the indeterministic reading of (R) include
Castañeda (1975, 134) and Thomas Hill (1985, 16–17).
7. See Richard Taylor (1964) and van Inwagen (1983, especially 152–161), for simi-
lar lines of argument. Taylor argues that a deliberator must not believe that she
cannot choose among undetermined alternatives rather than that such a deliberator
must believe that she can so choose. Van Inwagen adapts Taylor’s argument in order
to argue for the presence of a belief. As mentioned in my note 6, van Inwagen argues
that all deliberators necessarily believe that multiple alternative possibilities are
within their power.
8. This is not the only line of reasoning that might be used in support of the inde-
terministic reading of (R). One might begin with the claim that one can deliberate
only if one believes oneself to be the ultimate source of one’s actions, and infer that
one must believe determinism to be false in order to do so. Although I believe the
line of reasoning set out in the text to be the one most often deployed, much of what
I go on to say addresses this second line, as well.
9. My claim here is analogous in certain respects to Harry Frankfurt’s claim to have

found a counterexample to the ‘Principle of Alternate Possibilities’ (see Frankfurt
1969). That principle states that one cannot be responsible for an action if one cannot
do otherwise. I claim here to have a counterexample to what might be called the

‘principle of belief in alternate possibilities’, the claim that one cannot deliberate if
one believes that one cannot do otherwise.
10. See, for example, Davidson (1969, 179).
11. See Randolph Clarke (1992), for a similar line of reasoning against the indeter-
ministic reading.
12. It might also be objected that while the original voting case shows that (i) and
even (iii) are false, it is nevertheless a case in which the agent believes that there is
something she can do or forgo (e.g., voting for Gore in a certain way), even if this
is not the (primary) object of deliberation. Thus, the case leaves open the possibility
of finding reasons other than (i) through (iii) for accepting (R), including the rea-
soning described in note 7. The case to follow in the text can be used to address this
objection, as well.
13. An added virtue of this case is that it is immune to the sort of objection we saw
earlier concerning action individuation. For if the agent believes that her future
actions are determined by past physical states together with natural laws, then she
believes that conditions are sufficient for her performing a particular action in a
particular way (i.e., with certain causal antecedents).
14. A similar sort of reply is suggested by van Inwagen. However, he offers the reply
on behalf of free will skeptics who deliberate rather than on behalf of determinists
who deliberate. See van Inwagen (1983, 157–158).
15. This notion of rationalization is somewhat broader than that introduced by

Davidson (1963). In that paper, Davidson uses ‘rationalization’ to refer to a reason
consisting of a desire (or other pro-attitude) toward actions of a certain type together
with the specific belief that one’s action is of that type. My formulation allows for
additional ‘background’ beliefs that are necessary for seeing the behavior as
intelligible to count as rationalizing. The example which follows in the text is one in
which the belief rationalizes the action in my sense even though it is not a belief of
the specific form: ‘my action is of a certain type’.
16. This passage also provides further support for the claim that Aristotle did not
hold the view that deliberation requires indeterminism. For if he had held that view,
then it would have been natural for him to identify the claim that we could not
deliberate as an absurd consequence of determinism. But in fact, he identifies the
weaker claim that deliberation would have no point as the absurd consequence of
determinism.
17. There are ways to resist the indeterministic reading other than the one I have set
out here. Some have described the phenomenology of deliberation and decision in
a way that undermines the indeterministic reading (see, for example, Mele 1995,
133–136 and Strawson 1986, 115, note 30), and others have argued against the
128 D. K. Nelkin
indeterministic reading by providing diagnoses of why we mistakenly take the view

to be true. One such diagnosis is that the indeterministic reading of (R) is confused
with the true claim that we human beings have a sense that our actions are undeter-
mined. J. David Velleman (1989), for example, accepts this weaker thesis, and offers
a psychological account of the phenomenon. A second kind of diagnosis is that
the belief in the contingency of one’s future actions has been confused with a be-
lief in the epistemic or, alternatively, the doxastic contingency of those actions. (See
Dennett 1984, 112–113; Pereboom 1995, 32–33; Kapitan 1986; and Pereboom 2001,
136–137.) A related diagnosis has been offered by Philip Pettit (1989), who argues
that the indeterministic reading has been confused with the following true claim: for
each of a set of alternative courses of action under deliberation, a deliberator must
not believe that it is not possible. Hilary Bok (1998) argues that while we have reason
to regard ourselves as having genuine alternatives, this means that we must see our-
selves as having alternatives in the sense that there are multiple actions we would
perform if we chose (110–114). George Thomas (unpublished manuscript) defends
what I take to be the most promising of these accounts: rational deliberators must
believe that they can choose among alternate possibilities where the alternate possi-
bilities are those actions that are not made impossible by anything that does not pro-
ceed via the process of deliberation itself. (See also Dennett 1984, 115–122 for a
related suggestion.) I discuss all of these suggestions in Nelkin (in preparation).
18. See Honderich (1988) for the view that both compatibilists and incompatibilists
make the mistake of focusing on one family of important human attitudes to the
exclusion of another with which it is inconsistent. I cannot do justice to Honderich’s
rich discussion here. For our purposes here, it is worth noting that it is consistent
with there being a single concept of freedom, and even one to which all rational de-
liberators are in some way committed, that human beings often possess inconsistent
attitudes concerning particular conceptions of freedom.
19. See, for example, Roderick Chisholm (1964).
20. See Watson (1975).
21. In this connection, see Frank Jackson (1998), who writes: “I find compelling
Peter van Inwagen’s argument that . . . determinism is inconsistent with free will.
What compatibilist arguments show, or so it seems to me, is . . . that free action on
a conception near enough to the folk’s [i.e., common-sense] to be regarded as a
natural extension of it, and which does the theoretical job we folk give the concept
of free action in adjudicating questions of moral responsibility and punishment, and
in governing our attitudes to the actions of those around us, is compatible with
determinism” (44–45). Although the concept that I offer is neutral as between com-
patibilism and incompatibilism, the spirit of Jackson’s point applies to it, as well.
Even if one doubts that the concept I offer is the concept of freedom, one can still ac-
cept that it can do the theoretical work we want it to do, including supporting our
attributions of moral responsibility.
22. See, for example, Frankfurt (1971).
23. Bok has recently defended what might naturally be thought of as an explicitly
compatibilist reading of (R). In particular, she argues that being practical reasoners
gives us reason to regard ourselves as free in a compatibilist sense. We are free in the
relevant sense when we can determine our conduct through practical reasoning, and
we have genuine alternatives among which we can choose, where genuine alterna-
tives are those actions we would perform if we chose to perform them. See especially
Bok (1998, 118–119).
24. Strawson’s rejection of (R) may ultimately rest on an understanding of rational

deliberation that differs from mine on just this point. For example, he describes the
rational deliberation of an imaginary being, the Spectator, who is “experientially
detached from her desires—from her motivations generally—in some curious way”
(Strawson 1986, 234). For her, rational deliberation is a series of “practical-rational
calculations going on in” a person in such a way that the person need have no sense
that she is the decider and rational planner of action (Strawson 1986, 235). Although
Strawson sees this as an anomalous case of rational deliberation, it reveals a possible
divergence between his conception of rational deliberation and my own. For a less re-
cent, but very explicit, example of a conception of rational deliberation very differ-
ent from mine on this point, see Hobbes, who describes deliberation in the following
way: “When in the mind of man appetites and aversions, hopes and fears, concern-
ing one and the same thing arise alternately, and diverse good and evil consequences
of the doing or omitting the thing propounded come successively into our thoughts,
so that sometimes we have an appetite to it, sometimes an aversion from it . . . the
whole sum of desires, aversions, hopes and fears, continued till the thing be either
done or thought impossible, is that we call DELIBERATION” (Hobbes 1668/1994, 33).
25. Frederick Adams develops a similar idea in Adams (1995, 552). Adams there
argues that all intentional action requires an attempt, and that trying to perform an
action requires the lack of a belief that success is impossible. Thus, it would seem to
follow that intentional action requires the lack of a belief that success is impossible.
Sometimes Adams also seems to endorse the stronger claim that intentional action
requires the belief in the possibility of success (see 553–554, for example). And the
stronger claim fits well with his reasoning that intentional action requires that one
have beliefs about how to achieve one’s end.
26. As Burge has pointed out, one might have “incomplete mastery” of a concept,
have false beliefs about even some of the essential properties of its instances, and yet
have genuine beliefs employing the concept nevertheless. For example, one might
believe that one suffers from arthritis, even if one believes it is not a disease of the
joints. (See Burge 1979.) However, it may be that there are certain true beliefs that
one cannot lack, and still be said to have the concept. The case at hand appears to be
of this kind: it is constitutive of having the concept of having reasons that one
believe one ought to act in certain ways if one has reasons to act.
130 D. K. Nelkin
27. This idea, too, recalls Kant: “That our reason has causality, or that we represent
it to ourselves as having causality, is evident from the imperatives which in all mat-
ters of conduct we impose as rules upon our active powers. ‘Ought’ expresses a kind
of necessity and connection with grounds which is found nowhere else in the whole
of nature . . .” (Kant 1781/1965, 472–473 [A457/B575]).
28. Along lines similar to the objection in the text, it might be argued that in ratio-
nally deliberating, one is sometimes guided only by the purpose of finding ‘the best
thing to do’. Yet, as I argued earlier, if one’s activity toward the goal of finding ‘the
best thing to do’ is to constitute genuine rational deliberation (as opposed to other
activities that might aim at that goal, such as making a sacrifice to the gods), one must
also be guided by the purpose of finding good reasons for acting.
29. For a small sampling of the literature on the connections among intentional
action, intention, and belief, see Davidson (1985) who argues that intentional action
requires the belief that one can succeed, Grice (1971), Harman (1976), and Velleman
(1985) who argue that having an intention requires the stronger belief (or accep-
tance, in the case of Grice) that one will succeed, and Bratman (1986) who argues
that (normally) having an intention and being rational requires that one not believe
that one will not succeed. Many have argued against one or more of these claims. For
example, McCann (1986) argues against all of these claims. Ludwig (1992) and
(1995) also argues against all of these claims, and goes one step further. He defends
the claim that one can be rational in both intending and acting intentionally even
though one believes that one cannot succeed. (I am grateful to Kirk Ludwig for email
correspondence about his view.)
30. See Ludwig (1992, 263).
31. See Ludwig (1995, 566–568).
32. Albritton (1985) argues that one can try to do what one believes is impossible,
but, like Ludwig, also recognizes the need to account for cases like the flying case. For
example, he agrees that there are certain things he cannot try to do, including trying
to jump over a building and even trying to do fifty push-ups. His explanation for this
fact is that “in his present cognitive position and state of mind that description of
him [as trying] would be inept whatever he did,” and “It’s that nothing I can think
of to do this evening would be rightly described as trying to jump over this building,
in a straightforward sense, unless, for example, my beliefs were to alter or go very
dim” (245). It seems to me that something important must be contained in the
phrases ‘present cognitive position’ and ‘my beliefs’. It is tempting to take them to
include the lack of belief that success is possible or the lack of belief that one has the
general ability to perform these kinds of actions.
33. One consequence of this reasoning is that there is a certain kind of irrationality
in simultaneously being a rational deliberator and a ‘practical reasons nihilist’ or
even a skeptic about practical reasons. Against this, it might be argued that a skeptic
about reasons could rationally deliberate (and be perfectly rational) simply by seeking
reasons if they happen to exist. In reply, as argued earlier, one could not deliber-
ate if one lacked the belief that one can succeed in at least some typical situations;
otherwise, one would lack a conception of what one was doing in deliberating. Since
deliberation requires such a conception, one must believe that one can sometimes
succeed in finding reasons for acting. Significantly, this leaves open the possibility of
a rational skepticism about the possibility of finding reasons on a particular occa-
sion. See also Burge (1998) for a different argument that finds a similar albeit more
general target in a broad reasons skepticism. Burge argues persuasively that for one
to fully understand reasons, “one must be susceptible to reasons” and “one must
recognize” the effect of reasons on one’s judgments and inferences (250).
34. The initial appeal of the examples also fades upon examination of the growing
body of work on infant and child development. Developmental psychologists are
divided on the question of exactly when many of the relevant concepts, such as ‘self’,
‘goal’, ‘desires’, and ‘desires as reasons for acting’ emerge in the human infant and
child. And while psychologists see the field as one that is need of a great deal more
research (see, for example, Wellman and Iagaki 1997, 2 and Meltzoff et al. 1999, 19),
there is an increasing consensus that such concepts appear much earlier in human
development than was previously thought. (See, for example, Wellman and Iagaki
1997 and the essays in part 2 of Malle, Moses, and Baldwin 2001. For example, it is
argued that by the age of 18 months, many children not only have the mental state
concepts of ‘desire’ and ‘intention’ but also understand that others have desires dif-
ferent from their own, and that others can intend to perform actions even though
they are prevented from performing them (Meltzoff et al. 1999). Thus, there is good
evidence that very young children have rich mental state concepts. Of course, it is
true that even if this evidence were conclusive, it does not provide a positive argu-
ment for the belief-concept reading of (R). At the same time, the available evidence
detracts considerably from the initial appeal of the alleged counterexamples.
35. The objection might be pressed in the following way: consider theoretical
deliberation. When engaged in it, we can suppose that we have a sense that we can
believe or judge for good reasons. But there is nothing like a sense of freedom
associated with theoretical deliberation; to the contrary, we do not choose our judg-
ments as we do many of our actions. So perhaps the sense of freedom as I have char-
acterized it is not really a sense of freedom either. This way of pressing the objection
raises a number of interesting issues regarding the relationship between rational
deliberation and its theoretical parallel. For the objection makes a number of presup-
positions including these: (1) that rational deliberation and its parallel in the realm
of judgment differ in that freedom and ‘chosen-ness’ are associated with the former
but not the latter, and (2) that the two forms of deliberation do not differ in any rel-
evant way in the implications of their respective requirements that agents represent
themselves as capable of responding well to good reasons. Both presuppositions
might be challenged, and I discuss both options in Nelkin (in preparation).
132 D. K. Nelkin
References
Adams, F. 1995. “Trying: You’ve Got to Believe.” Journal of Philosophical Research 20:
563–570.
Albritton, R. 1985. “Freedom of the Will and Freedom of Action.” Presidential

Address. Proceedings of the American Philosophical Association 59: 239–251.
Anscombe, G. E. M. 1957. Intention. Oxford: Blackwell.
Aristotle. 1984a. The Complete Works of Aristotle. J. Barnes, ed. Princeton, N.J.:
Princeton University Press.
. 1984b. Eudemian Ethics. J. Solomon, trans. In Aristotle (1984a).
. 1984c. Nicomachean Ethics. W. D. Ross, trans., revised by J. O. Urmson. In

Aristotle (1984a).
. 1984d. De Interpretatione. J. L. Ackrill, trans. In Aristotle (1984a).
Bok, H. 1998. Freedom and Responsibility. Princeton, N.J.: Princeton University Press.
Bratman, M. 1986. Intention, Plans, and Practical Reason. Cambridge, Mass.: Harvard
University Press.
Burge, T. 1979. “Individualism and the Mental.” Midwest Studies in Philosophy 4:

73–121.
. 1998. “Reason and the First Person.” In C. Wright, B. Smith, and C.

MacDonald, eds., Knowing Our Own Minds. Oxford: Clarendon Press.
Castañeda, H. 1975. Thinking and Doing. Dordrecht: Reidel.
Chisholm, R. 1964. “Human Freedom and the Self.” The Lindley Lecture. Lawrence,
Kans.: University of Kansas Department of Philosophy. Reprinted in Watson
(1982).
Clarke, R. 1992. “Deliberation and Beliefs About One’s Abilities.” Pacific Philosophical
Quarterly 73: 101–113.
Davidson, D. 1963. “Actions, Reasons, and Causes.” Reprinted in Davidson (1980).
. 1969. “The Individuation of Events.” In N. Rescher, ed., Essays in Honor of

Carl G. Hempel. Dordrecht: Reidel. Reprinted in Davidson (1980).
. 1980. Essays on Actions and Events. New York: Oxford University Press.
. 1985. “Reply to Pears.” In B. Vermazen and M. Hintikka, eds., Essays on

Davidson Actions and Events. Oxford: Clarendon Press.
Dennett, D. 1984. Elbow Room. Cambridge, Mass.: MIT Press.

Fine, G. 1981. “Aristotle on Determinism: a Review of Richard Sorabji’s ‘Necessity,

Cause, and Blame’.” Philosophical Review 90: 561–580.
Frankfurt, H. 1969. “Alternate Possibilities and Moral Responsibility.” Reprinted in

Frankfurt (1988).
. 1971. “Free Will and the Concept of a Person.” Reprinted in Frankfurt (1988).
. 1988. The Importance of What We Care About. Cambridge: Cambridge

University Press.
Grice, P. 1971. “Intention and Uncertainty.” Proceedings of the British Academy.

London: Oxford University Press.
Harman, G. 1976. “Practical Reasoning.” Review of Metaphysics 29: 431–463.
Hill, T. 1985. “Kant’s Argument for the Rationality of Moral Conduct.” Pacific
Philosophical Quarterly 66: 3–23.
Hobbes, T. 1668/1994. Leviathan. E. Curley, ed. Indianapolis, Ind.: Hackett.
Honderich, T. 1988. The Consequences of Determinism 2. Oxford: Clarendon Press.
Jackson, F. 1998. From Metaphysics to Ethics. Oxford: Oxford University Press.
Kant, I. 1781/1965. The Critique of Pure Reason. N. Kemp Smith, trans. New York:
St. Martin’s Press.
. 1785/1981. Grounding for the Metaphysics of Morals. J. Ellington, trans.

Indianapolis, Ind.: Hackett.
Kapitan, T. 1986. “Deliberation and the Presumption of Open Alternatives.”

Ludwig, K. 1992. “Impossible Doings.” Philosophical Studies 65: 257–281.
. 1995. “Trying the Impossible: A Reply to Adams.” Journal of Philosophical

Research 20: 563–570.
Malle, B. and J. Knobe. 2001. “The Distinction Between Desire and Intention: A Folk-
Conceptual Analysis.” In B. Malle, L. Moses, and D. Baldwin, eds., Intentions and
Intentionality. Cambridge, Mass.: MIT Press.
Malle, B., L. Moses, and D. Baldwin, eds. 2001. Intentions and Intentionality.
Cambridge, Mass.: MIT Press.
McCann, H. 1986. “Rationality and the Range of Intentions.” Midwest Studies in

Mele, A. 1995. Autonomous Agents: From Self Control to Autonomy. New York: Oxford
University Press.
134 D. K. Nelkin
. 2001. “Acting Intentionally: Probing Folk Notions.” In B. Malle, L. Moses,

and D. Baldwin, eds., Intentions and Intentionality. Cambridge, Mass.: MIT Press.
Meltzoff, A., A. Gopnik, and B. Repacholi. 1999. “Toddlers’ Understanding of Inten-

tions, Desires and Emotions: Explorations of the Dark Ages.” In P. Zelazo, J. Astington,
and D. Olson, eds., Developing Theories of Intention. London: Lawrence Erlbaum.
Nelkin, D. In preparation. Rational Deliberation and the Sense of Freedom.
Pereboom, D. 1995. “Determinism al Dente.” Nous 29: 21–45.
. 2000. Living without Free Will. Cambridge: Cambridge University Press.
Pettit, P. 1989. “Determinism with Deliberation.” Analysis 49: 42–44.
Rawls, J. 1971. A Theory of Justice. Cambridge, Mass.: Belknap Press of Harvard

University Press.
Reid, T. 1788/1983. Essays on the Active Powers of Man. In R. E. Beanblossom and

K. Lehrer, eds., Inquiry and Essays. Indianapolis, Ind.: Hackett.
Sorabji, R. 1980. Necessity, Cause, and Blame. Ithaca, N.Y.: Cornell University Press.
Strawson, G. 1986. Freedom and Belief. Oxford: Oxford University Press.
Taylor, R. 1964. “Deliberation and Foreknowledge.” American Philosophical Quarterly

1: 73–80.
Thalberg, I. 1962. “Intending the Impossible.” Australasian Journal of Philosophy 40:

49–56.
Thomas, G. Unpublished manuscript. “Deliberation and Determinism.”
Van Inwagen, P. 1983. An Essay on Free Will. Oxford: Oxford University Press.
Velleman, J.D. 1989. “Epistemic Freedom.” Pacific Philosophical Quarterly 70: 73–97.
. 1985. “Practical Reflection.” Philosophical Review 94: 33–61.
Watson, G. 1975. “Free Agency.” Reprinted in G. Watson (1982).
Watson, G., ed. 1982. Free Will. Oxford: Oxford University Press.
Wellman, H., and K. Inagaki, 1997. The Emergence of Core Domains of Thought: Chil-
dren’s Reasoning About Physical, Psychological, and Biological Phenomena. San Francisco,
Calif.: Jossey-Bass.
Wellman, H., and A. Phillips. 2001. “Developing Intentional Understandings.” In B.

Malle, L. Moses, and D. Baldwin, eds., Intentions and Intentionality. Cambridge, Mass.:
MIT Press.
6 Libertarian Openness, Blameworthiness, and Time
Ishtiyaque Haji
1 Introduction
Construe ‘action’ broadly to include mental actions such as decisions, as

well as garden-variety actions like lighting a cigar, and intentional omis-
sions. For the purposes of this paper, assume that being morally responsible
for an action consists in being either blame- or praiseworthy for that action.
According to an intuitively plausible thesis, moral responsibility is essen-
tially ‘retrospective’, in that, necessarily, if a person is morally responsible
for an action, then the time at which the person is responsible is later than
(or the same as) the time at which the action is performed. It has been
thought, for example, that when we speak of a person’s desert in connec-
tion with blameworthiness, the ‘desert base’—the fact or facts to which we
appeal in order to explain this person’s desert of blame—necessarily pre-
cedes what is deserved. So proponents of this thesis are committed to its
implication that a person cannot now be deserving of blame for an action
yet to be performed by her.1 The thesis can be formulated more exactly in
this way:
Retrospective Responsibility (thesis RR): Necessarily, for any agent, S,
time, t, and action, A, if S is morally responsible at t (and thereafter)
for A, then for some t* identical to or earlier than t, S does A at t*.
If this thesis is true, then, of course, the thesis of Prospective Responsibility

(thesis PR):
Possibly (i.e., it can be the case that), for some agent, S, time, t and
action, A, if S is morally responsible at t (and thereafter) for A, then for
some t* later than t, S does A at t*,
is false.
136 I. Haji
Not only is thesis RR intuitively attractive, it is invoked as a vital plank in

debates over significant moral issues. For example, in arguing for the strik-
ing view that one’s past voluntary actions are the only bases of desert,
James Rachels (1991, 142–143) invokes the premise that people are morally
responsible only for their past voluntary actions. John Cottingham (1992,
662) proposes that we cannot rectify an injustice prior to its occurrence. He
thinks that this is so because “Verbs like ‘to rectify’ and ‘to correct’ share
with many other verbs (including ‘to punish’, ‘to blame’, ‘to thank’, ‘to
regret’, ‘to renounce’) what we might call an inherently ‘retrospective’
logic: we cannot understand such verbs without grasping that their use
involves an intrinsic and automatic reference back to some past event or
state of affairs.”
In addition to its innate interest, views such as those of Rachels and
Cottingham reveal that thesis RR is of substantive importance and merits
close scrutiny. Contrary to conventional wisdom, though, I believe that the-
sis PR is true, and that the view that responsibility is essentially retrospective
is thus to be eschewed. In the interests of brevity, in this paper I confine
attention to the version of thesis RR concerning blameworthiness (‘blame
past’) that says that, necessarily, if one is blameworthy for performing an
action, then the time at which one is blameworthy is later than (or the same
as) the time at which one performs the action. The analogous version of
thesis PR (‘blame future’) says that one can, as of a time, be blameworthy for
an action—a ‘future action’—that one will perform later than that time.
In what follows, I first advance what I take to be one of the strongest lines
of support for blame past. I then utilize Frankfurt-type examples to cast
doubt on this line of support. Finally, I propose that a certain conception of
moral responsibility—roughly, the conception that when one is morally re-
sponsible for an action, one discloses or expresses what one morally stands
for in relation to that action—provides motivation for blame future.
Before launching into a discussion of prospective blameworthiness, I
should lay to rest the concern that I’m addressing a will-o’-the-wisp. At-
tending to this concern will lead to further refinement of the thesis of
prospective responsibility of interest. The concern is that if we allow that
agents can be blameworthy for the consequences of their actions, as we
surely should do, and we allow that a future decision or an act can be a
certain consequence of one’s present action, why should we not be able to
infer that agents can be blameworthy for their future mental or other
Libertarian Openness, Blameworthiness, and Time 137
actions? Why should the prospective blameworthiness thesis require

special justification?
A number of considerations sustain the need for such justification. For
one thing, if it is uncertain that a particular consequence will occur from
one’s present action—there is a chance that it will occur and there is a
chance that it will not—then even should the consequence occur, we
might well wonder whether one can, at the time of performing the action,
be blameworthy then for its future occurrence. For another thing, it is im-
portant to bear in mind that blameworthiness for the consequences of our
actions is, in Michael Zimmerman’s (1988, 55) terminology, “essentially
empty.” Directly free actions are free independently of the agent’s having
freely done something that brought about these actions; indirectly free
actions derive their freedom from the freedom of other acts. Blameworthi-
ness for a consequence that issues from an action (including a mental
action such as a decision) with respect to which one is directly free, is
wholly exhausted by (‘direct’) blameworthiness for that action, in the sense
that blameworthiness for the consequence corresponds “with no entries in
a person’s ledger above and beyond those already entered as a matter of di-
rect culpability” for the pertinent action (1988, 54). Whereas it is relatively
unproblematic that a person can now be blameworthy for certain conse-
quences of, say, a decision for which he is directly blameworthy now, it is
an open question whether a person can now be (directly) blameworthy for
a decision that he will directly freely make an hour from now.
This slight digression highlights the fact that the thesis of prospective
responsibility to be argued for in this essay is really the thesis that we can
be blameworthy (or, more generally, responsible) for future actions of ours
that we can avoid performing. It can be recast in this fashion:
Thesis PR*: Possibly (i.e., it can be the case that), for some agent, S, time, t,
and action, A, if S is morally responsible at t for A, then for some t* later
than t, S does A at t* and S could have avoided doing A at t*.
2 Libertarian Openness and a Defense for Blame Past
Although blame past is widely affirmed, it is less than transparent what

argument could be given in its support. Reflection, though, on remarks
against ‘prepunishment’—which is punishment for an offense before the
offense is committed—are suggestive. Fred Feldman (1997), for instance,
138 I. Haji
offers a number of reasons against a legal system’s permitting prepunish-

ment. One has to do with ‘openness’ of the future that is, he suggests,
supposedly presupposed by responsibility. Pondering a case in which it
seems likely that a person will commit some crime, Feldman (1997, 189)
writes:
We think he will deserve the legally mandated punishment only if he will be re-
sponsible for the crime; and we think he will be responsible for the crime only if he
will commit it ‘freely’; and we think that if he will commit it ‘freely’, then it cannot
yet be quite certain that he will commit it. There must still be some possibility that
he will decide not to commit it. So we insist upon a legal system that prohibits
punishment-in-advance.
One might exploit the suggestion in this passage that legally mandated
punishment presupposes a sort of openness of the future to argue against
blame future and, hence, to generate support for blame past. Suppose that
although it is now true that, say, Bob will steal Ann’s car at noon, it is still
‘open’ to Bob, as of now, not to steal at noon. But then, it might be claimed,
Bob is not now deserving of blame for stealing later. For, as of now, there
is no relevant desert base. If Bob is blameworthy for doing something, he is
blameworthy partly in virtue of the fact that he freely does that thing.
But as of now, he has not committed any crime, and as of now, it is open
whether he will commit any crime. So he cannot, as of now, be blamewor-
thy for stealing Ann’s car at noon.
I believe it is some libertarian notion of openness that lies at the heart of
this entire line of reasoning. The notion requires elucidation before this
reasoning can be clarified and evaluated. The basic idea underlying liber-
tarian openness (hereafter ‘L-openness’) is that if it is L-open to an agent
to do something, A, at a time, t, then the agent can do something other
than A at t consistent with the past and the laws of nature remaining un-
changed. Somewhat more rigorously, it is L-open to S to do A at t if and
only if S is not causally determined to do A at t. S is not causally determined
to do A at t when S does A at t if and only if S does A at t, and there is a pos-
sible world that is (a) past-wise indiscernible (with respect to ‘genuine’ or
nonrelational facts) from the actual one until just prior to either the
moment at which S does A, or an earlier moment at which S does some-
thing else, B, such as forming a judgment to A or an intention to A; (b)
governed by the same laws of nature as the actual world, and (c) such that,
in it, S does other than A at t.2
It’s important to distinguish between ‘indeterministic decision making’

and L-openness. I’ll use a Frankfurt-type example to illustrate the distinction.
The typical Frankfurt-type example is one where an agent seemingly has
various alternatives from which to choose. He makes his choice and per-
forms the action in question not knowing that, had he revealed even the
slightest inclination to act differently, something, a ‘counterfactual inter-
vener’, would have forced him to act as he in fact did. Since he acted on his
own, in that the intervener did not play any role in his action—it’s just that
the intervener would have forced him to act as he did had he shown any
signs of acting differently—it seems that nothing stands in the way of his
being morally responsible for his deed.3 It will be useful, for later discus-
sion, to summarize an imaginative nonstandard version of a Frankfurt-type
case recently advanced by Alfred Mele and David Robb (1998, 101–102) in
which ‘counterfactual intervention’ plays no role:
At t1, Black initiates a certain deterministic process P in Bob’s brain with the inten-
tion of thereby causing Bob to decide at t2 (an hour later, say) to steal Ann’s car. The
process, which is screened off from Bob’s consciousness, will deterministically cul-
minate in Bob’s deciding at t2 to steal Ann’s car unless he decides on his own at t2
to steal it or is incapable at t2 of making a decision (because, for example, he is dead
by t2). (Black is unaware that it is open to Bob to decide on his own at t2 to steal the
car; he is confident that P will cause Bob to decide as he wants Bob to decide.) The
process is in no way sensitive to any ‘sign’ of what Bob will decide. As it happens, at
t2 Bob decides on his own to steal the car, on the basis of his own indeterministic
deliberation about whether to steal it, and his decision has no deterministic cause.
But if he had not just then decided on his own to steal it, P would have determinis-
tically issued, at t2, in his deciding to steal it. Rest assured that P in no way influ-
ences the indeterministic decision-making process that actually issues in Bob’s
decision.4
The case provides healthy support for the view that Bob is morally respon-
sible for deciding to steal the car (and, subsequently, for stealing the car)
even though he could not have decided (and then done) otherwise. Fur-
ther, the case highlights the distinction between indeterministic decision
making and L-openness. We can assume that it is causally undetermined
that certain of Bob’s nonoccurrent beliefs will enter into his deliberations
about whether to steal the car, and that if the beliefs do come to Bob’s
mind, he will decide (on his own) to steal the car.5 In this sort of case, in-
determinism is located relatively early in the causal pathway of an action at
some point prior to the making of a decision, and the event that is
140 I. Haji
nondeterministically caused, such as the forming of a judgment to do

something, that, in turn, causes the relevant decision, is not a mental
action. Alternatively, more apt for our purposes, it could be that the event
that is nondeterministically caused is the making of a decision itself. Given
this latter type of indeterminacy assumed to be present in Bob’s scenario,
Bob’s decision to steal the car has no deterministic cause. His deliberation
or decision making that culminates with the decision to steal is thus ‘inde-
terministic’. However, as Mele and Robb (1998, 103) emphasize, given the
details of the case, any future open to Bob after the initiation of determin-
istic process P in which Bob is capable of making a decision includes his
deciding at t2 to steal the car. It is not L-open to Bob at t2, then, to decide
to steal the car at t2. For Bob cannot, as of t2, refrain from deciding to steal
the car at t2.
Appealing to the notion of L-openness, a defender of blame past (or a
foe of blame future) might claim that such openness is incompatible with
blameworthiness for a future action. For suppose it is L-open to S, as of t0,
to do A at t1. Then, as of t0, both S can do A at t1 and S can refrain from
doing A at t1. So, as of t0, there is a chance that S will not do A at t1. How,
then, it might be challenged, can S at t0 already be blameworthy for what
S will do at t1 (or, in short, for doing A at t1)?
Prior to picking up the gauntlet, let’s note that if it is L-openness that is
meant to underpin the view that blame future is false, some causal deter-
minists may well regard the explanation of blame future’s falsity that in-
vokes such openness with suspicion. Determinists hold that, for any given
time, a complete statement of all the nonrelational or ‘genuine’ facts about
that time, together with a complete statement of the laws of nature, entail
every truth as to what happens after that time. Arguably, if causal deter-
minism is true, then no one can ever do other than what he or she in fact
does.6 If this is so, causal determinism is incompatible with L-openness.
Yet, such determinists who are compatibilists about determinism and re-
sponsibility champion the view that people can be responsible for at least
some of their actions. Suppose causal determinism is true, people can be
blameworthy for some of their actions, and blame future is false. Then
compatibilists can reasonably claim that the explanation of why blame
future is false cannot, in any way, depend on L-openness. In the rest of this
paper, though, I limit discussion to the L-openness rationale for the alleged
falsity of blame future.
3 Concerns with the L-Openness Rationale
The central claim embedded in the reasoning under consideration against

blame future is this: if it is L-open to S as of t0 to do A at t1, then S cannot
be blameworthy at t0 for doing A at t1. But I believe this claim can be re-
sisted as the following example suggests. Suppose it is causally determined
that Sid will enter a certain room at 11:30 A.M. and not leave that room by
noon. Suppose wishing to kill Sid, Rob has hidden a bomb in the room.
Nobody save Rob knows about the whereabouts of this bomb nor has any
inkling about Rob’s intentions to murder Sid. The bomb can be activated
by a remote control device in Rob’s possession. On the basis of indeter-
ministic deliberation, Rob decides at 11:00 to activate the bomb at 11:30 by
using the device. Once activated, the bomb will detonate at noon. Assume
that, as of 11:00, it is L-open to Rob to act on his decision to activate the
bomb at 11:30. He is resolute about his decision; once made, nothing what-
soever will change his mind. Further, nobody else will intervene as no one
save Rob knows about his plan to kill Sid, and no one will uncover that
plan even after Sid’s demise. Having executed his deadly decision at 11:30,
Rob boards a plane to Brussels. The bomb detonates at noon and Sid is
killed. The judgment that, as of 11:00 (and thereafter), Rob is blameworthy
for Sid’s death at noon seems perfectly reasonable. It seems reasonable pri-
marily because, as of 11:00, for all rhyme and reason, Sid’s fate is as good
as sealed. Alternatively, we should note that, granting libertarianism, it is
arguable that, even if one does in fact freely perform an action at some
time, t2, then it was true though not settled at some earlier time, t1, that
one would perform that action at t2. In this case (call it ‘bomb1’) Rob is
responsible for, among other things, a ‘distal’ consequence of an action
(activating the bomb) that he performed earlier.
Suppose, to vary the case a bit, Rob decides resolutely at 11:30, on the basis
of his own indeterministic deliberation, to push the ‘kill now’ button on the
remote device at 11:45, a button that, if pushed, will result in the bomb’s
going off at just about when the button is pushed. Assume that, as of 11:30,
it is L-open to Rob that he will push the button at 11:45. But once he has
arrived at the pertinent decision, Rob won’t waver; nothing will change his
mind, and nobody will intervene. By 11:30 Sid’s fate is as good as sealed. In
this variation, we are assuming that Rob acts with libertarian free will and
freely pushes the button at 11:45. But then, arguably, it was true at, say, 11:30
142 I. Haji
even though it wasn’t settled then, that Rob would push the button at 11:45.
Call this case ‘bomb2’. Yet again it seems that the verdict that, as of 11:30
(t2), Rob is blameworthy for pushing the button at 11:45 (t3) is reasonable.
Not everyone will agree with this verdict. One might contest that, in
this context, ‘as good as sealed’ can only mean ‘nearly sealed’, and so ‘not
sealed’. As long as the action is L-open, it cannot be settled or truly
inevitable but only highly improbable, in which case the agent cannot be
responsible for it until it occurs. In this respect, the case differs from one
in which, say, should one now perform some action, it is inevitable that
an upshot of that action will occur.
In defense of my verdict, compare bomb2 with a modification of Mele
and Robb’s Frankfurt-type case. In the modified case, at t1, Black initiates a
deterministic process P in Rob*’s brain with the intention of thereby caus-
ing Rob* to decide at t2 to press the ‘kill now’ button at t3 on a remote de-
vice, one just like the device that Rob has. Pressing the button will detonate
the bomb in Sid*’s room. P, screened off from Rob*’s consciousness, will
deterministically culminate in Rob*’s deciding at t2 to press the ‘kill now’
button at t3 unless he decides on his own at t2 to press the button at t3 or
is decisionally incapacitated at t2. (Black is unaware that Rob* can decide
on his own at t2 to press the button at t3; he is confident that P will cause
Rob* to decide as he wants Rob* to decide.) It happens that Rob* decides on
his own at t2 to press the button at t3, on the basis of his own indetermin-
istic deliberation. Rob* is resolute about this decision, and once made, no-
body or nothing will intervene. But if he had not just then decided on his
own to press the button, P would have deterministically issued, at t2, in his
deciding to press it. P in no way influences the indeterministic decision-
making process that actually issues in Rob*’s decision.
Now let’s add one more layer of complexity to the case. Imagine that, at
t1, Black also initiated another deterministic process, Q , in Rob*’s brain
with the intention of thereby causing Rob* to press the ‘kill now’ button
at t3. Q, just like P, screened off from Rob*’s consciousness, will determin-
istically culminate in Rob*’s pressing the ‘kill now’ button at t3 unless he
presses it on his own at t3 or is physically incapacitated at t3. (Black is,
again, unaware that Rob* can press the button on his own at t3; he is con-
fident that Q will cause Rob* to act as he wants Rob* to act.) It turns out
that Rob* acts on his own at t3, pressing the button at t3 on the basis of his
own indeterministic intention-forming capacities. We can suppose that,
having made the indeterministic decision at t2 to press the button at t3,

Rob* had to form various other intentions to see to it that he would press
the ‘kill now’ button at t3. For example, he had to form the intention to
switch on the remote device and to have the device handy so that he could
press the ‘kill now’ button at t3. Assume that the processes culminating in
these intentions are indeterministic. Had Rob* not, at t3, pressed the but-
ton on his own, Q would have deterministically issued, at t3, in his press-
ing it. Q in no way influences the indeterministic processes that actually
issue in Rob*’s pressing the button at t3.
In this modified version (‘bomb3’) of the Mele/Robb scenario, it certainly
seems that Rob* is morally blameworthy at t2 for deciding at t2 to press the
button at t3. Further, it also seems that, as of the time he makes his deci-
sion, he is blameworthy then, too, for his future action of pressing the but-
ton at t3. He is blameworthy, roughly, in virtue of the fact that, given the
decision at t2 to press the button, Rob*’s pressing the button at t3 is
inevitable. In bomb3, though, Rob* cannot refrain from making the decision
that he does at t2 and he cannot refrain from pressing the button at t3. So it
is not an option for an advocate of blame past to rejoin that, at t2, Rob*
cannot already be blameworthy for pressing the button at t3 as it is L-open
to Rob*, at t2, to press the button at t3.
Assume that the actual sequence of events that culminates in Rob*’s press-
ing the ‘kill now’ button at t3 in bomb3 is type-identical to the actual
sequence of events that culminates in Rob’s pressing the ‘kill now’ button at
t3 in bomb2. The assumption is reasonable: In bomb3, Rob* decides at t2,
on the basis of his own indeterministic deliberation, to press the ‘kill now’
button at t3; and, at t3, he presses the button on his own. Deterministic
processes P and Q play no role whatsoever in the actual sequence of events
that culminate in Rob*’s decision to press the button and Rob*’s pressing
the button. Similarly, in bomb2, Rob decides on the basis of his own
indeterministic deliberation, to press the ‘kill now’ button at t3; and, at t3,
he presses the button on his own.
One of the deep morals of Frankfurt-type cases is that responsibility
depends on what unfolds in the actual sequence of events leading to action
and not on ‘alternative’ sequences (such as, for instance, the alternative se-
quence in bomb3 in which process P influences Bob’s decision at t2). As the
actual sequence of events culminating in action—the decision to press the
‘kill now’ button and pressing the ‘kill now’ button—in bomb2 and bomb3
144 I. Haji
are (arguably) type-identical, if Rob* is blameworthy, as of t2, for pressing

the button at t3 in bomb3, then Rob should also be blameworthy, as of t2,
for pressing the button at t3 in bomb2. I see no reason to believe that Rob*
is not, at t2, deserving of blame for his decision to press the button at t2
and for pressing the button at t3; indeed, there is every reason to believe
that he is blameworthy, at t2, for these things. (Again, one cannot appeal to
L-openness to argue against Rob*’s being blameworthy, as of t2, for pressing
the button at t3. For in bomb3, it is not L-open to Rob*, as of t2, to press
the button at t3.) But then, given type-identity of the actual sequences of
events in bomb2 and bomb3, there appears to be no reason to deny that
Rob (in bomb2) is also blameworthy, as of t2, for pressing the button at t3,
even though in bomb2 it is L-open to Rob, as of t2, to press the button at t3.7
4 Blame Future and Conceptions of Responsibility
I want to end with the suggestion that the controversy over which of blame
future or blame past is true can be explained partly by endorsement, at
some level, of different conceptions of moral responsibility. Consider, for
example, R. J. Wallace’s Strawsonian account of responsibility (1994, 91).
On Wallace’s view, a person is morally responsible for some action if and
only if it would be appropriate to hold her responsible for that action. As
moral norms of fairness, Wallace contends, set the standards of appropri-
ateness for responsibility ascriptions, this view is to be construed as the
normative one that a person is morally responsible for an action if and only
if it is fair to hold her responsible for that action. Wallace proposes that to
hold someone morally responsible is essentially to be subject to the moral
sentiments of guilt, indignation, or resentment in one’s interactions with
the person. In cases in which the standard excuses such as coercion, phys-
ical constraint, duress, mistake, accident, and inadvertence apply, Wallace
contends that the agent has not violated a moral obligation, and conse-
quently, does not deserve to be subjected to the reactive emotions and the
forms of sanctioning treatment that express them. With the exemptions—
psychopathy, behavior control, stress, insanity, addiction, childhood, hyp-
notism, deprivation, or torture—Wallace submits that the agent has been
deprived of the ability to grasp and apply moral reasons, and the ability to
control her behavior by such reasons. But if holding an agent responsible
involves judging it appropriate to sanction her morally, and to sanction her
as a result of the violation of moral obligations for which there are moral
reasons, then it will, once again, be unfair to do this if she lacks the ability
to apply the moral reasons that sustain the obligations violated.
Of pertinence to our present concerns is that, if one subscribes to such a
normative conception of responsibility, then it is quite natural to affirm
blame past. For just as it may be unfair to blame a person for having done
something that isn’t wrong in the first place or (as with prepunishment) for
not yet having committed a crime, so, it may reasonably be contended, it
would be unfair to hold a person blameworthy now for an action that she
has yet to perform. As we rarely know precisely what the future will bring,
there is almost always the chance that the person won’t perform the perti-
nent action. It would, hence, seem unfair to react to this person, in advance
of her performing this action, with resentment or indignation.
John Fischer (1994), in an elegant paper, has recently proposed that the
intuitive picture or conception of responsibility that fuels the view that
alternative possibilities are required for moral responsibility is the concep-
tion that moral responsibility requires the ability to make a sort of difference
to the world. Fischer explains that if one makes a difference in this sense, one
selects which path the world will take among various paths that are gen-
uinely accessible. One’s selection determines which way the world unfolds
and one thereby makes a crucial difference. Again, if one is driven by this
conception of responsibility, then one will be attracted to blame past. A pro-
ponent of this conception of responsibility might sensibly propose that as a
prospective action has still to be performed, its agent has not yet made a dif-
ference regarding the pathway the world may take by selecting that action.
Hence, the agent cannot, prior to the fact, be responsible for that action.
Contrast these two conceptions of responsibility with a third according
to which, roughly, when a person is responsible for what she does, she
discloses what she morally stands for in performing that action. This sort
of self-disclosure view has been proposed, among others, by John Dewey
(1957, 160–161) who wrote:
[W]hen any result has been foreseen and adopted as foreseen [by the agent], such
result is the outcome not of any external circumstances, not of mere desires and
impulses, but of the agent’s conception of his own end. Now because the result thus
flows from the agent’s own conception of an end, he feels himself responsible for it
. . . . The result is simply an expression of himself; a manifestation of what he would
have himself to be. Responsibility is thus one aspect of the identity of character and
146 I. Haji
conduct. We are responsible for our conduct because that conduct is ourselves objectified in
actions.8
Regarding blameworthiness, I have elsewhere (1998, ch. 9) argued for the

condition that a person is blameworthy for an action only if she performs
it in light of the belief that she is doing wrong. If this condition is in the
right ballpark, then when a person is indeed blameworthy for an action,
there is a very real sense in which she ‘discloses’ what she morally stands for
in performing that action, as when she acts, she does so in light of certain of
her moral beliefs. Responsibility has this sort of self-expressive requirement.
To elaborate briefly, according to the self-expressive conception, when one
is responsible for what one does, one manifests or expresses in conduct
what one would have oneself to be. Acting in light of certain moral beliefs,
one discloses one’s moral stance vis à vis a particular episode in one’s life.
When one is morally blameworthy for an action, for instance, one ex-
presses ill will toward another, and it is the expression of such ill will that
(in part) sanctions the judgment that one’s moral worth with respect to the
relevant deed has been diminished. Analogously, when one is morally
praiseworthy for an action, one expresses good will toward another, and it
is the expression of such good will that (in part) validates the judgment that
one’s moral worth vis à vis the relevant deed has been augmented.
If this intuitive picture of responsibility is correct, then the time at which
one discloses or expresses what one morally stands for in performing an
action seems irrelevant to ascriptions or correct judgments of blamewor-
thiness. In the sorts of cases considered in which prospective blameworthi-
ness seems apt, at a time prior to the performance of the relevant act, the
agent has, in a manner of speaking, disclosed his moral stance in relation
to the act to be performed. In the modified version of bomb2, for example,
Sid has, so to speak, disclosed at 11:30—the time at which he decides res-
olutely to press the ‘kill now’ button at 11:45—what he morally stands for
in relation to his future act of pressing the ‘kill now’ button at the time
when he does. Metaphorically speaking, the ‘statement’ about what he
morally stands for—the statement that reveals something about his moral
worth in connection to his future action—has already been ‘made’ at 11:30
but will not be ‘announced’ or ‘revealed’ to the world until the deed is ac-
tually done. Given this intuitive picture of responsibility, there is nothing
untoward about, indeed, there is a lot to recommend, blame future.
Needless to say, this is not the place to decide among the various possible
conceptions of moral responsibility. My proposal is simply that if one finds
the self-disclosure picture of responsibility promising, one may well be
willing to renounce blame past in favor of blame future.
In conclusion, focusing on central cases in which persons are morally to
blame for their actions, a number of people have contended that blame-
worthiness, and responsibility in general, are essentially retrospective. I
have proposed that this view is called into question by Frankfurt-type
examples.9 Puzzling over why so many have been attracted to this mis-
taken view, I have further proposed that proponents of this view may be in
the grips of a certain conception of responsibility, like Wallace’s normative
conception, or the make-a-difference conception, that renders dim the pos-
sibility of prospective blameworthiness. Finally, I have suggested that there
is an alternative, intuitive conception of responsibility—the self-disclosure
conception—which lends support to the contrasting view that persons can
be prospectively responsible for some of their actions.
Acknowledgments
I thank Randolph Clarke, Roderick Long, Michael McKenna, and Michael J.

Zimmerman for their criticisms, comments, and suggestions.
Notes
1. A number of people have endorsed this view or at least something very similar to
it. For example, Wojciech Sadurski (1985, 117) says “desert considerations are always
past oriented. When talking about desert, we are evaluating certain actions which
have already happened.” Brian Barry (1965, 111) says, “Desert looks to the past—or
at most to the present—whereas incentive and deterrence are forward looking no-
tions . . . .” David Miller (1976, 93) claims that “Desert judgments are justified on the
basis of past and present facts about individuals, never on the basis of states of affairs
to be created in the future.”
2. Conditions (a), (b), and (c) are advanced and explained by Peter van Inwagen
(1983, 136).
3. See Harry G. Frankfurt (1969).
4. Derk Pereboom (2000, 119–137) has advanced another sort of Frankfurt-type case
involving indeterministic decision making. In his case, the counterfactual intervener
148 I. Haji
makes the agent (in a nondeterministic world) decide to cheat on his tax deduction
if he doesn’t (nondeterministically) do so on his own. The agent’s psychology is such
that a causally necessary (but not causally sufficient) condition for the agent to
decide not to cheat is that a moral reason occurs to him with a certain force. Such a
reason’s occurring to the agent is the cue for intervention.
5. Mele (1995, 211–221) discusses this sort of doxastic indeterminacy.
6. For a defense of this view see, for example, Carl Ginet (1983), John Martin Fischer
(1994) and Peter van Inwagen (1983).
7. What of the following case though? In bomb4, it is open to Job, as of t0, to decide
at t1 to press the ‘kill now’ button at t6 and to press the ‘kill now’ button at t6. If the
button is pressed at t6, Cid will be blown to bits. However, unlike Rob and Rob*, Job
is unsure about whether to kill Cid. Let’s suppose that he first decides at t1, on the
basis of his own indeterministic deliberation, to press the button at t6. A little while
later, he changes his mind. At t3, he decides, again on the basis of indeterministic de-
liberation, not to press the button at t6. But he is still not settled in his mind about
what to do. Just before t6, he decides on his own once again to press the button, and
at t6 he presses the button. Is Job at, say, t2, blameworthy for pressing the button at
t6? I confess to being unsure about this case. I’m inclined to say that he isn’t blame-
worthy at t2. If this verdict is correct, then there is a kernel of truth in the idea that
L-openness is incompatible with blameworthiness for future actions. It’s this: both
Rob and Rob* are resolute about their decision to kill. They successfully execute
every element in their plan to kill without indecision or unsettledness of mind. It is
as if, from their perspective, L-openness does not matter; from their perspective,
every future includes the decision to kill and the subsequent killing. Similar things
are not true of Job.
8. Gary Watson (1996) discusses aspects of Dewey’s view.
9. It appears that both Fred Feldman and Michael Zimmerman would agree with the
conclusion that blame past is false. Feldman (1997) rejects the following view, DT: If
at t, S deserves x in virtue of the fact that S did or suffered something at t’, then t’ can-
not be later than t, primarily by appealing to putative counterexamples. He says that
several people, including John Cottingham, have accepted DT or something very
similar. Feldman claims that Cottingham seems to be asserting, among other things,
that I cannot blame you for doing something unless you have already done it. How-
ever, Feldman claims: “If, as a result of your failure to drain the pipes, there is going
to be damage to the plumbing later tonight when the temperature drops, I can blame
you for the damage that is going to occur” (1997, 180–181). This example, though
suggestive, falls short of establishing prospective blameworthiness both for future
actions with respect to which one is directly free, and for noninevitable upshots of
one’s present actions. Michael Zimmerman (1988) asks: if S wills (that is, roughly, de-
cides) at t that e should occur, at what time is it true to the facts that S is blameworthy
for willing e at t? The answer, he says, is: at that time at which it is true that (to speak
tenselessly) S wills e at t. Zimmerman adds that if S wills e at t and is culpable for
this, then S is culpable at t and forever thereafter. In addition, Zimmerman says that
“if there can be true propositions about the free volitions and actions of persons in
the future, then we should also admit that persons are forever (period) (that is, not
just forever after, but also forever before) culpable for those volitions for which they
are culpable.” (1988, 2, 40). Zimmerman seems to be suggesting that responsibility is
not essentially retrospective. Even on this tenseless approach, one might still want
an explanation of what conception of responsibility, if any, can accommodate
prospective blameworthiness.
References
Barry, B. 1965. Political Argument. London and New York: Routledge and Kegan Paul.
Cottingham, J. 1992. “Justice; Rectificatory.” In Lawerence Becker, ed., Encyclopedia

of Ethics. New York and London: Garland Publishing.
Dewey, J. 1957 (1891). Outlines of a Critical Theory of Ethics. New York: Hillary House.
Feldman, F. 1997. “Desert: Reconsideration of Some Received Wisdom.” In F. Feldman,

Utilitarianism, Hedonism, and Desert. New York: Cambridge University Press.
Fischer, J. 1994. The Metaphysics of Free Will. Cambridge, Mass.: Blackwell.
. 1999. “Responsibility and Self-Expression.” Journal of Ethics 3: 277–297.
Frankfurt, H. 1969. “Alternate Possibilities and Moral Responsibility.” Journal of

Ginet, C. 1983. “In Defense of Incompatibilism.” Philosophical Studies 44: 391–400.
Haji, I. 1998. Moral Appraisability: Puzzles, Proposals, and Perplexities. New York:
Mele, A. 1995. Autonomous Agents: From Self-control to Autonomy. New York: Oxford
University Press.
Mele, A. and Robb, D. 1998. “Rescuing Frankfurt-style Cases.” Philosophical Review

107: 97–112.
Miller, D. 1976. Social Justice. Oxford: The Clarendon Press.
Pereboom, D. 2000. “Alternative Possibilities and Causal Histories.” Philosophical

Perspectives 14: 119–137.
Rachels, J. 1991. “What People Deserve.” In John Arthur and William H. Shaw, eds.,
Justice and Economic Distribution. Englewood Cliffs, N.J.: Prentice-Hall.
150 I. Haji
Sadurski, W. 1985. Giving Desert Its Due: Social Justice and Legal Theory. Dordrecht:
Reidel.
van Inwagen, P. 1983. An Essay on Free Will. Oxford: Oxford University Press.
Wallace, R. 1994. Responsibility and the Moral Sentiments. Cambridge, Mass.: Harvard
University Press.
Watson, G. 1996. “Two Faces of Responsibility.” Philosophical Topics 24: 227–248.
Zimmerman, M. 1988. An Essay on Moral Responsibility. Totowa, N.J.: Rowman and

Littlefield.
7 Moderate Reasons-responsiveness, Moral
Responsibility, and Manipulation
Todd R. Long
I may be a bird in a cage, but at least it’s your cage.

—Old 97’s
Frankfurt-type examples have proven to be quite helpful in our think-

ing about the relation between freedom and moral responsibility. Harry
Frankfurt designed his own examples to show that an agent can be morally
responsible even if that agent could not have done otherwise (Frankfurt
1997). Following Frankfurt’s lead, many philosophers have developed their
own Frankfurt-type examples to extend our understanding and sometimes
to exercise our intuitions about the limits of moral responsibility and
freedom.1 In this essay, I join this enterprise by providing a novel
Frankfurt-type example designed to reveal the limits of moral responsibil-
ity, given a new theory of responsibility developed by John Martin Fischer
and Mark Ravizza. Although I am sympathetic to the mechanism-based
theory of moral responsibility that Fischer and Ravizza (hereafter, F & R)
argue for in Responsibility and Control, I believe there is trouble for the way
in which they apply their theory to Frankfurt-type examples. I will show
that F & R’s theory handles my test cases appropriately, but accepting my
conclusion requires us (and them) to deny the seemingly plausible intu-
ition that certain cases of rather severe external manipulation rule out
moral responsibility.
In what follows I first summarize F & R’s notion of guidance control with
an eye toward explaining their preferred notion of reasons-responsiveness
(namely, ‘moderate reasons-responsiveness’). Second, I show how F & R
want moderate reasons-responsiveness to hook up with the relevant delib-
erative mechanism. Third, I offer a new Frankfurt-type example that both
challenges F & R’s own application of their theory to Frankfurt-type
152 T. R. Long
examples and threatens a certain appealing intuition that some people have
about the relation between moral responsibility and external manipulation.
1 Guidance Control and Reasons-Responsiveness
In Responsibility and Control, F & R argue that, in the free will debate, guid-
ance control is the freedom-relevant condition that is sufficient for moral
responsibility.2 Guidance control does not require the sort of regulative con-
trol that depends upon the existence of genuine alternative possibilities.
Although regulative control often goes hand in hand with guidance control,
F & R hold that moral responsibility requires only guidance control. Thus, we
are morally responsible for our actions so long as we have guidance control,
even if we lack regulative control. Pointing out that guidance control is the
sort of control we find illustrated in the Frankfurt-type actual-sequence ex-
amples, F & R conclude that guidance control is what ultimately grounds our
ascriptions of moral responsibility. This is a familiar compatibilist claim.
What sets F & R’s theory apart is that F & R ascribe guidance control to an
agent insofar as that agent’s deliberative mechanism is appropriately respon-
sive to reasons. Hence, their theory hinges on a satisfactory account of the de-
liberative mechanism operant in agents. I will return to mechanisms shortly.
Another significant feature of F & R’s notion of guidance control is their
distinction between moral responsibility, on the one hand, and praisewor-
thiness or blameworthiness, on the other. F & R characterize their notion of
moral responsibility as a “Strawsonian view” that “allows for moral re-
sponsibility for ‘morally neutral’ behavior. For instance, one can be morally
responsible for simply raising one’s hand” (Fischer and Ravizza 1998, 8). All
that is required for moral responsibility is that one be “an appropriate
candidate for the reactive attitudes” (Fischer and Ravizza 1998, 7). So, one
can be morally responsible for an action without being an actual target of
a reactive attitude. Nevertheless, moral responsibility is connected to praise
and blame; after all, one would not even be a candidate for praise or blame
unless one were morally responsible, in F & R’s sense, for one’s actions. We
can summarize the key point here by saying that although guidance con-
trol (and, thus, moral responsibility) does not necessarily imply praisewor-
thiness or blameworthiness, the latter do imply the former.3
F & R refine their notion of guidance control by connecting it to the notion
of reasons-responsiveness.4 Their idea is that we are morally responsible for an
Moderate Reasons-responsiveness 153
action when the mechanism that issues in that action is reasons-responsive in

an appropriate way. We can get a quick handle on the concept of reasons-
responsiveness by thinking about contrasting cases. Consider Bill, who has a
neurological disorder, which causes him to throw his head back violently
every few seconds. Clearly, Bill lacks guidance control for his violent head
movements. F & R claim that there is some physical mechanism issuing in
Bill’s behavior that has nothing to do with responding to reasons; therefore,
we do not judge him to be morally responsible for his head movements. Now
consider Amy, who is deliberating about whether to give money to a worthy
charity; as she deliberates she may consider reasons for and against giving the
money. Suppose she decides to give the money. Since her giving the money
is not induced by a process such as a neurological disorder, a psychotic im-
pulse, or brain manipulation, F & R hold that Amy’s action is issued in by a
mechanism that is responsive to reasons, and thus we think it proper to praise
her for her charitable action; hence, she is morally responsible for her action.
F & R settle on a version of reasons-responsiveness that they call moder-
ate reasons-responsiveness (or, MRR).5 The following schema provides a fuller
explanation of MRR:
MRR:6 An agent’s responsibility relevant mechanism K is moderately

reasons-responsive iff
1. K is regularly receptive to reasons, some of which are moral; this requires

(a) that holding fixed the operation of a K-type mechanism, the agent
would recognize reasons in such a way as to give rise to an understand-
able pattern from the viewpoint of a third party who understands the
agent’s values and beliefs; and (b) that some of the reasons mentioned in
(1.a) are moral reasons; and
2. K is at least weakly reactive to reasons; this requires that the agent would
react to at least one sufficient reason to do otherwise (in some possible sce-
nario), although it does not follow that the agent could have responded
differently to the actual reasons; and
3. K is the agent’s own; being the agent’s own means ‘taking responsibility’
for K; this requires that the agent (a) sees herself as the source of her
behavior (which follows from the operation of K); and (b) believes that
she is an apt candidate for the reactive attitudes as a result of how she
exercises her agency in certain contexts; and (c) views herself as an agent
with respect to (3.a)–(3.b) based on her evidence for these beliefs.
154 T. R. Long
F & R hold that MRR is sufficient for moral responsibility.7 Claims of suffi-
ciency are customarily tested with specific cases. Any case in which an
agent’s mechanism is MRR but the agent is not morally responsible for an ac-
tion issued in by that mechanism will count as a counterexample to F & R’s
claim of sufficiency. I will provide test cases in section 3. But first a word
about mechanisms.
2 F & R’s Human Deliberative Mechanisms
Determining which of various mechanisms is actually operant in a given

case is an important issue for F & R’s account. What are the necessary and
sufficient conditions for a mechanism to be the relevant mechanism in a
given case of deliberation? F & R offer little help:
We cannot specify in a general way how to determine which mechanism is “the”
mechanism that is relevant to assessment of responsibility. It is simply a presupposi-
tion of this theory as presented here that for each act, there is an intuitively natural
mechanism that is appropriately selected as the mechanism that issues in action, for
the purposes of assessing guidance control and moral responsibility. (Fischer and
Ravizza 1998, 47)
Nonetheless, F & R provide examples intended to help. One of their exam-

ples, which is about a case of direct brain manipulation, is useful for my
present project:
Consider, again, our claim that certain sorts of “direct manipulation of the brain”
rule out moral responsibility. When the scientists intervene and manipulate the
brain in a way which is not reasons-responsive, this is sufficient to undermine an
agent’s moral responsibility for action. Similarly, when forms of subliminal advertis-
ing, hypnosis, brainwashing, and so forth yield physical mechanisms that are not
reasons-responsive, they rule out moral responsibility. Thus the theory that associ-
ates moral responsibility with actual-sequence reasons-responsiveness can help to
explain our intuitive view that certain sorts of direct interventions in the brain are
inconsistent with moral responsibility. (Fischer and Ravizza 1998, 48–49)
Here is the crucial idea that F & R want us to get: any MRR mechanism that
operates to issue in my actions is always a different mechanism from any
one that would operate if my brain were monkeyed with to the point that I
am no longer morally responsible for my actions. Hence, a brain-implanted
mechanism that issues in an action would always constitute a different
mechanism from an ordinary practical reasoning mechanism that issues in
an action.
So far, so good. However, F & R also think that there is a close connection
between the distinction we’ve just noted and Frankfurt-style examples.
Specifically, they hold that the mechanism operant in an actual-sequence
case is always to be distinguished from the mechanism operant in an
alternative-sequence case. Consider Fischer’s discussion of such cases in
The Metaphysics of Free Will:
In a Frankfurt-type case the kind of mechanism that actually operates is reasons-
responsive, although the kind of mechanism that would operate in the alternative
scenario is not. In the Frankfurt-type case discussed above (in which Jones votes for
Clinton on his own and Black does not actually intercede), Jones’ action issues from
the normal faculty of practical reasoning, which we can reasonably take to be
reasons-responsive. But in the alternative scenario, a different kind of mechanism
would have operated—one involving direct electronic stimulation of Jones’ brain.
And this mechanism is not reasons-responsive. Thus, the actual-sequence mecha-
nism can be reasons-responsive, even though the agent is not reasons-responsive.
(He couldn’t have done otherwise.) . . . Clearly, . . . it is crucial to distinguish between
the kind of mechanism that operates in the actual sequence and the kind of mecha-
nism that operates in the alternative sequence (or sequences).8 (Fischer 1995, 163)
I want to make two points about Fischer’s discussion here. First, let’s consider
a couple of features of this alternative-sequence case: (1) There is brain ma-
nipulation that leads to a particular outcome; and (2) The agent would have
done otherwise had there been no manipulation. Now, one might think that
what underlies our intuition that the agent lacks moral responsibility in
such a case is some kind of manipulation principle that addresses those
features. As we have seen, F & R hold that a brain-implanted mechanism
issuing in an action is a different mechanism from an ordinary practical
reasoning mechanism. F & R say that this is just intuitively the case, but per-
haps we can explain this intuition by drafting a manipulation principle that
addresses the two factors we just noted about alternative-sequence cases:
MAN: If a person S performs an action A at time tn only because another
agent B, acting completely unbeknownst to S, has directly manipulated
S’s brain earlier at t1 then S is not morally responsible for performing A.9
We might think that F & R would happily endorse such a principle, for it
helps to explain the intuition that a person is not responsible for an action
if her brain has been directly manipulated in a certain way. More specifically,
it helps to explain the appealing intuitive view that a person is not morally
responsible for performing an action if the action is such that he would have
refrained from performing it if his brain had not been directly manipulated
156 T. R. Long
by an external source. This is, after all, true of Fischer’s Frankfurt-style case,
wherein the agent is morally responsible in the actual sequence but is not
morally responsible in the alternative sequence; and MAN would explain
why Fischer concludes the quote above by saying that it is crucial to distin-
guish the mechanism operant in the actual-sequence case from the mecha-
nism operant in the alternative-sequence case.
This leads me to the second point I want to make. Clearly, it is crucial to
make this distinction only if it is true that the mechanism operant in the
alternative-sequence case is always a different mechanism from the one
operant in the actual-sequence case. But since F & R give us so little to go
on concerning what a relevant mechanism is, or how to differentiate mech-
anisms, I see no reason to think that the mechanism that operates in
the actual sequence of a Frankfurt-type example must be different from the
mechanism that operates in an alternative sequence of a Frankfurt-type
example. As I will show in section 3, we can tell a Frankfurt-style alternative-
sequence story in which the mechanism issuing in the action is the very
same mechanism as the one operating in the actual-sequence case. The
basic idea that I will develop there is that the results yielded by the human
deliberative mechanism will vary depending on its inputs. It is very plausi-
ble to think that differing inputs (e.g., reasons) will yield differing outputs
(e.g., actions). And in their latest account, F & R grant that their concept of
the relevant deliberative mechanism in a given case involves the idea that
the same mechanism may operate on a variety of inputs:
. . . a mechanism’s reacting differently to a sufficient reason to do otherwise in some
other possible world shows that the same kind of mechanism can react differently to
the actual reason to do otherwise. . . . The picture here is of one kind of mechanism
with different “inputs.” (Fischer and Ravizza 1998, 73–74)
F & R assume that in any Frankfurt-style example, the mechanism in an

actual-sequence story is always a different one from the mechanism in an al-
ternative-sequence story (call this F & R’s distinction thesis). But, given that
the same mechanism can yield different outputs depending on its inputs, we
can conceive an alternative-sequence case as one in which the only differ-
ence between it and the actual-sequence case is the inputs—the mechanism
remains the same.
In Fischer (1995), Fischer says little in support of his claim that the two
mechanisms cannot be the same. Fischer suggests only that the mechanisms
are different in the two cases because the causes are different; but that alone
is not a good reason to accept his claim. After all, a light can be turned off by
a human hand’s flicking a light switch, and the same light can be turned off
by falling plaster that strikes a light switch. Here we have two different causes
for the light’s going out, but we have one and the same mechanism (the light
switch). In Fischer and Ravizza (1998), F & R provide no additional reasons
for thinking that the two mechanisms cannot be the same; rather, they just
appeal to intuition to have us think that they must be different.10 I contend
that, given the dearth of explanation F & R provide about mechanisms, it
can be just as intuitive to think that some actual-sequence and alternative-
sequence scenarios involve the same, not different, mechanisms.
3 The Test Cases
In this section I will develop a basic Frankfurt-style actual-sequence exam-

ple and then show three possible types of ways to get a different outcome
in an alternative-sequence case. We will want to treat these examples as test
cases for F & R’s thesis that MRR is sufficient for moral responsibility, and we
will want to find out what light these examples shed on F & R’s distinction
thesis. The basic story, which takes place in Germany during the 1930s, will
involve some setup.11 Here is the basic story:
Schmidt is a high-ranking official in the German government just prior
to Hitler’s rise to dictatorial power. Having been raised by parents with a
deep inner moral sense and resolve, Schmidt learned from an early age the
difference between right and wrong. As he developed, so did his moral
wisdom. Suffice it to say that Schmidt developed into a man of impeccable
moral judgment.
Owing to Schmidt’s high rank in the German government, he is a mem-
ber of a secret group that will decide by vote whether or not Hitler is to be
given supreme power. Schmidt has known Hitler for many years. The two
became friends when each was a budding young politician. During those
early years, they shared with each other their dreams and ideas about how
to build a better Germany. Schmidt was impressed with both the ingenuity
of Hitler’s ideas and the deep moral sense that those ideas displayed. From
the things that Hitler said to Schmidt during those early years, Schmidt had
very good reasons to think that Hitler would be an honest, forthright, and
morally upstanding asset to Germany’s future. However, several years before
the vote is to take place, Schmidt begins to hear disturbing stories about
158 T. R. Long
Hitler. From reliable sources, he hears that Hitler has intentions of building
up a master race of pure-blooded Aryans, driving non-Aryans out of the
country, and perhaps killing those who do not leave. Schmidt develops deep
doubts about Hitler’s moral character.
Schmidt’s role in the decision is to cast the deciding vote, should the
voting members reach a deadlock. The members do reach a deadlock, so
Schmidt is called upon to cast the deciding vote. Schmidt deliberates.
Waiting in the wings is Block, a malevolent being with extraordinary causal
and predictive powers. Should it become clear to him that Schmidt is going
to vote against Hitler, Block will take effective steps to ensure that Schmidt
votes for Hitler. Otherwise, Block will do nothing. So, whatever Schmidt’s
initial preferences and inclinations, Block will have his way.12
We can divide the basic possibilities here into two: Schmidt may deliber-
ate and vote for Hitler. Although this would be a surprising outcome given
Schmidt’s moral character and his serious misgivings about Hitler, this
would be a possible case under F & R’s account of MRR; for F & R want their
theory of moral responsibility to be compatible with cases of ‘weakness of
will’.13 Note that in this case, which I will deem the Block/Schmidt actual
sequence, Schmidt is both morally responsible and blameworthy.
The other basic possibility is that Schmidt is such that, were Block not to
intervene, he would vote against Hitler.14 Now let us consider what course of
action Block will take, if he believes that Schmidt will vote against Hitler, to
ensure that Schmidt acts as Block wants. There may be a multitude of actions
that Block could take; however, as far as I can tell, there are only three types
of action that Block could take. I will apply each of these three possible types
in the following three cases respectively, each of which will be an example of
what I will call an alternative sequence.
Block/Schmidt Case 1: A minute before Schmidt casts his vote, Block
becomes convinced that Schmidt is going to vote against Hitler.15 Block
goes into action: rather than monkeying with Schmidt’s normal delibera-
tive voting mechanism, Block bypasses it. Block manipulates Schmidt in
some way such that Schmidt’s voting for Hitler is not related in any way to
the reasons that Schmidt has for or against voting for Hitler. We might
imagine him doing something directly to Schmidt’s nervous system that
physically causes him to vote for Hitler.
This is, of course, the kind of action that F & R think would occur should
Block believe that he must act; for, in this alternative-sequence case, the
mechanism that issues in Schmidt’s voting for Hitler (i.e., the one that Block
uses) is a different mechanism than the one that would issue in Schmidt’s
voting against Hitler. In the actual sequence, Schmidt’s action is issued in by
a deliberative voting mechanism that is MRR. But in this alternative se-
quence, Schmidt’s action is ushered in by a different mechanism, a mecha-
nism that has nothing to do with reasons-responsiveness. As F & R say, we
would hold Schmidt neither morally responsible nor blameworthy for his
voting for Hitler in this case. Furthermore, note that F & R are committed to
thinking that this is the only kind of action that could occur if Block should
intervene; for, as we saw earlier, F & R’s view is that we cannot hold Schmidt
morally responsible in this case precisely because the mechanism that
operates is not the same as the one at play in the actual sequence (which
Schmidt has taken responsibility for).16 However, I see no reason to think
that this type of action is the only possible way for Block to accomplish
his work. Cases 2 and 3 demonstrate other types of action that could get the
job done.
Block/Schmidt case 2: A minute before Schmidt casts his vote, Block be-
comes convinced that Schmidt is going to vote against Hitler. Block goes into
action: he adds new inputs into the very same mechanism that is operant in
the actual sequence. Suppose that these new inputs come in the form of
reasons for voting for Hitler. Block directly feeds into Schmidt’s delibera-
tive voting mechanism enough reasons, or reasons powerful enough, to
ensure that Schmidt will vote for Hitler. Because Block has extraordinary
predictive powers, he knows which reasons will ensure that Schmidt’s
deliberative voting mechanism yields a vote for Hitler, and Block provides
those reasons as inputs for Schmidt’s deliberative voting mechanism.17
I have two things to discuss about this case. First, note that F & R do not
consider this case as a possible way for Block to accomplish his desire. Their
emphatic claim, that whenever there is intervention (in an alternative
sequence) the mechanism is a different one from the mechanism that
would have been employed had there been no intervention, commits them
to thinking that this case is impossible. However, I do not see that this case
is any less possible than the one that they prefer. Second, this situation
prompts the key question: is Schmidt morally responsible for voting
for Hitler? According to F & R’s account of MRR, the answer is “yes,” for
Schmidt deliberates appropriately in this alternative sequence case; that is,
as he comes to make his decision, Schmidt’s deliberative mechanism is
160 T. R. Long
regularly receptive to reasons, weakly reactive to reasons, and the mecha-

nism is his own. In short, Schmidt meets all the conditions of MRR. More-
over, the deliberative voting mechanism that Schmidt employs is the same
mechanism as the one operant in the actual sequence—that mechanism can
yield varying outcomes in response to different reasons. So, despite Block’s
intervention, Schmidt is morally responsible.
Now, one might worry about whether Schmidt is, in fact, morally
responsible in case 2. We may assume that, had Block not intervened,
Schmidt would have voted against Hitler. Even though Schmidt chooses via
his own deliberative voting mechanism, something seems wrong about
saying that he is morally responsible for voting for Hitler. After all, we
might think, if Block hadn’t monkeyed with the inputs (reasons) that
Schmidt had on his own, Schmidt would have voted against Hitler. Let’s
pause to consider why we might be inclined to think that Schmidt is not
morally responsible in this case. I think it is because we are inclined to
think that our principle MAN (above) is true: according to MAN, Schmidt
is not morally responsible, since unbeknownst to Schmidt, Block manipu-
lated Schmidt’s brain a moment before Schmidt acts; and had Block not
manipulated Schmidt’s brain, Schmidt would have voted against Hitler. It
is these facts that drive the intuitive sense that Schmidt is not morally
responsible.18 It seems that Block’s action goes so far as to amount to the
manipulation we had in case 1, even if the manipulation here occurs in a
different way than it did in case 1. So case 2 appears to be a counterexam-
ple to F & R’s claim that MRR is sufficient for moral responsibility.
Block/Schmidt case 3: A minute before Schmidt casts his vote, Block
becomes convinced that Schmidt is going to vote against Hitler. Block goes
into action: he removes inputs from the very same mechanism in Schmidt
that would have operated if Block had not intervened.19 The inputs
that Block removes from Schmidt’s deliberative voting mechanism are
Schmidt’s own reasons for voting against Hitler. Block knows which reasons
to remove in order to ensure that Schmidt’s deliberative voting mechanism
yields a vote for Hitler. Suppose that these reasons are in the form of
Schmidt’s negative memories of Hitler during the last few years (the ones
that point to Hitler’s diabolical intentions). The idea is that Block removes
only the relevant memories—he does not remove whatever counts as the
deliberative mechanism.20 So, when Schmidt deliberates during that
moment before he casts his vote, the only memories he has of Hitler are
from the early days when Hitler seemed to be a thoroughly good chap (per-
haps Block would also leave in Schmidt’s brain morally neutral memories
involving Schmidt’s associating with Hitler during the past few years).
Thus, the only reasons that Schmidt has to go on as he deliberates are
reasons to vote for Hitler.
Is Schmidt morally responsible for voting for Hitler in this case? Accord-
ing to F & R’s account of MRR, the answer is “yes,” for in coming to make
his decision, Schmidt’s mechanism meets all the conditions of MRR. More-
over, the very same deliberative voting mechanism is operant in this case
as it is in the actual sequence. That mechanism can yield varying outcomes
in response to different reasons. Schmidt exhibits guidance control in that
the mechanism that actually issues in his action is MRR. So, F & R are com-
mitted to thinking that Schmidt is morally responsible, despite Block’s
intervention.
Nevertheless, it may seem that Schmidt is not morally responsible in this
case. Although it is true that Schmidt’s mechanism is MRR, it seems plausi-
ble to suppose that Block has deprived Schmidt of some of his own reasons
in such a way that he is not morally responsible for voting for Hitler.21
Moreover, note that if we are inclined to think that Schmidt is not morally
responsible in this case, it is because we think that MAN (or something very
much like it) is true. On MAN, Schmidt is not morally responsible, since he
was not aware of Block’s manipulation a moment before Schmidt acted,
and he wouldn’t have acted as he did if Block hadn’t intervened.
4 The Options for Fischer and Ravizza
What resources do F & R have to deal with these last two test cases? The
answer will depend on whether they concede them to be counterexamples.
If they do, then I doubt that they can fix their theory without denying that
MRR is sufficient for moral responsibility. Here’s why: the only way around
my counterexamples seems to involve construing the relevant deliberative
mechanism broadly enough that it includes the inputs (i.e., the reasons) as
part of the mechanism. On this construal, F & R would be entitled to con-
clude that Schmidt is not morally responsible in cases 2 and 3, because
Schmidt has not taken responsibility for a mechanism that includes inputs
that Block has either directly added or excised. This reply may seem
promising, but it turns out to be deadly to their theory; for F & R want a
162 T. R. Long
general account that allows us to discern when an agent is morally respon-

sible. Thus, what they need are mechanisms that are able to yield varying
outputs (actions) depending on the inputs (reasons). But, on the present
proposal, which has it that the mechanism includes the inputs, what we get
in each case is an invariant mechanism. But if each mechanism is invari-
ant, then we don’t have what is needed for a general theory. To their credit,
F & R maintain that the mechanisms they are talking about can, in fact,
deliver varying outputs in response to various reasons. However, this claim
opens them to the kinds of counterexamples that I raise. Here is a more
general point: no matter how broadly one tries to construe the mechanism
(i.e., no matter how many inputs are included as part of the relevant mech-
anism), one must allow for inputs that are not included in the mechanism.
We can then develop a counterexample that will exploit those nonincluded
inputs. Hence, if they accept cases 2 and 3 as counterexamples, then I can-
not think of any way to save their theory without dropping the claim that
MRR is sufficient for guidance control and thus for moral responsibility. So,
one option in dealing with my counterexamples is for F & R to affirm MAN
(or something like it) and add conditions to MRR so that the theory gets
the cases right.
However, F & R need not accept my cases as actual counterexamples. They
have another solution open to them that will save their theory. Here is the
alternative solution: my cases are not counterexamples at all; Schmidt is
morally responsible in cases 2 and 3, but he is not blameworthy. Although it
is unfortunate that Schmidt’s action put Hitler in a position to carry out his
horrible, unspeakable acts, and although we may assume for the sake of the
thought experiments that had Schmidt voted against Hitler, Hitler would
not have carried out his reign of terror, nonetheless in each of the two cases
Schmidt did the best he could given the reasons he had. Assuming that he
had no special reason to doubt his own faculties, it is clear that in each case
Schmidt had very good reasons to vote for Hitler. Moreover, Schmidt voted
in consonance with his reasons. Given his reasons, it would have been irra-
tional for him to vote against Hitler. These facts show that Schmidt is
morally responsible for his actions.
So where do these considerations leave F & R? Either they can accept my
examples as counterexamples, in which case they will need to go back to
the drawing board in order to hash out the missing necessary condition(s)
of MRR, or they can deny that my examples are counterexamples. If the
latter, they will need to acknowledge a false assumption they seem to be

under about Frankfurt-style cases. To make this point perspicuous, I want
to draw out what I take my discussion to have shown about Frankfurt-style
cases. The fundamental thing that Frankfurt designed these cases to show
is that one can be morally responsible even if one couldn’t have done oth-
erwise. But the alternative-sequence cases are diverse. Using my characters
as examples, the leading idea of an alternative-sequence case is that every
possible world where Schmidt on his own does not do the thing that Block
wants is a world where Block intervenes such that Schmidt does what Block
wants. I have shown that we can divide the alternative-sequence cases into
two subcategories, one of which nobody seems to have noticed. There is
the kind of case in which Block intervenes in such a way that Schmidt is
not morally responsible for his action—that is, the case in which Block
does something to Schmidt that physically (or otherwise) directly causes
Schmidt to vote for Hitler. This is the kind of case that F & R and others
have countenanced. The other kind of case is one in which Block inter-
venes to bring it about that Schmidt does what Block wants, but the inter-
vention is such that Schmidt does not lack moral responsibility for his
action. My examples show not only that Block can manipulate in a way
that eliminates moral responsibility, but they also show that Block can
manipulate in a way that does not eliminate moral responsibility. Now, I
suppose that everyone will agree that Block could have successfully manip-
ulated Schmidt by more ordinary means—by providing him with effica-
cious newspapers, flyers, and other documents, or by arranging a large
enough bribe, or by some other means—and Schmidt would be morally
responsible for his actions in each of those cases. The key point I have
made is that the manipulative intervention that does not eliminate moral
responsibility can be more severe than we might have thought. We are now
in a position to state what I said above is a false assumption by F & R. They
seem to assume that in all alternative-sequence cases, the intervenor’s
manipulation eliminates moral responsibility on the part of the agent. My
examples show that this assumption is false.
There is another lesson to be drawn from my examples, which involve
manipulation that does not rule out moral responsibility. This kind of
manipulation can be so severe that it clashes with an intuition that many
people have. That intuition is the driving force behind MAN, which says
that if an agent is unknowingly manipulated in such a way that she performs
164 T. R. Long
an action that she wouldn’t have performed had there been no manipula-
tion, then she is not morally responsible for her action. Despite the fact
that MAN may seem intuitively right, further reflection reveals that there is
little to recommend it. Consider Block/Schmidt cases 2 and 3. In each case,
not only did Schmidt meet the conditions of MRR, but he also intuitively
did the morally responsible thing, given the evidence he had to go on. And
if we knew the relevant facts about each case, we might even praise him for
his action, even as we deeply regret or deplore the trick Block played on
him and the consequences that followed. This shows that MAN is false, for
Schmidt did perform an action at time tn only because another agent, act-
ing completely unbeknownst to Schmidt, directly manipulated Schmidt’s
brain earlier at t1; and yet Schmidt is morally responsible for voting for
Hitler, since he acted in consonance with his reasons in an appropriate
way. The moral is that we are morally responsible for our actions when they
are consonant with our reasons in an appropriate way (perhaps, when MRR
is satisfied), but whether we are praiseworthy or blameworthy for our
actions depends on additional facts about the cases. Counterexamples to
MAN are plentiful. Although the intuition behind MAN may be appealing,
it does not survive scrutiny.
After doing this ground clearing, we are in a position to see that F & R’s
theory does get the cases right. It implies that when the deliberative process
as a whole is directly manipulated the agent is not morally responsible for ac-
tions issued in by that process. This is exactly right. The problem is in F & R’s
assumption about how their theory applies to Frankfurt-style examples; for
it does not, nor need not, imply that all alternative-sequence cases eliminate
moral responsibility on the part of the agent.
5 Trouble with Intuitions
However, Fischer himself has responded to my examples by saying that

intuitively Schmidt is not morally responsible in cases 2 and 3. Fischer infers
from this intuition that the manipulation of the inputs makes the mecha-
nism in those cases a different mechanism from the one operant in the
actual-sequence case.22 I find this response disappointing, for Fischer con-
cedes that he has no principled way to distinguish the mechanisms in ques-
tion. If MRR really is sufficient for guidance control, and thus for moral
responsibility, as F & R claim that it is, then an objection to my examples
stands a chance only if we have reason to doubt that Schmidt meets the third
condition of MRR. However, I think it is clear that Schmidt can easily meet
that condition. In my examples I suppose that Schmidt (a) sees himself as the
source of his behavior, which follows from the operation of the relevant
mechanism, that Schmidt (b) believes that he is an apt candidate for the
reactive attitudes (in the manner F & R specify), and that Schmidt (c) views
himself as an agent based on his evidence for (a) and (b). Nothing about
Block’s manipulation prevents Schmidt from meeting those conditions. To
think that Schmidt fails to satisfy those conditions is to suppose that
Schmidt would have to be responsible both for his prior evidence (or beliefs)
and for the way in which he acquired that evidence (or those beliefs). But,
surely this is to add too much to the notion of taking responsibility. Evidence
(and reasons) supporting beliefs come to us from various directions and in
many different ways. Some of it we are quite aware of, and some of it we are
less aware of. Earlier I mentioned that nobody would balk at the suggestion
that Schmidt could have been convinced to vote for Hitler by more ordinary
external means (news reports, personal testimony, etc.). What is the relevant
difference between manipulating the inputs in those ways and manipulat-
ing the inputs directly?
Consider a case in which a group of conniving folk manipulate the inputs
of someone’s deliberative mechanism by affecting the information that per-
son gets. Suppose a group of Young Republicans who work at a news press
manipulate the inputs that Karen’s deliberative voting mechanism receives
by changing the text of Karen’s newspaper every time something good is said
in print about a Democratic candidate. Would this prevent Karen from being
morally responsible for her voting? If you are inclined to answer “yes,” then
you owe us an explanation of why any of us are morally responsible for our
acts of voting. After all, very few of us control the political reports or the po-
litical opinions that are propounded in our newspapers and other media;
thus, most of us do not control the inputs our deliberative voting mecha-
nisms receive. You can bet your bottom dollar that much of the political talk
in newspapers is written with the intent of manipulating us to vote the way
the authors want us to vote. On what basis should we judge ourselves as
morally responsible for our voting but deny that Karen is morally responsible
for her voting? My point is even more convincing when we turn our attention
to advertising. The very livelihoods of advertisers depend on their ability to
manipulate us into buying whatever marketers are selling. Advertising is a
166 T. R. Long
sneaky, subtle business, but are we to suppose that most of us are not morally
responsible for our purchases? Despite the cunning powers of advertisers,
surely most of us are not helpless victims of manipulation. Likewise for
Schmidt in cases 2 and 3; he was no helpless victim. No one caused him to act
by bypassing his deliberative voting mechanism. Schmidt actually weighed
reasons in a regularly reasons-responsive way, and his deliberative voting
mechanism was weakly reactive to reasons; moreover, that mechanism,
which we may assume Schmidt used many times before, was one that he had
clearly taken responsibility for. So, according to MRR, Schmidt is morally
responsible in cases 2 and 3.
F & R want their theory to capture our intuitions about moral responsi-
bility. I grant that judging Schmidt to be morally responsible strains the
prereflective intuition that some people might have about the cases. How-
ever, it is fairly common for us to find that our prereflective intuitions
about a matter are wrong, after reflection on the details; and, I have found
that, after reflection, many people (philosophers and nonphilosophers)
judge the cases as I do. Even so, I am not inclined to hang my case on any
claim about how many people have a particular intuition. Instead, I put
forward the following evidence for your consideration: (1) F & R want vari-
ant mechanisms; that is, they want any MRR mechanism to be able to
operate on a variety of inputs and to produce a variety of outputs; and (2)
F & R explicitly distinguish between moral responsibility, on the one hand,
and praiseworthiness and blameworthiness, on the other. Consideration of
(1) reveals that it will not do to build all the inputs into an MRR mecha-
nism, but the only way to make sense of Fischer’s response to my examples
is to think that Schmidt would have to be responsible for his input acquisi-
tion as well as for the mechanism itself. This amounts to building the
inputs into the mechanism. Consideration of (2) reveals that F & R have
the resources for their theory to get the cases right. Schmidt is morally
responsible for voting for Hitler, although he is not blameworthy. If you are
still unconvinced, suppose that Schmidt walked into the voting room and
murdered the officials who voted against Hitler. I think that we would all
judge him to be morally responsible for the murders, even if we knew every
detail about Block’s manipulation. But, then, what principled reason is
there to deny that Schmidt is morally responsible for his voting?
In summary, given that F & R explicitly distinguish between any MRR
mechanism and its inputs, given that they explicitly distinguish between
moral responsibility and praiseworthiness and blameworthiness, and given

that they have no principled reason to judge Schmidt not to be morally
responsible in cases 2 and 3, I think it is clearly more reasonable to judge
Schmidt as morally responsible in those cases; furthermore, I think that
this judgment is correct.
6 Conclusion
F & R are aware that, in failing to provide effective differentiation parame-

ters for mechanisms, they open their theory to philosophical worries:
Clearly, on the approach to moral responsibility we have begun to sketch, it is
crucial to distinguish between the kind of mechanism that operates in the actual
sequence and the kind of mechanism that operates in the alternative sequence (or
sequences). We must confess that we do not have any general way of specifying
when two kinds of mechanisms are the same. This is a potential problem for our
approach; it will have to be considered carefully by the reader. (Fischer and Ravizza
1998, 40)
After carefully considering the point, I conclude that Block/Schmidt cases

2 and 3 reveal an actual problem for F & R’s characterization of their theory
with respect to Frankfurt-style examples. As I have shown, there may be no
distinction between the mechanism that operates in the actual sequence
and the mechanism that operates in the alternative sequence (or sequences).
The problem, however, is not insurmountable, although my examples show
that a reasonable solution requires F & R to give something up: either they
must give up the claim that MRR is sufficient for moral responsibility, or they
must give up their claim that MRR explains why all Frankfurt-style alterna-
tive-sequence cases yield agents who are not morally responsible for their ac-
tions. As I have shown, alternative-sequence cases need not yield agents who
lack moral responsibility for their actions. MRR does rule out moral resp-
onsibility when manipulation is pervasive across the whole deliberative
process, but it does not rule out moral responsibility when manipulation
touches only the inputs to the deliberative process (even though the manip-
ulation may be so severe that it conflicts with MAN). Since I think there is
something right about connecting moral responsibility to reasons-respon-
siveness, I think it best to take the second horn of the disjunction. Not only
is this solution plausible, but it also keeps F & R’s theory of moral responsi-
bility intact.
168 T. R. Long
Acknowledgments
Special thanks to Rich Feldman for many fruitful discussions about this
paper. Thanks also to John Martin Fischer, Tom Senor, Michael McKenna,
Joseph Keim Campbell, Allen Plug, Greg Wheeler, Tony Roark, Ken Himma,
and an anonymous referee for this volume, as well as participants in the
Rochester Philosophy Society (2000), the 52nd Northwest Conference on
Philosophy (2000), the American Catholic Philosophical Association,
Western New York (2000), the Society of Christian Philosophers, Eastern
Region (2001), the Philosophy Writing Seminar at the University of
Rochester (2001), and the Inland Northwest Philosophy Conference on
Freedom and Determinism (2001), for helpful comments on this paper.
Notes
1. For some of the works that have influenced my thinking, see Della Rocca (1998),
Mele and Robb (1998), Widerker (1995a, 1995b, 2000), and Vihvelin (1998).
2. See Fischer and Ravizza (1998, chaps. 1–2). See also Fischer (1995, chaps. 7–8).
3. I thank Michael McKenna and Joseph Keim Campbell for bringing to my atten-
tion that F & R also mention an epistemic condition as necessary for moral respon-
sibility. Though providing no analysis of the condition, F & R say that it is meant to
capture the “Aristotelian intuition” that “in order to be praiseworthy or blamewor-
thy, a person must know (or be reasonably expected to know) what he is doing”
(Fischer and Ravizza 1998, 12). To use their example, if you were to run over a kitten
hiding under your car but were ignorant of the kitten’s location, then, intuitively,
you would not be morally responsible for its death. After noting this epistemic
condition, F & R don’t mention it again but rather assume that it is met both in the
examples that they discuss and in their claim that guidance control is sufficient
for moral responsibility. In my cases 2–3 from section 3 below, I also assume that
F & R’s epistemic condition is met. It is possible that some epistemic condition
would make trouble for my test cases, but I find nothing in F & R’s discussion of their
epistemic condition that implies that there is trouble for my cases; moreover, I doubt
that there is a plausible, fully worked-out epistemic condition that would make
trouble for my test cases.
4. Here I summarize Fischer’s discussion of reasons-responsiveness (Fischer 1995,

161–164). See also Fischer and Ravizza (1998, 41–46).
5. See Fischer and Ravizza (1998, chap. 3). I will use MRR to stand for both ‘moder-
ate reasons-responsiveness’ and its cognate form, ‘moderately reasons-responsive’.
6. Although this schema does not appear in Fischer and Ravizza (1998), I have done
my best to reproduce faithfully the conditions of MRR as F & R present them. The
schema contains quotations or near quotations of various claims made by F & R that
are meant to explain what they summarize as follows: moral responsibility “requires
that an agent act on a mechanism that is regularly receptive to reasons, some of which
are moral reasons, and at least weakly reactive to reasons” (Fischer and Ravizza 1998,
82), and also that the agent takes responsibility for the mechanism (Fischer and Ravizza
1998, 242–243). Strictly speaking, on F & R’s theory, condition (3) is not a condition
of moderate reasons-responsiveness but is rather an additional necessary condition
for guidance control. However, for ease of presentation I speak of MRR as encompass-
ing all three conditions. This does not affect F & R’s theory in any relevant way.
7. F & R make this clear in numerous passages in Fischer and Ravizza (1998). Here is
a typical one: “Return to our account of moral responsibility for actions. On this
account, an agent is morally responsible for an action insofar as he has guidance
control of it, where guidance control consists in the action’s issuing from the agent’s
own, moderately reasons-responsive mechanism” (Fischer and Ravizza 1998, 230).
As I mention in note 4, F & R make such claims on the assumption that their
epistemic condition is met.
8. I use Fischer’s earlier account here because of the detail. Here is the parallel
account by F & R: “. . . in a Frankfurt-type case . . . the kind of mechanism that actually
operates is reasons-responsive, even though the kind of mechanism that would
operate—that is, that does operate in the alternative scenario—is not reasons-
responsive. . . . But in the alternative scenario, a different kind of mechanism would
have operated—one involving direct electronic stimulation of [the agent’s] brain. . . .
And this alternative sequence mechanism is not reasons-responsive” (Fischer and
Ravizza 1998, 38–39).
9. I thank Tom Senor for calling to my attention, though not necessarily endorsing,
a version of MAN.
10. F & R repeatedly say the following sort of thing: “the actual-sequence mecha-
nism is, intuitively, the agent’s own, whereas the alternative-scenario mechanism is
not” (Fischer and Ravizza 1998, 228).
11. This story is fictional, but it is supposed to express the character of the historical
Hitler.
12. Readers may note that Schmidt and Block are counterparts to Frankfurt’s Jones
and Black, respectively. See Frankfurt (1997, 162).
13. Condition (2) of MRR allows for cases of ‘weakness of will’. See Fischer and
Ravizza (1998, 42–44). In Fischer (1995), Fischer also wanted his theory to be com-
patible with cases of ‘actual irrationality’ (166–168), but it is unclear whether F & R
maintain this claim in Fischer and Ravizza (1998).
170 T. R. Long
14. I put this as I do for two reasons (and I thank Michael McKenna for bringing this
issue to my attention). First, I want to avoid the kind of criticism that David
Widerker and others have brought against designing Frankfurt-type examples that
both grant libertarian-free action and rule out alternative possibilities. Such exam-
ples would have it that the agent at least starts to choose before the intervenor inter-
venes. The concern is that if some kind of nondetermined act is allowed in the setup
of a Frankfurt-type example, then there is no reason to think that the agent
(Schmidt, in my cases) cannot do otherwise, regardless of the actions of the inter-
venor (Block, in my cases). So, as I characterize my cases, I don’t assume libertarian-
free action on the part of Schmidt. This leads to my second point: F & R do not
assume libertarian-free action on the part of agents in Frankfurt-type examples
either. Their view is that such examples show that alternative possibilities are not
required for moral responsibility even if determinism is true. My goal is to follow their
lead to see where it gets us. If it turns out that there is some general problem along
these lines, then my cases might be problematic as well. However, I think I avoid
problems in this neighborhood by setting up the cases as I do; for, all I need is for it
to be the case that every possible world in which Schmidt does not do what Block
wants is a world in which Block intervenes. I need not tell a story about how Block
does this. If someone demanded such a story, then it seems to me that luck on
Block’s part could account for it.
15. Although in all three of my alternative sequence stories Block goes into action
one minute before he thinks Schmidt will cast his vote, nothing significant turns on
Block’s timing.
16. F & R also consider a case in which scientists don’t directly cause the implantee
to perform an action but rather the scientists implant a mechanism that is itself
MRR, which issues in an action. Although this example is closer to my cases 2 and 3,
it is still a case 1 type in virtue of the fact that the mechanism itself is implanted.
17. Of course, in order to avoid psychic shock or noticeable incongruence, Block

may have to inject more than thoughts into Schmidt’s deliberative voting mecha-
nism. I just intend the story to be one in which Block adds in addition to reasons
whatever else is minimally necessary to ensure a vote for Hitler.
18. I am claiming neither that F & R explicitly endorse MAN nor that they are com-
mitted to it; rather, the point of MAN is to provide something other than mere
intuition to explain why it may seem to us that Schmidt is not morally responsible
in this case.
19. This possibility was suggested to me by Allen Plug in conversation.
20. Block may, of course, have to remove more than just some of Schmidt’s memo-
ries in order to avoid psychic weirdness in Schmidt, but the story is just that Schmidt
removes whatever in addition to the relevant memories is minimally necessary in
order to ensure a vote for Hitler.
21. In order to motivate the worry that Schmidt is not morally responsible in this
case, consider that it works as an amnesia case might. Consider Chuck, who has lost
all memories of the last five years. Chuck has been married for five years, and his
wedding anniversary was yesterday. Although Chuck knows the woman who is his
wife (for he remembers having dated her), he did not know—indeed, could not have
known—that yesterday was his wedding anniversary. So Chuck not only failed to
buy his wife anything for their wedding anniversary, but he also failed to acknowl-
edge their wedding anniversary at all. Assuming that Chuck had not been told earlier
about his impending wedding anniversary, it seems clear that no one, not even
Chuck’s wife, ought to hold him responsible for not doing something special for his
wife on their wedding anniversary (even if she is a bit disappointed). Although
Chuck’s relevant mechanism is MRR, we might think that he is not morally respon-
sible for his lack of appropriate celebratory action, precisely because his amnesia has
deprived him of the use of reasons that he once had. The idea here is that he is in no
way responsible for the loss of those reasons, and so he is not morally responsible for
failing to act appropriately. Compare case 3 to the reasoning in Chuck’s case:
Schmidt is in no way responsible for the loss of his reasons to vote against Hitler, and
so he is not morally responsible for voting for Hitler. However, as I argue in the main
text, Chuck may be morally responsible for whatever actions he does perform, so
long as he acts in an appropriate reasons-responsive way; just note that if he mur-
dered his wife, he would be morally responsible. So long as we are careful to distin-
guish moral responsibility from praiseworthiness and blameworthiness, we should
have no trouble seeing that Chuck is morally responsible for his actions, even
though he is not blameworthy for failing to do anything special for his wedding
anniversary.
22. Fischer made these claims during the question and answer session of my pre-
sentation of a version of this paper at the Inland Northwest Philosophy Conference
on Freedom and Determinism, University of Idaho and Washington State University,
April 2001.
References
Della Rocca, M. 1998. “Frankfurt, Fischer, and Flickers.” Noûs 32: 99–105.
Fischer, J. 1995. The Metaphysics of Free Will: An Essay on Control. Oxford: Blackwell.
Fischer, J. and Ravizza, M. 1998. Responsibility and Control: A Theory of Moral

Responsibility. Cambridge: Cambridge University Press.
Frankfurt, H. 1997. “Alternate Possibilities and Moral Responsibility.” In D.

Pereboom, ed., Free Will. Indianapolis, Ind.: Hackett.
Ginet, C. 1998. “Comments on John Fischer, The Metaphysics of Free Will.” Journal of
Social Philosophy 29: 126–134.
172 T. R. Long
Mele, A. and D. Robb. 1998. “Rescuing Frankfurt-style Cases.” Philosophical Review

107: 97–112.
Widerker, D. 2000. “Frankfurt’s Attack on Alternative Possibilities.” Philosophical

Perspectives 14: 181–201.
. 1995a. “Libertarian Freedom and the Avoidability of Decisions.” Faith and

Philosophy 12, no. 1 (1995): 113–118.
. 1995b. “Libertarianism and Frankfurt’s Attack on the Principle of Alternative

Possibilities.” Philosophical Review 104: 247–261.
Vihvelin, K. 1998. “John Martin Fischer, The Metaphysics of Free Will.” Noûs 32:
406–440.
8 Which Autonomy?
Nomy Arpaly
“When I make a word do a lot of work like that,” says Humpty Dumpty,
“I always pay it extra.” Some words in philosophy have been working over-
time, as, for example, anyone knows who has attempted to explain to a
bewildered student the relations between various ‘externalisms’ and ‘in-
ternalisms’. It has to be admitted, though, that the multitude of meanings
that ‘internalism’ has in our field is not a real problem. Real problems begin
when a term of art performs so many tasks that it becomes at least as
elusive and complex as the natural language terms it was supposed to help
clarify. I suspect that this is the case with the term ‘autonomy’. In philo-
sophical discussions of moral responsibility, attention has shifted some-
what from questions of praise and blame, responsibility and lack thereof, to
questions about autonomy: what it is, whether it exists, and so forth.
I would like to point out that there are at least eight things that we some-
times call ‘autonomy’, and there is a need to tell them apart. In discussion,
we sometimes act as if these eight things were one, or, more often, act as if
intuitions concerning any one of them should be equally relevant to the
agency theorist who is trying to figure out what true autonomy consists in.
The discussion in this paper is meant to raise questions about the general
usefulness of the concept of autonomy to our theorizing about moral
responsibility—to make us wonder if a moral responsibility theorist would
not be better off going back to investigating praiseworthiness and blame-
worthiness directly, ignoring intuitions regarding autonomy. (I do not
intend to criticize or question the use of ‘autonomy’ in political philoso-
phy, normative ethics, or any other context except the context of moral
responsibility and descriptive agency theory.)
Let us look at the various things that ‘autonomy’ means. One of these
things is what I will call, for convenience, agent autonomy. Agent autonomy
174 N. Arpaly
is a relationship between an agent and her motivational states which can

be roughly characterized as the agent’s ability to decide which of them to
follow: it is a type of self-control or self-government that persons (usually)
have and that nonhuman animals do not have. This, of course, is a very
broad construal of agent autonomy—different philosophers can agree that
we, unlike other animals, ‘govern’ ourselves while having very different
views as to what constitutes this governing and as to which parts of a
person, if any, constitute the ‘governing’ self and the ‘governed’ self (i.e,
a philosopher may think that our autonomy is a matter of our having
second-order desires or that it is a matter of our having the ability to
reflectively endorse our actions). Some idea of agent autonomy is at least
implicit in any Kantian account of moral psychology, and has been at the
center of the work of countless action and agency theorists of many
stripes—some clear examples being early Harry Frankfurt, Gary Watson,
David Velleman, Robert Noggle, Michael Bratman, and Keith Lehrer.
Occasionally, authors use ‘autonomy’ to mean simply ‘agency’, and
‘autonomous action’ to mean simply ‘action’ (or ‘full-fledged action’). This
is not the way I propose to use the term ‘agent autonomy’. The way I use
the term, not every theory of action is also a theory of agent autonomy nor does
every theory of action include a notion of agent autonomy. Take, for ex-
ample, Davidson’s theory of action, according to which a movement is an
action if it is caused, in a specific way, by a combination of a desire and a
belief. If my glancing at my watch is caused by a desire to know the time
and a belief that looking at my watch will tell me the time, it is thereby an
action. This theory of action stipulates that a person’s movement is her
action as long as a certain relationship obtains between the movement and
her mental states. Davidson’s theory of action does not go further and
require that some relationship obtain between the agent and the mental
events that cause her action. For my movement to be an action, it
is enough, for Davidson, that it is controlled in a certain way by
mental events that happen in my brain: there is no extra sense in which
the relevant mental events need to be controlled by me, the agent. All of the
desires that happen in me are my desires and can, with the help of beliefs,
produce actions: none of these desires is more “truly my own” than others.
While Davidson does, implicitly, have a notion of self-control—after all, he
talks about akratic actions versus nonakratic actions—there is no direct way
in which control over self (as opposed to control of desire and belief over
Which Autonomy? 175
movement) enters his definition of action. Akratic actions are irrational,

but they are still actions, still my own, still performed for a reason. Thus
Davidson’s theory of action is, in my terms, not a theory of agent auton-
omy, but a theory of action in which agent autonomy plays no role (as op-
posed to theories of action, like Velleman’s, in which it does play a role). It
is also possible for a philosopher to have a theory of agent autonomy that
is not identical to her theory of action—that is, a theory according to
which the phrase ‘autonomous action’ is meaningful but contains no re-
dundancy, as ‘nonautonomous action’ is also an action. I say more about
this possibility elsewhere (Arpaly forthcoming).
Next on the list is the quite distinct sense of ‘autonomy’ that designates
the absence of various kinds of dependence on other people—autonomy as
self-efficacy, or having the ability to get along well in the world without
requiring the help of others. In this sense of ‘autonomy’, a person can
make herself more autonomous by learning how to drive, becoming rich,
becoming knowledgeable, or gaining physical strength. This sense of the
word ‘autonomy’ should not be confused with agent autonomy or self-
control because, after all, a rich or educated agent is no more agent auto-
nomous than a poor or uneducated agent: agent autonomy, like Sartrean
freedom, is the kind of thing which the slave in chains has just as much as
her master.
Often described as lacking autonomy are those people who seem to lack
not efficacy, but various kinds of independence of mind: servile and submissive
people (Thomas Hill’s Deferential Wife in Hill 1991 is a classical example),
and people who blindly accept the views of their gurus or their ‘communities’.
The psychological situations of these persons are often undesirable, and in
many cases they involve irrationality or vice. However, they are not always
situations in which agent autonomy is lacking. In an old Eastern European
joke, a man is urged to stop obeying his wife in every matter, and to start fol-
lowing his own free will, whereupon he replies that he always follows his own
judgment, as freely as anyone, and it so happens that his own judgment tells
him to ask his wife what to do and follow her advice. This figure of fun may
be perfectly autonomous as an agent, perfectly self-controlled even though
his psychotherapist is likely to tell him that, in a different sense, he needs to
increase his autonomy. The feminist who argues that women should be made
more autonomous and who then cites early Frankfurt’s or Bratman’s defini-
tions of autonomy, or alternately faults Frankfurt’s definition of autonomy
176 N. Arpaly
for not explaining why deferential wives are nonautonmous, is most likely
making a sort of category mistake.
An important concept of autonomy that I have not yet mentioned is the
concept of normative, moral autonomy—the one invoked when people ask
to be allowed to make their own decisions and to be free from paternalis-
tic intervention. There are well-known Kantian arguments for a connec-
tion between moral autonomy and agent autonomy, and there is much to
be said about them (I will return to some of them later), but I would like to
point out that the word ‘autonomy’ in ‘respect for autonomy’—an almost
pretheoretical notion in the United States of America—does not simply
mean ‘agent autonomy’. To see this, it suffices to look at a paradigmatic
case of ‘autonomy violation’ in the normative sense. If someone steals
from you, that person violates your autonomy in the normative sense, but,
plausibly, you are not any less of an autonomous agent as a result. Or
imagine that a doctor lies to a patient or withholds information from a
patient about his condition, ignoring the fact that the patient is in the
process of making an important employment decision to which his condi-
tion would be relevant. Many people would say, in this case, that the doc-
tor made it impossible for the patient to make an autonomous decision.
This is, however, not the sense of ‘autonomy’ that autonomous-action the-
orists are looking for. Suppose that the patient did, based on the false or
incomplete information received from his doctor, make an employment
decision. Are we, as agency theorists, to say that his decision was not agent
autonomous? If we were to say that, we would have to hold that any deci-
sion based on mistaken beliefs or incomplete information is a nonau-
tonomous decision. Given that people err frequently and have very
incomplete information about the world, or the next move one’s opponent
plans to make in chess, for that matter, that would again make autonomy
too scarce a commodity, especially if one intends to make it the basis of
I have spoken of autonomy as self-control, autonomy as material and
psychological independence, and normative autonomy. David Velleman
makes an additional distinction between autonomy, which is some form of
self-control, and authenticity, or being true to oneself (Velleman 2002). His
distinction is driven by a reading of Frankfurt’s latest view of love, caring,
and volitional necessity. Until about 1987, Frankfurt’s key examples of
nonautonomous agents were examples of people whom we would normally
Which Autonomy? 177
regard as lacking in self-control or experiencing a lack of it—recall the

unwilling addict, the man who is subject to attacks of rage, and other
such weak-willed people. But as of “Rationality and the Unthinkable,”
Frankfurt insists that some people are autonomous whom we would ordi-
narily regard as weak-willed (Frankfurt 1988). For example, he talks about
people who resolve reflectively to press what they believe to be the button
that would facilitate the dropping of an atomic bomb, but who find that
they “just can’t do it,” and also of a man who decides to spy on his bride-to-
be and finds himself forced to stop because the idea revolts him too much.
These people are not paradigmatic cases of self-control, but Frankfurt sees
them not as weak-willed, but as constrained by their own deepest values,
their volitional essences. Whatever these people may have consciously
resolved, the values which stop them from dropping the bomb or snooping
are truly their own—otherwise they would not have found themselves
repulsed by their respective tasks. These agents are prevented from acting by
what they care about—which, according to Frankfurt, makes them what
they are. Velleman maintains that this is not truly a theory of autonomy, but
is rather a theory of authenticity, or personal identity, or the real self. Here is
what Velleman says:
Even if I believed that a person had a motivational essence of this kind, I would not infer
that his being governed by this essence was what made him autonomous. Being gov-
erned by such an essence might amount to authenticity, perhaps, but not autonomy.
To see the difference, consider the paradigm case of inauthenticity, the person who
manifests what D. W. Winnicott called a “False Self.” This person laughs at what he
thinks he is supposed to find amusing, shows concern for what he thinks he is sup-
posed to care about, and in general conforms himself to the demands and expecta-
tions of others. The motives that his behavior is designed to simulate are motives
that he doesn’t genuinely have. . . . But is this person lacking in self-control, self-
governance, or autonomy? To be sure, he has a problem with autonomy, but his
problem is one of excess: he is overly self-controlled, overly deliberate; his grip on
the reins of his behavior is too tight, not too loose. His failure to be motivated from
within his true self makes him inauthentic, but it seems to result from his being all
too autonomous. (Velleman 2002, 97)
Velleman draws a distinction between authenticity, which he takes to be

the same as being true to one’s values, and autonomy—self-control or self-
governance. He reminds us that there are some people who have a
strong measure of autonomy, are in fact known for their exemplary self-
governance, but whose very self-control allows them to live out of touch
178 N. Arpaly
with what one would be tempted to call their ‘real selves’. We can imag-
ine a person who, like the young E. T. A. Hoffman, represses his passion
for literature due to his conviction that he should become “a respectable
citizen.” This person may be very autonomous, but it is natural to say that
he is out of touch with his real self. On the other hand, Harry Frankfurt’s
agents, who surrender to ‘volitional necessity’, may be very true to them-
selves, but they do not control themselves—it may be said that, on the
contrary, their selves control them, as it were. This is not to suggest, by any
stretch of the imagination, the romantic thesis that the autonomous and
the authentic are always in conflict, but only to point out that sometimes
they are, and that therefore they are not one and the same.
Our talk of people being more ‘themselves’ or less so is not all of a piece,
either. Peter Strawson, in “Freedom and Resentment,” discusses contexts
in which we make statements such as “you must excuse her, she is not
herself—she has been under stress” (Strawson 1962). This is a statement
about what Velleman calls ‘authenticity’—the person’s action is claimed to
be less blameworthy because, presumably, it does not reflect her true will
but only the stress she has endured. Quite different is the kind of talk
in which a person reports that she experiences her feelings or desires as
alien, or feels ‘possessed’ or ‘violated’ by them or passive towards them, like
Frankfurt’s enraged man1 who feels as if his anger is an alien intruder and
thus not truly his own. It is the latter kind of talk that gave birth to the idea
of external desires, and while many accounts of external desires have been
given in terms of the structure of the agent’s will, the pretheoretical expe-
rience of ‘passivity’ toward one’s desires often serves as the intuition against
which these accounts are often measured. For example, the fact that some
of us do not experience our vices as alien led Frankfurt to refine his account
of externality so as to allow some unwanted traits (vices) not to be external,
and the fact that some thoughts are experienced as alien has led him to sug-
gest that there are external thoughts as well as external desires—thoughts
toward which we feel ‘passive’ (Frankfurt 1988). Because autonomy is some-
times perceived as the absence of externality, its phenomenological
face, the experience of passivity and alienation with respect to desires, is
often assumed to be relevant to autonomy.
The idea of external desires can be quite strange. If a Victorian lady
experiences her sexual desires as alien, intrusive, ‘not truly her own’, our
natural reaction is to tell her that she is wrong, that these desires are in fact
Which Autonomy? 179
her own, and only the false, asexual self-image that she acquired with her
upbringing makes her experience them as threatening to her integrity as a
person. We would think that she denies her real self. A person who first dis-
covers the presence of homosexual desires in himself is also likely to expe-
rience his desires as alien and he, too, would be urged to ‘accept himself for
who he is’. Similarly, the man whom Frankfurt used as his key example of
a person possessed by an external passion—namely, the man who feels pos-
sessed or violated by his outburst of rage—need not be the same person to
whom we would be inclined to give the Strawsonian excuse of ‘not being
himself’—he may in fact be revealing what we would call his true colors in
his outbursts, however honestly he experiences his anger as ‘not himself’.
It is perfectly possible for a person to wonder what came over him while his
friends tell each other ‘he is very much himself today’. In other words, ex-
periences of ‘being possessed’ are not always an indication that the desire
in question is inauthentic. Nor are they always an indication of lack of
agent autonomy or self control. Some addicts in religious twelve-step pro-
grams experience their sobriety as the will of God taking over, and their
akratic, ego-distonic falling off the wagon as a manifestation of their will-
fulness. Or consider a person who experiences herself primarily as weak-
willed, spineless, lazy, and passive. When this person, because of her lack of
self-control, takes a course of action or inaction which she regards as utterly
impermissible, she does not feel invaded or possessed at all. Instead, she
sadly thinks: Here I go again. I hate myself.2 Imagine now that suddenly, in
an emergency, our lazy person begins to act with admirable self-control. As
her desires obey her reason and her actions follow her decisions, for a
change, she may feel, paradoxically enough, as if somebody—somebody
cool and competent—has taken over her.
So far we have seen autonomy as self-control, autonomy as having power
in the world, autonomy as psychological independence, autonomy as hav-
ing moral rights, and autonomy as authenticity. The foregoing discussion
of the external desires literature shows that we need to introduce another
figure who, until now, has been lurking in the background of our discourse
on autonomy: the figure of the autonomous person as the self-identified
person, the person with a harmonious and coherent self-image who never
experiences her desires as an external threat.3
Also lurking in this background are concepts of what may be called heroic
autonomy, namely, ideals such as Spinoza’s freedom, stoic Apathia’s or
180 N. Arpaly
Ataraxia’s, Aristotle’s life of contemplation, Freud’s or Jung’s ideals of the

liberation that their methods were to aim at bringing, Nietzsche’s ideal of
the free spirit, the condition of the person who may truly state “I am the
master of my fate/I am the captain of my soul,” and other states which
are supposed to be desirable and only attainable by the very few. Some
notions of heroically autonomous persons may be related to some of the
autonomies mentioned before—that is, one may have an ideal of a heroic
authenticity, heroic self-efficacy, perfect inner harmony which leaves no
room for externality, or an ideal combining all of the above. One may even
imagine an übermench whose superhuman power consists of a truly in-
credible degree of agent autonomy—a person who always follows his inner
party line, always does what he reflectively endorses or what his higher-
order desires tell him, and so forth. However, agent autonomy per se, or at
least a substantial degree of it, is supposed to be the default condition of
the average adult human being, not merely of the übermensch. It is espe-
cially true that those who take agent autonomy to be a necessary condition
for moral responsibility cannot mean by agent autonomy any condition
that is too much of an ideal and too little of a human reality, unless they
wish to imply that very few people are morally responsible. Similarly,
those who take agent autonomy to be a necessary condition for action do
not wish to imply that we rarely ever act. Thus, caution must be exercised
when mixing intuitions about autonomy as an ideal condition for which
humans aspire and intuitions about autonomy as a property that a great
majority of humans actually possess to a large extent and most of the time.
I have left for last one more thing that has often been associated and im-
plicitly identified with the ability to respond to reasons—reasons in general
and moral reasons in particular. Autonomy has also been associated with
acting for good reasons—that is, with acting rationally. While it is not often
argued, since Kant, that autonomy is the same as rationality or reason-
responsiveness (although David Copp argues for autonomy being identical
with rationality in Copp forthcoming), intuitions about autonomy and the
lack thereof are often influenced by intuitions about rationality or reason-
responsiveness and the lack thereof, ‘rational autonomy’ being a common
expression and ‘irrational autonomy’ not being one. “When we act ratio-
nally,” says Elizabeth Anderson casually, “we act for reasons that we
endorse. We do not follow blindly where our desires suggest. This is the
autonomy of practical reason” (Anderson 1997, 92). On such a view, rational
Which Autonomy? 181
action is always agent autonomous action—is always an action in which

the agent, whoever she is, exercises control over her desires. Acting for rea-
sons we endorse is also many people’s idea of autonomy, and acting in
ways that diverge too much from what we endorse is often regarded as
nonautonomous. Rational and reason-responsive action is thought of as
requiring deliberation or endorsement, and autonomous actions are also
often regarded as the result of deliberation or endorsement. (Even some
very un-Kantian thinkers connect deliberation or endorsement to auton-
omy. See, for example, Mele 1995, who argues that being an adequate de-
liberator is a necessary condition for being an autonomous agent.) Things
that impair our ability to respond to reasons, such as alcohol and hypnosis,
are supposed to impair our autonomy. Animals are often said not to re-
spond to reasons, and they are often said not to be autonomous.
Identifying reason-responsiveness with autonomy may seem to be a rea-
sonable move if one thinks of acting for reasons as always being the result of
agent endorsement, or at least as accompanied by it. Elsewhere, I have
argued in various ways that acting for reasons as well as acting for good
reasons (acting rationally) are not necessarily related to agent endorsement,
self-control in the ordinary sense, or to deliberation: there are instances of
one without the other (Arpaly 2000). Related, though different, arguments
have been made in various talks by Thomas Scanlon and Peter Railton, and
also by the neuroscientist Antonio Damasio in Descartes’ Error (1994), and by
Robert Audi (1990).4 Harry Frankfurt in “Rationality and the Unthinkable”
tells that whereas a person who resolves to destroy the earth but cannot
bring himself to do is somewhat irrational, a person who decides to destroy
the earth and manages to follow through with his decision is truly insane—
thus opening the possibility that the ‘failure of self-control’ experienced
by the former may in fact be a triumph of his common sense (Frankfurt
1988). If any of these different arguments has merit, it seems less obvious
that intuitions about self-control and agent endorsement belong with intu-
itions about rationality and reasons-responsiveness as providing the prethe-
oretical grounding for one coherent concept called ‘autonomy’. Of course,
one would still be tempted to identify them if one assumes that whatever
moral responsibility depends on is called ‘autonomy’ and if one surmises,
correctly, that one cannot be morally responsible unless one acts for reasons.
So far, we have seen eight meanings ‘autonomy’ can have, or at least
eight things that the word ‘autonomy’ connotes in a way that invokes some
182 N. Arpaly
pretheoretical intuitions. Whenever we ask ‘is so-and-so autonomous’

instead of ‘is so-and-so morally responsible’ we expose ourselves to a ten-
dency to equivocate some or all of these eight meanings of ‘autonomy’, or
at the very least, to be unduly influenced by intuitions about irrelevant
senses of ‘autonomy’. Thus, phrasing questions in terms of the agent’s
autonomy instead of in terms of her moral responsibility serves not to
clarify, but to further obscure, the already difficult question of her moral
responsibility.
Let us look at an example. It involves one kind of use, in the moral re-
sponsibility literature, of stories involving people who have acquired their
current desires, beliefs, and values via being brainwashed. The premise,
assumed to be obviously true, that these people lack autonomy, has been
used in arguments for certain theories of responsibility such as that of
Alfred Mele, or the theory of John Martin Fischer and Mark Ravizza.5 In his
Autonomous Agents, Mele brings up the issue with an entertaining tale of
two agents:
Ann is an autonomous agent and an exceptionally industrious philosopher. She puts
in 12 solid hours a day, seven days a week; and she enjoys almost every minute of it.
Beth, an equally talented colleague, values a great many things above philosophy,
for reasons that she has refined and endorsed on the basis of careful critical reflection
over many years. She identifies with and enjoys her own way of life—one which, she
is confident, has a breadth, depth and richness that long days in the office would
destroy. Their dean, who shall remain nameless, wants Beth to be like Ann. Normal
modes of persuasion having failed, he decides to circumvent Beth’s agency. Without
the knowledge of either philosopher, he hires a team of psychologists to determine
what makes Ann tick and a team of new-wave brainwashers to make Beth like Ann.
The psychologists decide that Ann’s peculiar hierarchy of values accounts for her
productivity and the brainwashers instill the same hierarchy in Beth while eradicat-
ing all competing values—via new-wave brainwashing, of course. Beth is now, in the
relevant respect, a ‘psychological twin’ of Ann. She is an industrious philosopher. . . .
When she carefully reflects on her preferences and values, Beth finds that they fully
support a life dedicated to philosophical work, and she wholeheartedly embraces
such a life and the collection of values that supports it. (Mele 1995, 145)
It is often taken for granted that someone like Beth is not an autonomous
agent—after all, Beth has been brainwashed by the Dean Who Shall Not be
Named. The reaction that we have to Beth’s story, along the lines of ‘surely
she cannot be autonomous if her state of mind has been compelled’, often
serves as an intuition against which to check theories. It is used as such by
proponents of historical views of autonomy, such as Fischer and Ravizza,
Which Autonomy? 183
and Mele himself is eventually moved by it to include in his own theory of

autonomy the condition that the agent have no ‘compelled’ or coerced
mental states which the agent did not arrange to have produced in her
(Mele 1995, 187). But consider the following puzzle. There are cases of pro-
found conversions in a person’s values that are just as inexplicable to their
possessors, just as uninvited, and just as irrational as Beth’s conversion. For
instance, many people have undergone shifts from being self-endorsed
party animals to being self-endorsed industrious workers because of myste-
rious factors they regard as ‘age’ or the ‘drying up of hormones’. Other peo-
ple begin to value parenthood—value, not just like—the moment their
(formerly unwanted) children are born. Other people have converted from
atheism to religion as the result of an experience of extreme loneliness and
pain, and many peoples’ convictions, however well rationalized, are the
result of irrational or a-rational factors so unconscious that the agent had
about as much opportunity to subject them to reflective criticism as Beth
had to scrutinize her brainwashing process. Those people are not nearly as
obviously nonautonomous as Beth and are unlikely to be used as paradig-
matic examples of lack of autonomy. And yet, the only thing that distin-
guishes them from Beth is that their irrational conversions are not the result
of a deliberate and wrongful action by another human being, but her conversion
is the result of such an action. A person who converted from atheism to
religion as an irrational reaction to pain and extreme loneliness does not
strike one as particularly nonautonomous, but a person who was deliber-
ately placed in a state of pain and loneliness by a brainwasher to obtain the
same result sounds considerably less autonomous. Why is that?
I cannot but suspect that this has to do with the fact that brainwashed
Beth, and not the ordinary irrational convert, lacks autonomy in two
senses. First, she is quite obviously lacking autonomy in the sense of lack-
ing independence of mind—in the same sense that a person who received
all of her convictions from her mother is lacking in autonomy, only worse:
after all, there exists another person, the dean, who has given her a subs-
tantial part of her value system. Second, it is obvious also that in the moral,
normative sense, her autonomy has been violated, because she did not give
her permission for the brain-washing operation; the dean violated her
autonomy in the same sense that he would have violated her autonomy if
he were to switch her long distance carrier behind her back. In these two
senses, it is obvious that Beth is lacking autonomy. In other senses, it is not
184 N. Arpaly
obvious at all that she lacks autonomy, or that she lacks it more than the
other converts mentioned.
Is present-day Beth autonomous in the sense of agent autonomy or self-
control? We would have to say she is—Mele himself states that she is
the same as Ann in this way. She does things that she reflectively endorses,
and cannot be regarded as ‘wanton’, compulsive, and so forth. Similarly,
present-day Beth is not lacking autonomy in the sense of being particularly
irrational or particularly unresponsive to reasons. It is true that the shift in
her basic worldview was in fact irrational, or unresponsive to reasons, but
again, even very rational people have had such shifts, and that does not
make their current actions less rational or less responsive to reasons. In her
current actions, including staying in the office for long days, Beth may be
as responsive to reasons as Ann is. One might wish to introduce a new
sense of autonomy—a sense in which a person is autonomous only if all of
her convictions are the result of reason-responsive processes or at very least
intentionally induced by her in a self-controlled act of self-indoctrination.
But this would be a heroic notion of autonomy, of little use to the moral re-
sponsibility theorist. If one wishes to maintain that only autonomous agents
are morally responsible, one cannot say that the relevant sense of ‘autonomy’
requires a history that does not include irrational and uninvited changes of
outlook due to growing up, growing old, falling in love, loneliness, parent-
hood, hormones, Prozac, or fatigue, because this would mean that very few
people would be morally responsible at all. At the end of the day, the only
ways in which Beth may be said to be nonautonomous involve the fact that
her convictions are secondhand, the fact that the dean is in the habit of
violating her rights, and the fact that she is not some sort of super-rational
creature who never undergoes irrational or uninvited changes. Yet, neither
independence of mind nor having one’s rights respected nor heroic auton-
omy seemed, prima facie, when considered by themselves, particularly rel-
evant to the question of moral responsibility or lack thereof. Anyone who
wishes to argue that Beth is not morally responsible for her actions would
need to explain why having been irrationally influenced by an evil human
being exempts her from responsibility in a way that having been influenced
in a similar way by some unlucky chance of a force of nature does not.
Is Beth responsible for her actions? I shall not answer this question here.
For now, let me just point out that this is a real question (as Mele acknowl-
edges) and intuitions run both ways, at times disturbed by personal identity
Which Autonomy? 185
concerns. One need only remember that Patty Hearst, after changing her
values as a result of brainwashing, was still convicted for her crimes, to re-
alize that the issue is complex. My point is that the language of autonomy
obscures the complexity of the issue. It is not clear whether Beth is morally
responsible. It seems obvious that in at least two senses, she is not auto-
nomous. By substituting the question ‘Is Beth autonomous?’ for the ques-
tion ‘Is Beth morally responsible?’—treating the questions as if they were
identical—an author may prejudice the reader towards a negative answer to
the question of Beth’s responsibility, turning a complex question into some-
thing trivial.
Let us look at another case study—this time, not a quote from a philoso-
phy book but a story by John Le Carre. Oliver Single is a young lawyer in a
prestigious firm—an heir apparent to his illustrious father, Tyger Single,
who heads the firm. The firm is heavily involved in organized crime. Oliver
had always known that his father makes shady deals, but only gradually
does he learn quite how shady. Even when he knows quite well that his
father performs immoral actions, he still has no plans to leave his post. He
enjoys the money and the prestige, and in addition, he loves his father
deeply and wishes to please him. At a critical moment in the novel, Oliver
defects from the side of his criminal family to the side of the law, in whose
service he remains to perform some courageous, intuitively admirable acts.
Here is how Le Carre describes Oliver’s defection:
It is Midwinter, and Oliver is a little mad. That much he knows about himself, no more.
The Origins of his madness, its causes, duration and degree, are not within his grasp,
not now. They are out there, but for another time, another life, another couple of
brandies. The neon-lit gloom of a December’s night at Heathrow airport reminds
him of a boy’s changing room in one of his many boarding schools. Garish card-
board reindeer and taped carols compound his mood of unreality. Snowy lettering
dangles from a washing line, wishing him peace and joy on earth. Something amaz-
ing is about to happen to him and he is eager to find out what it will be. He isn’t drunk
but technically he isn’t sober either. A few vodkas on the flight, a half of red with the
rubber chicken, a remy or two afterwards have done no more in Oliver’s eyes than
bringing him up to speed with the furor already raging inside him. He has only hand
luggage and nothing to declare except a reckless ferment of the brain, a firestorm of
outrage and exasperation starting so long ago that is origins are impossible to peg,
blowing through his head like a hurricane while other members of his internal congre-
gation stand around in timid twos and threes and ask one another what on earth Oliver is
going to do to put it out. He is approaching signs of different colors and instead of
wishing him peace, joy on earth and goodwill among men, they are requiring him
186 N. Arpaly
to define himself. Is he a stranger in his own country? Answer, yes he is. Has he
arrived here from another planet? Answer, yes he has. Is he blue? Red? Green? His
eye drifts to a tomato-colored telephone . . . It is familiar to him. Perhaps he noticed
it on his way out three days ago and, unknown to himself, recruited it as a secret ally.
Is it heavy, hot, alive? A notice beside it reads “If you wish to speak to a Customs
officer, use this phone.” He uses it. That is to say, his arm reaches out for it unbidden, his
hand grasps it and puts it to his ear, leaving him with the responsibility of what to
say. (Le Carre 1998, 90)
Much of the novel is built upon the reader’s admiration of Oliver and his
defection, and this admiration assumes moral praiseworthiness—hence,
moral responsibility. The sense that Oliver is morally responsible, that his de-
fection is not just an accidental drunk action, comes from the feeling that
his action is, after all, reason-responsive: it is not an accident that the ac-
tion occurs right after his discovery of his father’s worst crimes, it is not an
accident that Oliver has noted the red phone long ago, as if knowing with-
out knowing that some day he’ll use it, and one even gets the impression
that Oliver’s drinking on the airplane is taken as a preemptive tranquilizer
to assist him in the hard task to come. Oliver’s action also appears to be
authentic: something deep and real in him, his moral sensibility, finally re-
volts against his life as his father’s heir. On the other hand, as for self-control,
Oliver does not seem to have much during his action, and Le Carre stresses
again and again the passivity and turmoil which he experiences. He is
alienated from his actions, and does not seem to identify with them. Le
Carre’s description of Oliver’s dark night of the soul is rather credible, and
may ring true to many people who have left a father, a religion, a marriage.
Is Oliver autonomous? Is there any clear meaning to this question—or any
good reason to ask it?
“Of course,” some agency theorists will answer. “After all, there is a rea-
son to ask if Oliver is morally responsible for his action, and the asking if
he is autonomous is simply the same as asking if he is morally responsible.
And may we point out that when we use the word ‘autonomy’, we do not
mean any of eight things you have listed, for by ‘autonomy’ we mean sim-
ply that on which moral responsibility depends, whatever it turns out to be.
Thus, there is also a reason to ask what it is that constitutes autonomy—
this is simply the same as asking what it is that makes an agent morally
responsible.”
But I take myself to have demonstrated why this may not be the best way
to pose the question of moral responsibility, as at least some of the meanings
Which Autonomy? 187
I have listed invariably get smuggled in, mixed together, whenever we use
the word autonomy, thus inviting confusion. Furthermore, there is also
the risk that ‘What is autonomy?’ is a bit too loaded a way to bring about the
question of moral responsibility, focusing our attention on some kinds of
solutions and not others. Perhaps we should simply ask what it is that makes
an agent morally responsible, or praiseworthy, or blameworthy, and give a
much overworked term of art a well-deserved rest.
Acknowledgments
This work, originally written for the Inland Northwest Philosophy Confer-
ence, appears in a modified form as a part of my book Unprincipled Virtue:
An Inquiry Into Human Agency, forthcoming, Oxford University Press.
Notes
1. Described in Frankfurt (1976).
2. It may be that it is common in North American culture to viscerally expect self-

control from one’s self and experience any failure of self-control as disturbing, baf-
fling, and surprising. On the other hand, Russian novels are full of people who
perceive themselves and humans in general as weak and passive, and are pleasantly
surprised, and grateful to God, for any exception to this (for example, Lebedeyev in
The Idiot).
3. Timothy Schroeder and I presented an extended version of these arguments in

Arpaly and Schroeder (1999).
4. Scanlon (2002), Damasio (1994), and Audi (1990). Railton’s work is yet unpub-
lished, though it has been presented in many talks.
5. See Fischer and Ravizza (1998).
References
Anderson, E. 1997. “Practical Reason and Incommensurable Goods.” In R. Chang, ed.,

Incommensurability, Incomparability and Practical Reason. Cambridge, Mass.: Harvard
University Press.
Arpaly, N. 2000. “On Acting Rationally Against One’s Best Judgment.” Ethics 110:
488–513.
. Forthcoming. Unprincipled Virtue: An Inquiry Into Human Agency. Oxford:

188 N. Arpaly
Arpaly, N. and T. Schroeder. 1999. “Alienation and Externality.” Canadian Journal of

Audi, R. 1990. “Weakness of Will and Rational Action.” Australasian Journal of

Copp, D. In preparation. Rationality and Autonomy.
Damasio, A. 1994. Descartes’ Error: Emotion, Reason, and the Human Brain. New York:
G. P. Putnam.
Fischer, J. and M. Ravizza. 1998. Responsibility and Control: A Theory of Moral Respon-
sibility. New York: Cambridge University Press.
Frankfurt, H. 1976. “Identification and Externality.” In A. Rorty, ed., The Identities of

Persons. Berkeley, Calif.: University of California Press.
. 1988. “Rationality and the Unthinkable.” In H. Frankfurt, The Importance of

What We Care About. Cambridge: Cambridge University Press.
Hill, T. 1991. “Servility and Self-respect.” In T. Hill, Autonomy and Self-respect.

Le Carre, J. 1998. Single and Single. New York: Charles Scribner’s Sons.
Mele, A. 1995. Autonomous Agents. New York: Oxford University Press.
Scanlon, T. 2002. “Reasons and Passions.” In S. Buss and L. Overton, eds., Contours of
Agency. Cambridge, Mass.: MIT Press.
Strawson, P. 1962. “Freedom and Resentment.” Proceedings of the British Academy

48: 1–25.
Velleman, D. 2002. “Identification and Identity.” In S. Buss and L. Overton, eds.,

Contours of Agency. Cambridge, Mass.: MIT Press.
9 The Transfer of Nonresponsibility
John Martin Fischer
1 Introduction
In ancient times—some fifteen years ago—I suggested that Frankfurt-type

examples call into question the principle of transfer of nonresponsibility
(which I then called, a bit too narrowly, the ‘principle of transfer of blame-
lessness’, following John Taurek’s usage in his fascinating Ph.D. disserta-
tion at UCLA in 1972).1 In the introductory essay to my anthology, Moral
Responsibility, I presented a somewhat informal version of van Inwagen’s
modal principle (which he called principle ‘B’), and (following van Inwagen)
explained how it could be employed as part of a ‘direct’ argument for the
incompatibility of causal determinism and moral responsibility (i.e., an
argument for the incompatibility claim that does not employ the claim that
causal determinism rules out alternative possibilities):
. . . if you are not morally responsible for one thing, and you are not morally re-
sponsible for that thing’s leading to another, you are not morally responsible for the
other. [This, roughly speaking, is the principle of transfer of nonresponsibility.] Now,
an argument . . . can be generated to show that causal determinism rules out moral
responsibility. Given that you are not morally responsible for the past, and you are
not morally responsible for the laws of nature, and assuming the principle of trans-
fer of blamelessness [the principle of transfer of nonresponsibility], causal determin-
ism seems to rule out moral responsibility. (Fischer 1986, 60)
Also, I told the following story:
. . . a man [Green] walks along a beach and, noting that there is a child drowning,
dives into the water and rescues the child. Though Green has had a device implanted
in his brain [by scientists in a research institute in California—one might now say
‘crazed neurophilosophers in La Jolla’], the device does not play any role in Green’s
decision to save the child (and his subsequent action). That is, the device monitors
190 J. M. Fischer
Green’s brain activity but does not actually intervene in it. Let us suppose that this
is because the scientists can see that Green is about to decide to save the child and to
act accordingly [they are morally good, albeit crazed, neurophilosophers]. But let’s
also suppose that the scientists would have intervened to bring about a decision to
save the child if Green had shown an inclination to decide to refrain from saving the
child. That is, were Green inclined to decide on his own not to save the child, the
scientists would ensure electronically that he decide to save the child and also that
he act to carry out this decision. (Fischer 1986, 41)
Of course, this case contains the distinctive characteristics of a Frankfurt-

type case: a fail-safe arrangement that plays no actual role but the presence
of which nevertheless ensures the actual result.2
I then suggested that the Frankfurt-type examples are plausible coun-
terexamples to the principle of transfer of nonresponsibility, even though
they would not be counterexamples to the parallel modal principle em-
ployed in the argument for the incompatibility of causal determinism and
alternative possibilities (the principle of the transfer of powerlessness):
. . . Green is not morally responsible for the fact that the scientists are ready to inter-
vene, and he is not responsible for the fact that, if they are so ready, he will save the
child. But he does seem to be morally responsible for saving the child. . . . So a com-
patibilist about determinism and moral responsibility might accept the fixity of the
past . . . the fixity of the laws . . ., and the principle of transfer of powerlessness but
might reject the principle of transfer of blamelessness. (Fischer 1986, 60–61).
Thus, semicompatibilism was born. Here I wish to defend the basic intu-
ition, which I still believe is correct, that the principle of transfer of blame-
lessness (or, more broadly, nonresponsibility) is called into question by the
Frankfurt-type cases, and that it cannot be employed in an uncontroversial,
decisive argument against the compatibility of causal determinism and
2 The Principle of Transfer of Nonresponsibility and a Preliminary Critique
It will be helpful to have a slightly more careful formulation of the princi-

ple of transfer of nonresponsibility (Transfer NR):
If
(1) p obtains and no one is even partly morally responsible for p; and
(2) if p obtains, then q obtains, and no one is even partly morally respon-
sible for the fact that if p obtains, then q obtains;
The Transfer of Nonresponsibility 191
then
(3) q obtains, and no one is even partly morally responsible for q. (Fischer
and Ravizza 1998, 152)
In Responsibility and Control: A Theory of Moral Responsibility, Mark Ravizza

and I criticize Transfer NR by presenting the following sort of Frankfurt-
type scenario:
[In “Erosion”], Betty plants . . . explosives in the crevices of [a] glacier and [intu-
itively speaking, freely] detonates the charge at T1 causing an avalanche that crushes
the enemy fortress at T3 [a result intended by Betty]. Unbeknownst to Betty . . . how-
ever, the glacier is gradually melting, shifting, and eroding. Had Betty not placed the
dynamite in the crevices, some ice and rocks would have broken free at T2, starting
a natural avalanche that would have crushed the enemy camp at T3. (Fischer and
Ravizza 1998, 157)
Our contention is that Betty is (intuitively) morally responsible for bring-

ing it about that there is an avalanche that crushes the enemy base at T3,
despite the fact that no one is even partly responsible for the fact that the
glacier is eroding, and no one is even partly responsible for the fact that if
the glacier is eroding, then there is an avalanche that crushes the enemy
base at T3. Thus, we reject Transfer NR.3
Now it might be objected that the potency of ‘Erosion’ as a counterex-
ample to Transfer NR depends on its being a case of preemptive overdeter-
mination. Consider, now, Transfer NR:
If
(2) (i) p is part of the actual sequence of events e that gives rise to q at T3,
(ii) p is a part of e that is causally sufficient for the obtaining of q at T3,
and
(iii) no one is or ever has been even partly responsible for (2i and ii);
then
(3) q obtains at T3, and no one is or ever has been even partly morally
responsible for q.4
But a case similar to Erosion, Erosion*, shows Transfer NR to be problematic.
In Erosion*, the conditions of the glacier do actually cause the ice and rocks
to break free, triggering an avalanche that arrives at the fortress precisely at
the same time as the independent avalanche triggered freely by Betty. Each
avalanche is sufficient for the destruction of the fortress, and yet Betty seems
192 J. M. Fischer
to be morally responsible for bringing it about that there is an avalanche

that destroys the enemy fortress at T3, just as in Erosion. Erosion* is analo-
gous to Erosion in every respect, except in Erosion* the overdetermination is
simultaneous, rather than preemptive. This difference does not appear to
make a difference with respect to Betty’s moral responsibility. Thus, we reject
Transfer NR.
Consider the typical sort of example van Inwagen employs to motivate the
acceptance of a principle such as Transfer NR or Transfer NR, ‘Snake Bite’
(van Inwagen 1983, 187). Imagine that John was bitten by a cobra on his thir-
tieth birthday, and no one was even partly responsible for this. Suppose, also,
that if John was bitten by a cobra on his thirtieth birthday, then John died on
his thirtieth birthday, and no one was even partly responsible for this fact. It
seems to follow that John died on his thirtieth birthday, and no one was even
partly responsible for the fact that John died on his thirtieth birthday.
Similarly, suppose that an earthquake takes place in the middle of the
Pacific Ocean, and that no one is even partly morally responsible for this.
Also suppose that if this earthquake occurs, then a tsunami will hit the coast
of California, and no one is even partly responsible for this fact. It appears
to follow that a tsunami will hit the coast of California, and no one is even
partly morally responsible for this. Call this case ‘Tsunami’. These sorts of
examples seem to support a principle of transfer of nonresponsibility.
But Mark Ravizza and I contend that these examples, like the others
invoked by van Inwagen (and other proponents of the idea of transfer of
nonresponsibility), appear to support a transfer principle only because one
pictures there being only one path to John’s death (or the tsunami’s hitting
the coast of California), along which no agents act freely (Fischer and
Ravizza 1998, 166). But the one-path cases are only a proper subset of all
the cases involving a putative transfer of nonresponsibility. The mistake of
the proponents of transfer is to generalize from a mere proper part of the
space of relevant possibilities.
3 McKenna’s Critique of the Fischer and Ravizza Attack on Transfer
The Critique
In an interesting recent paper, Michael McKenna has argued that the criti-
cism of the transfer idea developed above is unsuccessful (McKenna 2001).
He says:
As Fischer and Ravizza observe, their reply relies exclusively upon examples that involve
‘two-path’ cases, i.e., cases in which the obtaining of the event is ensured by two dif-
ferent causal pathways to the same event. They do not object to Transfer NR as it applies
to one-path cases, but hold that Transfer NR restricted to one-path cases cannot be
used to show that moral responsibility is incompatible with causal determinism.
This is because causal determinism does not rule out cases of overdetermi-
nation. They therefore hang their criticism of Transfer NR on counterexamples in-
volving two-path cases of simultaneous overdetermination. (McKenna 2001, 45)
But McKenna is not sanguine about the prospects for our approach; he goes
on as follows:
This is not a convincing strategy. No doubt, Fischer and Ravizza are correct that
causal determinism does not rule out overdetermination. But if determinism is true,
then the manner in which the facts of the past and the laws of nature entail one
unique future is not analogous to the manner in which one set of independently
existing causally sufficient conditions (for example, an erosion) ensures a subse-
quent event also ensured by some distinct set of independently existing sufficient
conditions (i.e., Betty’s action). Assuming determinism, the pertinent facts (consist-
ing in the deterministic order of things) are not independent of an agent’s reasons for
action, they constitute them! Therefore, at a deterministic world involving a typical
case regarding a judgment of moral responsibility, the case is relevantly like a one-
path, not a two-path case. (McKenna 2001, 45)
Although we had chastised van Inwagen for attending only to a proper sub-
set of the relevant cases, the one-path cases, it is precisely McKenna’s point
that these are the only cases relevant to causal determinism. Thus, he seeks
to formulate a version of the modal principle that captures a restriction to
one-path cases. Here is a slight reformulation of McKenna’s proposal, Trans-
fer NR :
If
(1) p obtains and no one is or ever has been even partly morally
responsible for p; and
(2) (i) p is part of the actual sequence of events e that gives rise to q at T3;
(ii) p is causally sufficient for the obtaining of q at T3; and any other
part of e that is causally sufficient for q either causes or is caused by p;
and
(iii) no one is or ever has been even partly responsible for (2i and ii);
then
194 J. M. Fischer
(3) q obtains, and no one is or ever has been even partly morally res-
ponsible for the fact that q obtains at T3. (McKenna 2001, 45)5
McKenna points out that Transfer NR does not rule out moral responsibil-
ity in cases of causal overdetermination, even in a causally deterministic
world. But he says that it would be very implausible for Ravizza and me to
hold that an agent can be morally responsible only in cases of simultaneous
overdetermination. Indeed, McKenna credits both Carl Ginet and Eleonore
Stump with pointing out that it would be a significant problem for the
Fischer/Ravizza approach if moral responsibility turned out to be incom-
patible with causal determinism in all one-path cases, which are, presum-
ably, the vast majority of cases (McKenna 2001, 46 and note 39, 51).
A Reply to McKenna
McKenna’s paper contains many insightful and helpful points. He makes
an important observation when he points out that the manner in which
determinism ensures a unique future may well be crucially different from
the manner in which one set of independent causally sufficient conditions
(say, erosion) ensures a subsequent event also ensured by some distinct set
of independently existing sufficient conditions (for example, Betty’s free
action). I will return to this point below, and also in my discussion
of Eleonore Stump’s defense of the transfer of nonresponsibility idea. But
here I wish to disentangle different points. One contention of McKenna is
that the validity of Transfer NR would pose significant problems for a com-
patibilist about causal determinism and moral responsibility (like me).
Granted. But I think a rather different issue is whether Transfer NR can be
employed to establish the incompatibility of causal determinism and moral
responsibility.
Mark Ravizza and I contend that it is futile to seek to restrict the modal
principle to one-path cases (as in Transfer NR ) precisely because this would
render the principle unable to generate incompatibilism about causal deter-
minism and moral responsibility (which is, after all, the conclusion of van
Inwagen’s argument).6 To point out that Transfer NR poses problems for
developing a plausible compatibilist theory of moral responsibility is one
thing; to suppose that it generates a successful argument for incompatibil-
ism is quite another. One possibility is that there are significant problems for
both compatibilism and incompatibilism, as Saul Smilansky has recently
contended (Smilansky 2000).
To see an additional problem with moving from there being a prob-

lem with compatibilism to the general thesis of incompatibilism, suppose
causal determinism is true in a world, but that in the world in question
some (morally significant) behavior is the result of simultaneous overdeter-
mination. Now that behavior, for all Transfer NR tells us, is such that the
relevant agent can be held morally responsible for it. Thus, we do not (yet)
have the incompatibility of causal determinism and moral responsibility—
only the incompatibility of causal determinism with moral responsibility
for some behavior. If there were a world—unusual as this would be—with all
of the morally significant behavior occurring as a result of simultaneous
overdetermination, then Transfer NR would be entirely consistent with
moral responsibility coexisting peacefully with causal determinism in this
world. I find this somewhat strange.
Note what is happening with Transfer NR : the principle entails that when
behavior is the result of one causally deterministic sequence, there cannot
be moral responsibility, but when the behavior is the result of two or more
such sequences, there may be moral responsibility. But if one causally de-
terministic sequence rules out moral responsibility, it would seem that two
or more would be even worse. Transfer NR , however, gives us precisely the
opposite result. Indeed, Transfer NR seems to entail, for the relevant sorts of
normative notions, that two wrongs may well make a right.7
I conclude, then, that whatever problems such a principle poses for com-
patibilism, it cannot be employed as it is—without further resources—to
generate a successful argument for the incompatibility of causal determin-
ism and moral responsibility. But someone might reply that Transfer NR
might be the first step in an argument for incompatibilism. On this ap-
proach, one would employ Transfer NR to rule out moral responsibility in
one-path cases, and then supplement it with the contention that merely
adding causally deterministic paths (to a context in which there is no
responsibility) cannot issue in moral responsibility.
My reply to this sort of more complex argumentative strategy is to insist
that I need not grant the first step: the acceptability of Transfer NR in the
context of one path. But can I produce any convincing counterexamples to
Transfer NR ? Let us say that there is just one path to an outcome, and that
this path is causally deterministic. Then of course it would be question-
begging, in the context in which Transfer NR is presumably employed (i.e.,
in an argument for the incompatibility of causal determinism and moral
196 J. M. Fischer
responsibility), to contend that nevertheless an agent is morally re-

sponsible for that outcome. But that would seem to be the only way one
could produce a counterexample to Transfer NR . Thus, if McKenna’s
Transfer NR actually succeeds in capturing a restriction of Transfer NR to
one-path cases, then it is so constructed that it is impossible for there to be
a non-question-begging counterexample to it.
Now some philosophers might think this a good thing. But I believe that
it can lead to a certain distinctive kind of stalemate—what I have called a
‘dialectical stalemate’. McKenna is aware of this possibility, but he never-
theless contends that Transfer NR can be supported; I will return to this
claim below. I shall now turn to a discussion of Eleonore Stump’s defense of
the transfer of nonresponsibility, which will raise similar issues to those we
have just discussed.
4 Stump’s Defense of Transfer NR
Stump’s Defense
In two separate places, Eleonore Stump has sought to defend the basic idea
of the transfer of nonresponsibility, and thus the direct argument for the
incompatibility of causal determinism and moral responsibility, against the
attacks mounted by Ravizza and me. In the first piece, “The Direct Argu-
ment for Incompatibilism,” Stump agrees that cases such as Erosion and
Erosion* show the inadequacy of Transfer NR and Transfer NR (Stump
2000). But she, like McKenna, complains that these are two-path cases. She
contends that no one-path case can similarly show the inadequacy of
a suitably restricted transfer principle—a principle whose application is
restricted to one-path cases.
In the new piece, “Control and Causal Determinism,” she offers what she
characterizes as a “significant revision” of the argument in her first paper
(Stump 2002, note 16, p. 57). She says: “In their response . . . Fischer and
Ravizza concede the general conclusion of my argument but argue that it
constitutes only a stalemate between the incompatibilist and their posi-
tion. Fischer and Ravizza are not right in this view, as I hope to show in the
revised version of the argument I give in this essay” (ibid.).
But this is a bit puzzling. I do not know what Stump means in saying that
Ravizza and I “concede the general conclusion” of her argument. At the end
of that first paper, Stump says, “Consequently, Fischer and Ravizza have not
succeeded in showing that the direct argument for incompatibilism fails.” I
do not think we agreed with this conclusion. We did agree that we could not
decisively show that a transfer principle restricted to one-path cases is unac-
ceptable. But we also pointed out that it is not our obligation decisively to
prove such a principle problematic, in order to show that one need not
accept the argument for incompatibilism. Given that there are considerable
attractions to compatibilism (for example, a compatibilist need not think
that our very basic views of ourselves and others as morally responsible and
as persons are held hostage to a possible scientific discovery by a consor-
tium of scientists that causal determinism is indeed true), and given (what
we argued) that a restricted transfer principle cannot be uncontroversially
established (i.e., that one reaches a dialectical stalemate in seeking to argue
for—or against—such a principle), we concluded that the incompatibilist’s
argument is not persuasive. That is, in light of the fact that its central prin-
ciple is essentially contested and cannot uncontroversially be supported,
we showed that one need not accept the incompatibilist’s argument. We did
think we were showing that the incompatibilist’s argument fails, insofar as
we thought we were showing that it does not succeed within the relevant
dialectic context.
It will be useful to consider Stump’s new formulation of her argument,
and to consider it in light of the discussion of McKenna’s Transfer NR in
the previous section. She proceeds by considering Erosion* and construct-
ing a dilemma. On the first horn of the dilemma, one is to assume that the
world of Erosion* is not causally deterministic and that Betty acts indeter-
ministically (but in a way that supports the judgments that she is acting
freely). Now Stump is willing to concede that this sort of example shows
that Transfer NR, unqualified or restricted, is invalid; but she thinks that
the principle can simply be qualified so as to rule out the counterexample
and still get the desired incompatibilistic result. She says,
. . . Erosion* shows that in cases of overdetermined effects, one of whose causes is an
agent acting indeterministically, Transfer NR fails. But this is not enough to show
that Transfer NR should be rejected. Because Transfer NR is a highly plausible princi-
ple, we can take Erosion* to show just that the principle needs to be restricted, to
exclude such unusual cases as that in which an indeterministic agent is one of the
causes of an overdetermined effect. (Stump 2002, pp. 40–41)
On the other horn of the dilemma, the assumption is that the world of
Erosion* is causally deterministic. But now of course it will be question-
begging, in the relevant dialectical context (in which Transfer NR is being
deployed in order to establish that causal determinism is incompatible
198 J. M. Fischer
with moral responsibility, and thus the compatibility of causal determin-

ism and moral responsibility is in doubt) simply to assert that Betty is
morally responsible for her behavior (and its upshots). But, as we saw in our
discussion of McKenna’s Transfer NR , only if Betty is morally responsible
can there be a counterexample to Transfer NR. Stump contends, then, that
on the assumption of causal determinism, it would be question-begging to
assert that Erosion* is a counterexample to Transfer NR.8
Reply to Stump
I would contend that there is trouble for Stump’s incompatibilistic con-
clusion, on either horn of the dilemma. Consider, first, the assumption of
causal indeterminism. Here Stump suggests a restriction of Transfer NR to
one-path contexts. She doesn’t actually present the principle, but if such
a principle can be crafted, it will have exactly the same problems as
McKenna’s Transfer NR . As I explained above, it will not be possible to em-
ploy this sort of one-path modal principle in a general argument for incom-
patibilism about causal determinism and moral responsibility. And if the
single path contains some insuring factor as seems required in order to trig-
ger the application of the modal principle, then it would seem that the
principle is implicitly presupposing causal determinism. If this is so (and it
explicitly is the case on the second horn of Stump’s dilemma), then it is
admittedly impossible, within the relevant dialectical context (in which
the compatibility of causal determinism and moral responsibility is at
issue) to produce a non-question-begging counterexample to Transfer NR.
But I contend that it is equally impossible (within the relevant dialectical
context) to provide examples that give the necessary sort of support to the
principle. That is, I believe that the current dialectical context is what I
have characterized in previous work as a dialectical stalemate (Fischer 1994,
83–85). This is my characterization of a dialectical stalemate:
Frequently in philosophy we are engaged in considering a certain argument . . . for
some claim C. The argument employs a principle P. Allegedly, P supports C. Now the
proponent of the argument may be called upon to support the principle, and he may
do so by invoking a set of examples (or other considerations). Based on these exam-
ples (or other considerations), he argues that the principle and thus also the philo-
sophical claim are to be accepted.
But the opponent of the argument may respond as follows. The examples are not
sufficient to establish the principle P. One could embrace all the examples and not
adduce P to explain them; rather, it is alleged that a weaker principle, P* is all that is
decisively established by the examples (or other considerations). Further, P*, in con-
trast to P does not support C. Finally, it is very hard to see how one could decisively
establish P. One reason it is so difficult is that it at least appears that one cannot
invoke a particular example that would decisively establish P without begging the
question in a straightforward fashion against either the opponent of P or the oppo-
nent of C. Further, it also seems that one cannot invoke a particular example that
would decisively refute P without begging the question against the proponent of P or
the proponent of C. These conditions mark out a distinctive—and particularly
precarious—spot in dialectical space. (Fischer 1994, 83)
But if in our discussion of the principles of the transfer of nonresponsibil-

ity we have reached one of these precarious spots—dialectical black holes,
as it were—it must be the case that there is some alternative modal
principle—an alternative to a suitably restricted Transfer NR, such as Trans-
fer NR —which explains the relevant examples equally well but does not
entail compatibilism.
What is this competitor modal principle? In order to frame the principle,
I will pause to describe what I believe is the basis of our intuitions in the
two-path cases (and thus what explains why these cases are counterexam-
ples to Transfer NR). Having done this, I will be in a position to frame an
alternative modal principle that explains all of the uncontroversial cases as
well as Transfer NR , but does not (together with other uncontroversial
ingredients) entail incompatibilism.
5 A Compatibilistic Transfer Principle
Return to Erosion*. I would say (to a first approximation, at least) that Betty
is morally responsible for her behavior, and its upshot (the destruction of
the enemy camp at T3), because she engaged in unimpaired practical
reasoning—practical reasoning undistorted by factors that uncontrover-
sially threaten moral responsibility, such as significant delusions, psychoses,
coercion, compulsion, irresistible impulses, clandestine manipulation, hyp-
nosis, subliminal advertising, and so forth. I am inclined to say that she is
morally responsible, in other words, because no uncontroversially responsi-
bility undermining factor impairs (or in any way affects) her deliberations,
her formation of an intention, and her action in accordance with it. This
is so, even though there is a second path to the same upshot that is a no-
responsibility path. The fact that the no-responsibility path is entirely sepa-
rate insulates Betty’s path from it, and allows all parties to the disagreement
200 J. M. Fischer
about causal determinism and moral responsibility to agree that Betty is

morally responsible here (apart from any special assumption, such as that
causal determinism is true). Insofar as the no-responsibility path is separate,
it is sequestered and cannot (in the economists’ term) “crowd out” the
factors that intuitively ground moral responsibility. Erosion, which involves
mere preemptive overdetermination, is an even clearer example of this sort
of sequestration or partitioning.
The problem with a one-path case is that there is a danger posed by the
phenomenon of crowding out: one cannot so easily insulate what puta-
tively grounds moral responsibility from alleged responsibility undermin-
ing factors. This is what McKenna has in mind when he says that the
manner in which the facts of the past and the laws of nature ensure one
unique future under causal determinism is not analogous to the way in
which one unique future is ensured in a case in which there are two sepa-
rate pathways (one of which is indeterministic) that overdetermine the
behavior and consequences in question. As McKenna puts the point,
“Assuming determinism, the pertinent facts (consisting in the determinis-
tic order of things) are not independent of an agent’s reasons for action, they
constitute them!” (McKenna 2001, 45).
I concede McKenna’s contention about constitution, but I do not
thereby accept that moral responsibility is necessarily crowded out. A one-
path case could be thought of as two separate paths that are superimposed.
Alternatively, I would recommend that we think of the one path as having
(at least) two separate sets of features. Suppose Betty deliberates just as she
actually does in the Erosion cases, but there is no “second path”—no
glacier that will independently cause the upshot (the destruction of the
enemy camp at T3). But imagine further that causal determinism obtains in
the actual path that leads to Betty’s decision and action. One set of features
entails that there is along this path just the sort of unimpaired deliberation
that obtains in Erosion and Erosion*—practical reasoning and subsequent
action undistorted by factors that uncontroversially threaten moral re-
sponsibility. But of course another set of features renders the actual sequ-
ence of events causally deterministic, and causal determinism is alleged by
the incompatibilist to rule out moral responsibility.
Note, however, that causal determination is not a factor that uncontrover-
sially rules out moral responsibility. The features in virtue of which the
actual path is causally deterministic are thus importantly different from
a feature such as an avalanche caused by movements of a glacier, which
uncontroversially fails to confer responsibility. A proponent of the incom-

patibility of causal determinism and moral responsibility is not entitled to
help himself, at this point in the disagreement, to the claim that causal de-
terminism crowds out moral responsibility. Note that if he were allowed to
help himself to this contention, then there would be no point to looking
for a modal principle, such as Transfer NR (or some restriction of it),
because the incompatibilistic result would already have been achieved
(startlingly easily) right from the start.
On my view, then, causal determination and unimpaired deliberation
can be superimposed upon each other without distortion of the delibera-
tion. This is analogous to the possibility of superimposing (the precise tech-
nology is not important) two audio tapes—perhaps a tape of a woman
singing and a tape of a piano accompaniment. When the tapes are put
together (and of course this is how much music is actually produced), the
piano need not distort, impair, or otherwise etiolate the singing. The same
of course is true about the relationship between the singing and the piano.
Similarly, a video tape of one character in a film can be superimposed on a
video tape of another character; what emerges may in no way obscure or
change the original depiction of each of the characters. My claim is that the
relationship between causal determination and undistorted human practi-
cal reasoning is relevantly similar: we can see the determination as being
superimposed on the practical reasoning without thereby distorting it.
Of course, it would also not be fair for me to say that it is obvious and in-
disputable that Betty is morally responsible for her behavior and its upshot
in the one-path case under consideration; that would straightforwardly beg
the question against the incompatibilist. Rather, my claim is that it cannot
simply be assumed, within the dialectical context we are exploring, that
causal determinism crowds out moral responsibility. Granted: the relation-
ship between the causal determination and the practical reasoning is more
intimate than (say) the relationship between the singing and the piano; but
this difference in itself cannot be assumed to make a decisive difference.
The above considerations suggest an alternative modal principle; call it
Transfer NRC:
If
(2) if p obtains, then q obtains, and no one is even partly morally respon-
sible for the fact that if p obtains, then q obtains; and
202 J. M. Fischer
(3) on the actual path that leads from p’s obtaining to q’s obtaining, either
there is no factor that at least prima facie could be thought to ground moral
responsibility, or there is some factor that uncontroversially undermines
moral responsibility (e.g., a factor that distorts or impairs the distinctive
process of human practical reasoning); then
(4) q obtains, and no one is even partly morally responsible for q.
It will sometimes be helpful to refer to a path that meets conditions (1) and
(3) as a “no-responsibility” path, just for simplicity’s sake. Transfer NRC
is a plausible and attractive modal principle that explains why we transfer
nonresponsibility in the clear and uncontroversial cases. For example, it ex-
plains why no one is even partly morally responsible for the fact that John
dies (from a snakebite) on his thirtieth birthday in Snake Bite, and it ex-
plains why no one is even partly morally responsible for the tsunami’s hit-
ting the coast of California in Tsunami. It has this implication in all of the
examples invoked by van Inwagen in an attempt to support Transfer NR.
Note that it does not entail that Betty is not morally responsible in Erosion
or Erosion*. Note, further, that Transfer NRC cannot be employed as part of
a direct argument for the incompatibility of causal determinism and moral
responsibility that would appeal to those not antecedently committed to
incompatibilism. Transfer NRC thus explains all the uncontroversial exam-
ples just as well as (say) Transfer NR (or an otherwise qualified version of
Transfer NR that only applies to one-path cases), but does not help to yield
incompatibilism. It is thus precisely the sort of principle whose availability
shows that the attempt to employ some appropriately restricted version of
Transfer NR to establish incompatibilism leads to a dialectical stalemate.
McKenna is (to some degree) aware of the dialectical difficulties faced by
someone who wishes to employ Transfer NR to argue for incompatibilism
about causal determinism and moral responsibility (McKenna 2001, note
45, 51).9 In response, McKenna says that Transfer NR is not merely ad hoc,
but is independently confirmed by a range of cases: “. . . notice that cases
like Snakebite can be used to confirm the plausibility of Transfer NR
as readily as they can the leaner Transfer NR (ibid.). I grant that these cases
support Transfer NR as readily as they do Transfer NR. But this is not the
issue. The point is that Snake Bite (and Tsunami and all the cases invoked
by van Inwagen and other incompatibilists) are unable to support Transfer
NR over Transfer NRC. So, as with dialectical stalemates in general, there is
a competitor principle—Transfer NRC—which must be considered when
we are evaluating Transfer NR . It is crucial here that the relevant data do

not support Transfer NR over Transfer NRC, and Transfer NRC does not
help to generate incompatibilism.
I claim then that the attempt to employ Transfer NR, restricted so that
it applies only to one-path causally deterministic contexts, as part of an
argument for the incompatibility of causal determinism and moral respon-
sibility, issues in a dialectical stalemate. This implies that such an attempt
does not generate a successful argument for incompatibilism. I believe that
there are certain reasons why compatibilism is attractive. Given these, I am
inclined toward compatibilism. I would give up compatibilism if I believed
there were a successful argument, employing principles that are broadly
and strongly appealing (clearly appealing even to those not already com-
mitted to incompatibilism), that has as its conclusion incompatibilism. But
I have yet to see such an argument. The argument that employs some
restriction or qualification of Transfer NR, employs a principle that can
only be established by begging the question against the compatibilist; it
does not employ a principle that is broadly appealing.
Now I do not claim to have proved such a principle unacceptable. Indeed,
I have admitted that I do not see how this would even be possible, given the
dialectical niche in which it is embedded. Thus, I have certainly not refuted
the direct argument for incompatibilism. But I do not feel any need to do so.
It is not my obligation to provide a refutation of this sort of argument. Given
that there are considerable attractions to compatibilism, and given that I
have shown that there is no successful direct argument (of the sort we have
been considering) for incompatibilism, I find no reason (as yet) to abandon
the compatibility claim.10 It is van Inwagen and Stump who claim to have
a direct argument that purports to establish the untenability of compatibil-
ism. If I have shown that their approach issues, at best, in a dialectical stale-
mate, then I have shown that they are wrong, at least if they are interested
in doing something more than simply preaching to the converted.
6 Conclusion
Certain incompatibilists about causal determinism and moral

responsibility—such as van Inwagen and Stump—have argued that one
can employ a valid transfer principle to yield the incompatibilistic result.
Transfer NR, originally suggested by van Inwagen, can be seen to be invalid
204 J. M. Fischer
by reference to two-path cases. Others, including Stump and McKenna,

have suggested that if the modal principle is either reformulated or re-
stricted in its applicability, the incompatibilistic result can still be achieved.
In this paper I have sought to identify some problems with their strategy.
It might be useful explicitly to distinguish between a ‘reformulation’ of
Transfer NR and a ‘restriction’ of it. Both McKenna and Stump have com-
mended a reformulation of Transfer NR so that it only applies to one-path
cases. I have argued that such a principle does not yield the general incom-
patibilistic result, and that it is implausible insofar as it has the ‘two-wrongs-
make-a-right’ problem. Further, I pointed out that, insofar as it is also
understood so that it requires a causally deterministic context, then it issues
in a dialectical stalemate; this provides an additional reason why it cannot be
employed in an argument for incompatibilism, and also a reason why it
need not be accepted as valid. Finally, Stump commends (as one horn of her
dilemma) that Transfer NR be restricted in its application to contexts of causal
determinism; here, again, this restriction issues in a dialectical stalemate.
Restriction shares this latter problematic feature with reformulation.
In her more recent paper on this subject, Stump reminds her reader that
Ravizza and I criticize van Inwagen for focusing exclusively on one-path
cases (Stump 2002, p. 42). We pointed out that among the two-path cases
there is a subclass in which one of the paths to the outcome contains the
sort of control that grounds moral responsibility. Stump says, “The ques-
tion for Fischer and Ravizza, then, is whether causally determined deci-
sions can be shown to belong to the relevant subclass of two-path cases”
(ibid.). Stump goes on to argue that causally determined decisions cannot
be shown to be part of the relevant subclass of cases.
I do not see why causally determined decisions would need to be shown
to belong to the sub-class of two-path cases to which Stump points. Rather,
it would seem that they need to be shown to be analogous in relevant ways
to these two-path cases. In the two-path cases in question, there is a path
to the outcome along which there are no uncontroversially responsibility
undermining factors. (For example, perhaps there is Betty’s unimpaired
practical reasoning.) Similarly, in the one-path causally deterministic case,
I have argued that there need not be any factor that is uncontroversially
responsibility undermining. For example, Betty’s undistorted practical rea-
soning might issue in the decision in question, and the subsequent action.
The fact that the actual flow of events here is causally deterministic is not
in itself a factor that uncontroversially undermines moral responsibility, and

it cannot be assumed to crowd out Betty’s unimpaired deliberation. If it is
taken to do so, then (as I argued above), a modal principle such as Transfer
NR is rendered entirely supernumerary. Thus, certain causally deterministic
cases are analogous in the relevant ways to the indicated subclass of two-
path cases.
I have contended that in the relevant two-path cases and in certain
causally deterministic one-path cases there are no uncontroversially res-
ponsibility undermining factors. This is all that I need for my purposes in
this paper. I acknowledge however that there is a gap between this point
and the further point that present in these cases is a sort of control that
grounds moral responsibility. I do wish to say that the sort of control
that grounds ascriptions of moral responsibility can be present even in a ca-
usally deterministic sequence, but I certainly concede that the considera-
tions invoked in this paper do not in themselves establish this result. If,
however, I am correct about this further point, then this would show why
all of the examples typically invoked by incompatibilists seem—but do not
really—establish a transfer principle that could yield incompatibilism: they
have carefully selected examples (such as Snake Bite and Tsunami) in which
there is no path to the actual result that contains the appropriate sort of
control. But even in a causally deterministic world, the actual sequence of
events can contain such control, as I have argued elsewhere.11 In my view,
the relevant sort of control consists in behavior that issues from the agent’s
own, suitably reasons-responsive mechanism.
Consider, finally, Thomas Nagel’s well-known skeptical worries about
agency and moral responsibility in a causally deterministic world first ex-
pressed in his essay, “Moral Luck”:
The area of genuine agency, and therefore of legitimate moral judgment, seems to
shrink under this scrutiny [from a perspective assuming causal determinism] to an
extensionless point. Everything seems to result from the combined influence of
factors, antecedent and posterior to action, that are not within the agent’s control.
Since he cannot be responsible for them, he cannot be responsible for their results—
though it may remain possible to take up the aesthetic or other evaluative analogues
of the moral attitudes that are thus displaced. (Nagel 1979, 35)
Later, Nagel elaborates, saying, “The effect of concentrating on the influ-

ence of what is not under his control is to make this responsible self seem
to disappear, swallowed up by the order of mere events” (ibid., 36).
206 J. M. Fischer
Part of Nagel’s worry—perhaps the basic thrust of it—is that it is unclear

how human actions could be mere events, or persons mere things.12 This
is a deep challenge, whether or not causal determinism obtains, and I have
nothing particularly helpful to say about it. I simply wish to suggest that
the ruminations in this paper can suggest a way of conceptualizing at least
some—even if they are not the deepest—of the issues Nagel raises. Nagel
appears to rely upon a transfer of nonresponsibility principle when he says,
“Since he cannot be responsible for them [certain factors not within one’s
control], he cannot be responsible for their results.” I agree that it is an
effect of concentrating on what is not under one’s control that the scope
of our moral responsibility seems to disappear. For example, when one fo-
cuses on the glacier and the natural forces that start the avalanche, and the
way in which they lead to the destruction of the enemy camp, Betty’s re-
sponsibility is in danger of disappearing. But, as I have argued above, it
would be a mistake to concentrate exclusively on those factors out of
Betty’s control and the way in which they issue in the destruction of the
enemy camp; that path is completely consistent with the existence of a
path on which Betty’s agency and responsibility are robust.
There can be cases—such as Erosion*—in which factors entirely out of an
agent’s control and thus for which the agent is not even partly responsible
result in certain upshots in a manner that is also entirely out of the agent’s
control; yet, in these same cases, there can be a path to the relevant upshots
along which the agent exercises control. If one focuses exclusively on the
no-responsibility paths, the responsibility paths can seem to disappear, and
nonresponsibility can seem to transfer to the upshots. But it is a mistake—
a kind of metaphysical depression—selectively and exclusively to focus
on the negative. If there is robust control in the actual sequence leading to
an upshot, I cannot see why the agent should not be morally responsible
for it. In a two-path case, it clearly is a mistake to focus solely on the
nonresponsibility path, giving it hegemony and implicitly assuming that it
crowds out the responsibility-conferring features found on a separate path.
If there is just one path leading to the upshot in question, and this path
is deterministic, it may nevertheless be the case that on that path the agent
displays the distinctively human capacity for practical reasoning in a way
that is neither impaired nor distorted by factors uncontroversially thought
to rule out moral responsibility. This sort of one-path case is then rele-
vantly similar to the two-path cases discussed above. Even in a casually
deterministic context, the agent may act from his own, appropriately
reasons-responsive mechanism. Thus, in my view, in such a context the
agent may exhibit the characteristic sort of control—guidance control—
that grounds moral responsibility, even though he is not morally responsi-
ble for conditions prior to his birth, and he is not responsible for the fact
that those conditions issue in his behavior. In place of Nagel’s metaphysi-
cal depression, I have sought to lift our spirits (if offering the possibility of
freedom and responsibility could do this). Although it might be supposed
that my suggestion pertains more to Pandora than Pangloss, I nevertheless
propose that the glass is half full.
Acknowledgments
I have read versions of this paper at the University of California, Santa

Barbara; the Inland Northwest Philosophy Conference at the University of
Idaho, Moscow; and the University of Texas, San Antonio. I have especially
benefited from helpful comments by Nathan Salmon, David Robb, Anthony
Brueckner, Francis Dauer, Richard Glass, Kevin Falvey, Voula Tsouna, and
Harry Silverstein.
Notes
1. Fischer (1986, 60–61).
2. van Inwagen presented his principle, and the argument for incompatibilism that
employs it in van Inwagen (1980), reprinted in Fischer (1986, 241–249).
3. I am indebted to Nathan Salmon for pointing out that, if the connective in (2) is
interpreted as a material conditional, then Transfer NR is clearly valid. This is
because any counterexample would have to be a case in which someone is at least in
part morally responsible for q. It would seem to follow that someone is at least in
part morally responsible for ‘q or not-p’. Thus (2) must be false. So any putative coun-
terexample to the principle would be a case in which the second premise would have
to be false, and thus it would not be a genuine counterexample to Transfer NR. I
accept this point, but it also nicely shows why Transfer NR, so interpreted, cannot be
employed in a sound, uncontroversial argument for incompatibilism. This is because
in any “ordinary” context in which a compatibilist is inclined to say that an agent is
morally responsible for q and causal determinism is assumed to be true, the compat-
ibilist will point out that (2) is false, given that the connective is interpreted as a
material conditional. So (given the dialectical situation) the incompatibilist will not
be entitled to apply Transfer NR to get to the incompatibilistic conclusion.
208 J. M. Fischer
4. We consider such a principle in Fischer and Ravizza (1998, 161–163); also, for the
formulation of the principle, I am indebted to McKenna (2001).
5. McKenna credits Al Mele and Ish Haji for help in formulating this principle. I am
not certain that this formulation works to secure the desired restriction to one-path
cases, but I shall accept it for the purposes of this paper.
6. Fischer and Ravizza (1998, 167).
7. Of course, Transfer NR does not entail that in a context of causal overdetermina-

tion there is moral responsibility; rather, it fails to entail that there is no responsibil-
ity. This seems to me to be sufficiently problematic.
8. Stump (2002); she goes on to explore whether Ravizza and I have available to us
some reason to call the incompatibilist’s judgment into question, but she concludes
(lamentably for us) that there is no such reason.
9. McKenna also credits Michael Zimmerman for bringing to his attention the
dialectical difficulty in question.
10. As Michael Bratman has pointed out, there is an interesting parallel between my
methodology here and John Perry’s approach to the mind–body problem in Perry
(2001). Here Perry defends what he calls ‘antecedent physicalism’. By this he means
that he finds physicalism antecedently plausible and sees his job as blocking dualis-
tic challenges.
11. For a defense of this view, see Fischer (1994); and Fischer and Ravizza (1998).
12. He also articulates this worry in Thomas Nagel (1985, 126). Here he is explicit
that the challenge appears whether or not causal determinism obtains.
References
Fischer, J. 1986. “Introduction: Responsibility and Freedom.” In J. Fischer, ed., Moral

Responsibility. Ithaca, N.Y.: Cornell University Press.
. 1994. The Metaphysics of Free Will: An Essay on Control. Cambridge, Mass.:

Blackwell Publishers.
Fischer, J. and Ravizza, M. 1998. Responsibility and Control: A Theory of Moral Respon-
sibility. Cambridge: Cambridge University Press.
Frankfurt, Harry G. 1969. “Alternate Possibilities and Moral Responsibility.” Journal

of Philosophy 66: 829–839. Reprinted in Fischer (1986).
McKenna, M. 2001. “Source Incompatibilism, Ultimacy, and the Transfer of Non-

Responsibility.” American Philosophical Quarterly 38: 37–52.
Nagel, T. 1976. “Moral Luck.” Proceedings of the Aristotelian Society, supplementary

volume 50. Reprinted in T. Nagel, ed. 1979. Mortal Questions. Cambridge: Cambridge
University Press.
. 1985. The View from Nowhere. New York: Oxford University Press.
Perry, J. 2001. Knowledge, Possibility, and Consciousness. Cambridge, Mass.: MIT Press.
Smilansky, S. 2000. Free Will and Illusion. Oxford: Clarendon Press.
Stump, E. 2000. “The Direct Argument for Incompatibilism.” Philosophy and

Phenomenological Research 61: 459–466.
. 2002. “Control and Causal Determinism.” In S. Buss and L. Overton, eds.,

Contours of Agency: Essays in Honor of Harry Frankfurt. Cambridge, Mass.: MIT Press.
Van Inwagen, P. 1980. “The Incompatibility of Responsibility and Determinism.” In

M. Bradie and M. Brand, eds., Bowling Green Studies in Applied Philosophy 2: 30–37.
Bowling Green, Ohio: Bowling Green State University Press. Reprinted in J. Fischer
(1986): 241–249.

The Compatibility Problem
10 Van Inwagen on Free Will
Peter van Inwagen
I can remember very clearly the first time van Inwagen encountered the
problem of free will. In the autumn of 1965 he was talking with a fellow
graduate student at the University of Rochester, one Myles Brand, and made
some remark that presupposed the incompatibility of free will and deter-
minism. Brand told him—second-year graduate student to first-year gradu-
ate student—that most philosophers believed that free will and determinism
were compatible, and outlined some of the currently popular arguments for
that position. As Athena from the head of Zeus, the argument that van
Inwagen was to publish ten years later in “The Incompatibility of Free Will
and Determinism” (1975) sprang from his head pretty much full-grown—
although it made its entrance into the world by way of his ever-active mouth
and not by Athena’s rather more unorthodox route.
The argument had it roots in the following reflections. If free will and de-
terminism coexist, then someone is able to do something not contained in
that one possible future that is consistent both with the past and the laws of
nature. Suppose that Alice, an inhabitant of a deterministic world, is able to
do something she is not in fact going to do; suppose, to be specific, that
although she is going to remain a prisoner, she is able to escape from her
prison. Her ability to escape can be looked upon as an ability to divert the
river of coming events into a channel through which it is not in fact going
to flow; to realize, that is, or to cause to be actual, a possible future that is not
the future that lies before her, to cause to be actual one of those possible
futures in which she escapes. And what would these possible but nonactual
futures be like? Let’s say that the past of a possible but nonactual future is
that possible past that would be the actual past if that future were the actual
future. Any given future in which Alice escapes must either be a future
whose past is the actual past and which is discontinuous with that past, or a
214 P. van Inwagen
future that is continuous with its past but whose past is not the actual past
(a past different from the actual past all the way back to the Big Bang), or,
finally, a future in which the world is governed by laws of nature that are not
the actual laws. But Alice can bring about futures of none of these sorts. If it
is insufficiently evident as it stands that she can bring about none of these
futures, here is an argument. Let ‘Clio’ be a proper name for the actual past—
thus, when I imagine, as I am about to, Alice using the name ‘Clio’ in an-
other possible world, I imagine her referring to the past as it is in our world,
not the past as it is in hers. Similarly, let ‘Nomos’ be a proper name for the ac-
tual laws of nature (which, remember, we are assuming to be deterministic).
It would seem that if Alice is able to escape, she must be able in some sense
to cause the actuality of or bring about or realize a future in which she could
say, and in so saying speak truly, one of the following three things:
There has been a causal break; the present state of affairs is not continu-
ous with the past.
The present is continuous with the past, but that past is not Clio; it is
some other past (a past different from Clio all the way back to the Big Bang).
The laws of nature are not Nomos, but some other set of laws.
And, obviously, Alice is not able to get herself into a future in which she
can say any of these things and be right.
When van Inwagen got around to writing down the argument that had
occurred to him in his conversation with Brand—he first did this in a doc-
toral thesis he wrote under the supervision of Richard Taylor (de jure) and
Keith Lehrer (de facto)—the argument he wrote down did not look much
like the argument I have just set out. (It looked very much like the argu-
ment he would later publish in “The Incompatibility of Free Will and
Determinism.”) Nevertheless, the central idea of both arguments was the
same, and they no doubt stand or fall together.
When van Inwagen had got his first academic job and was trying to publish
this argument, he found it extraordinarily hard to do so. The reason was
simple: the conclusion of the argument was known to be false. The unanimous
position of the referees whose reports were enclosed with the rejection letters
could have been expressed in the following words, which I take from an essay
by Donald Davidson, words as well known as they are ungrammatical:
I shall not be directly concerned with such arguments [i.e., arguments for the incompat-
iblity of free will and determinism], since I know of none that is more than superficially
Van Inwagen on Free Will 215
plausible. Hobbes, Locke, Hume, Moore, Schlick, Ayer, Stevenson, and a host of others
have done what can be done, or ought ever to have been needed, to remove the confu-
sions that can make determinism seem to oppose freedom. (1973, 137)
It’s not that no one was an incompatibilist in those days, of course. The
volume that contained the essay from which my quotation from Davidson
is taken also contained David Wiggins’s “Toward a Responsible Libertari-
anism.” Roderick M. Chisholm was an incompatibilist, as were Richard
Taylor, Carl Ginet, Elizabeth Anscombe, and Peter Geach. Still, it can hardly
be denied that incompatibilists were thin on the ground in the sixties and
early seventies.
Eventually, however, van Inwagen was able to publish two papers in
which he argued for the incompatibility of free will and determinism. (He
has always suspected that “The Incompatibility of Free Will and Determin-
ism,” which had been rejected by many journals, was accepted by Philo-
sophical Studies only because Sellars chose Keith Lehrer as its referee, but he
has never known whether this suspicion was correct.) His other paper on
the subject, “A Formal Approach to the Problem of Free Will and Deter-
minism” (1974), had, in respect of publication, the advantage of making
use of the then new and exciting apparatus of possible worlds. (This paper
was strongly influenced by Montague’s “Deterministic Theories”(1974); the
impetus for its writing was a suggestion of Rolf Eberle’s.)
Van Inwagen summed up his thought on free will in his book An Essay on
Free Will (1983), and has pretty much avoided learning anything about the
problem since—other than by sitting about and thinking it over. (The pub-
lication of this book by Oxford University Press was due to the good offices
of Tony Kenny and Derek Parfit, neither of whom could have had any
sympathy whatever with its content, and van Inwagen has always been
grateful to them for their generosity and has tried to imitate it. He has done
his best to learn from them the hard lesson that a philosophical book that
he regards as thoroughly wrongheaded can nevertheless be a good book.)
Van Inwagen likes to think that this book bears a significant share of the re-
sponsibility for the fact that incompatibilists are now much more common
than they were thirty or forty years ago. In a paper that he read at a con-
ference in the early nineties, van Inwagen made a remark to the effect that
compatibilism was the standard view among philosophers. Michael Slote,
who was in the audience, said that he thought that, on the contrary, in-
compatibilism had become the standard view, or at least the majority view.
216 P. van Inwagen
A few years later, van Inwagen asked Ted Warfield whether he thought that
was right. Warfield, who comes as close as is humanly possible to knowing
what every analytical philosopher thinks about anything, replied that he
thought that the majority of analytical philosophers who had actually
worked on the free-will problem were incompatibilists, and that the major-
ity of analytical philosophers (full stop) were compatibilists.
There was one passage in An Essay on Free Will that, after the publication
of the book, van Inwagen became more and more worried about.1 (These
worries were aggravated by pointed questions from Alex Rosenberg, and
also, curiously enough, by a science-fiction novel by Larry Niven called
Protector—a novel about which van Inwagen is ready to tell you consider-
ably more than you want to know.) Van Inwagen had said in that passage
that although no one was able to render a physical law false, it seemed that
if there were psychological laws, and if we had free will, we had to be able to
render these laws false. But that raises the question: if one has it within
one’s power to render some proposition false, in what sense can that propo-
sition be a law? As a sort of schematic example of a psychological law, van
Inwagen proposed, “No one who has received moral training of type A in
early childhood ever spreads lying rumors about his professional colleagues
in order to advance his career.” He imagined someone asking, “Why does the
pattern of behavior described in this statement occur if people don’t have to
conform to it?” He answered this question as follows: “Perhaps it is just the
people who have received moral training of type A in early childhood who
see the point in not spreading lying rumors about their colleagues.”
He gradually came to see, or to think he saw, that this response to the dif-
ficulty was facile, and that the difficulty he had his finger on was broader
and deeper than the original puzzle about psychological laws. He gradually
came to the conclusion that if one was faced with the necessity of doing
either A or B, and that if one saw every reason to do A and no reason what-
ever to do B, then one would simply not be able to do B. From this
conclusion it was no great leap to the slightly stronger conclusion that, if
one was faced with a choice between A and B, and one was aware of con-
siderations that could be brought in support of both alternatives, and if the
considerations that supported A seemed to one clearly and decisively to
outweigh the considerations that supported B, then one would simply not
be able to do B. Van Inwagen defended, in “When Is the Will Free?” (1989),
the thesis that the general principles about ability that lead philosophers to
incompatibilism should lead anyone who accepts them to accept these

conclusions as well. And he went on to argue that, since occasions that
call for serious deliberation—occasions, that is, on which one is choosing
between alternatives and it does not seem to one that (once all the purely
factual questions have been settled) the reasons that favor either alternative
are clearly the stronger—at best only a small proportion of the occasions on
which we make a choice are occasions on which we make a free choice.
(‘At best’ because there may be no free will at all; perhaps determinism is
true, or perhaps—as Broad believed—free action is incompatible both with
determinism and indeterminism and is therefore a self-contradictory idea.)
Van Inwagen concluded that no action is free unless it is the outcome of
deliberation in which one considers reasons that support that act, reasons
that support various alternative acts, and in the course of which one finds
no obvious answer to the question, ‘Which set of reasons should prevail?’
To take one example among many different sorts of possible example of the
consequences of this position, if you answer the telephone ‘automatically’,
if you answer the telephone without so much as considering the question
whether you should answer it, your act is not a free act: you could not have
done otherwise than answer the telephone; you were not able to let it ring
till it fell silent; it was not within your power not to raise the receiver.
After presenting arguments for this thesis, van Inwagen went on to at-
tempt to show that it does not, or does not obviously, have a certain unto-
ward consequence that it might be thought to have. He contended that from
the premise that at best a very small proportion of our acts are free acts, the
conclusion does not follow that only a very small proportion of our ascrip-
tions of moral responsibility are correct. For, he maintained, although there
is an inseverable connection between free will and moral responsibility, this
connection, inseverable though it be, can be stretched exceedingly fine. An
example will illustrate his point. Suppose a man is driving drunk and that a
pedestrian suddenly looms before him. He attempts to swerve, but too late:
he hits and kills the pedestrian because his reflexes are impaired by alcohol.
Compare his case with the case of a sober, able, and alert driver whose car
strikes and kills a pedestrian in circumstances in which swerving in time to
avoid the pedestrian would have required a reaction time smaller than that
allowed by the speed of propagation of human neural impulses. In neither
case was the driver able to avoid hitting the pedestrian who suddenly
loomed before him, but when we consider the former case, we hold the
218 P. van Inwagen
driver morally responsible for the pedestrian’s death, and when we consider
the latter case we do not.
The relevant difference is, of course, that the man whose reflexes were
impaired by drink, was, so to speak, able to avoid being unable to avoid hit-
ting the pedestrian, and the sober and alert driver was unable to avoid being
unable to avoid hitting the pedestrian. At the moment he first saw the
pedestrian, the drunk driver was unable to avoid hitting him, but he had
earlier been able, or so we should suppose if we were judge or jury, to avoid
driving with the impaired reflexes that were the cause of his fatal inability
at the time of the accident. Van Inwagen suggested that this sort of case
could serve as a model for the relation between ability and responsibility.
Here is a second case, a case in which the inseverable connection between
ability and responsibility, though it remains unsevered, as inseverable con-
nections do, is stretched considerably further than it is in most philoso-
phical examples concerning moral responsibility and the ability to do
otherwise. Consider a man who is, in middle age, a corrupt politician and
is, owing to his corrupted nature, unable to refuse bribes when he believes
there is no significant likelihood of the bribery coming to light. That is how
he is, but how did he get that way? Suppose the answer is this: as a young
man, he made a certain series of free choices, choices preceded by genuine
deliberation, which collectively had the effect of establishing him in set-
tled and unbreakable habits of venality. Van Inwagen argued—guided, I
suppose, by Aristotle—that this politician can properly be held morally res-
ponsible for the baleful effects on the public welfare of the informal ser-
vices he renders to his political cronies in return for money. And this despite
the fact that he is unable, in middle age, to reject the bribes he is offered. He
can properly be held responsible for, say, the deaths of the four children in
the fire in the building that wasn’t up to code, because he could, as a young
man, have avoided becoming the sort of man who would be unable to resist
the bribe offered by the slumlord who owned the building.
Several philosophers have disputed van Inwagen’s conclusion that the
principles that lead philosophers to incompatibilism entail that free acts, if
they exist at all, are extremely rare, but van Inwagen has never been able to
see any force in their arguments. Although he has published answers to
them (1994), he is of the opinion that no answers were needed; that his
original arguments were untouched by the arguments of his critics. One
philosopher, who generally disagrees with van Inwagen about free will,
Dan Dennett, agrees with van Inwagen that these arguments are unan-
swerable. As Dan put it, referring to “When Is the Will Free?”, “Thank you,
Peter, for the lovely reductio of incompatibilism.”
Now van Inwagen’s arguments for this conclusion, whether they are
good or bad, presuppose that there is an inseverable connection between
moral responsibility and the power to do otherwise, however flexible this
connection may be. The inseverable connection is this: if one is morally
responsible for anything, it follows logically that one has had a free choice
about something. But Harry Frankfurt has presented a famous argument
that some have taken to refute this thesis.2 A significant proportion of van
Inwagen’s work on free will has been devoted to Frankfurt’s argument.
And Frankfurt’s argument is important. If it is indeed true that one might
be morally responsible for various things, despite one’s never having been
able to do otherwise than one has done, then the problem of free will
loses much of its interest—for the simple reason that most people would
find the thesis that we lack free will much less unappealing if this thesis
could be shown not to entail that we can never be held to moral account
for anything.
I have said that Frankfurt’s argument has been taken by some to show
that it is possible for one to be morally responsible for something even
though one has never been able to do otherwise. The actual conclusion of
Frankfurt’s argument, however, is this: the so-called principle of alternative
possibilities is false, or at least not a necessary truth. (I’ll call it the princi-
ple of alternative possibilities. Frankfurt has recently presented an ill-
advised defense of his use of the adjective ‘alternate’ in his name for the
principle. It is, I concede, uncharitable of me to mention this. I’ll attempt
to atone for my lapse by very charitably saying nothing further about it.)
This is the principle of alternative possibilities:
A person is morally responsible for what he has done only if he could

have done otherwise.
Van Inwagen has always thought that Frankfurt’s argument—which, of

course, consists in the presentation of a certain sort of counterexample to the
principle of alternative possibilities—has a great deal of force and has never
been shown conclusively to be mistaken. His position has never been that
Frankfurt’s proposed counterexamples to the principle of alternative possi-
bilities fail; his position has been rather that even if these counterexamples
220 P. van Inwagen
succeed, even if the principle of alternative possibilities is false, the existence

of moral responsibility nonetheless requires the existence of free will.
I have used the qualified phrases ‘has a great deal of force’ and ‘has never
been shown conclusively to be mistaken’ because, although van Inwagen is
inclined to think that Frankfurt’s counterexamples show that the principle
of alternative possibilities is false if it is meaningful at all, he’s also inclined
to think that it’s meaningless. I’ll briefly say something about why he’s
inclined to think it’s meaningless. If the principle of alternative possibili-
ties is meaningful, the following must be one of the particular statements
the general principle endorses:
Bill is morally responsible for lying under oath only if he could have
done otherwise.
Van Inwagen has never been able to convince himself that he understands
sentences like ‘Bill is morally responsible for lying under oath’. It has al-
ways seemed reasonably plain to him that what one is morally responsible
for is not one’s acts but the consequences of one’s acts, or, more exactly, certain
of the consequences of one’s acts—for no one would suppose that one
could be responsible for all the consequences of one’s acts. (When I say ‘it
has always seemed plain to him’, I mean that it has seemed plain to him
since he first encountered the idea that what a person is morally responsi-
ble for is the consequences of his acts and not the acts themselves. Van
Inwagen is sure that he first encountered this idea in something written by
P. H. Nowell-Smith; he cannot now remember where it was that Nowell-
Smith made this valuable point.) And the consequences of one’s acts, it
would seem, are members of the same ontological category or categories as
the consequences of anything that takes place within the causal and tem-
poral order; whatever ontological category one thinks the consequences of
a person’s acts should be assigned to, one should assign them to the same
category or categories as the consequences of an earthquake or a scientific
discovery or a rise in the prime lending rate. There would seem to be two
serious candidates for this categorial office: ‘concrete event’ (for example,
Caesar’s death), on the one hand, and something proposition-like on the
other, ‘fact’, perhaps, or ‘state of affairs’ (for example, the fact that Caesar dis-
regarded the soothsayer’s warning, or Caesar’s having chosen to believe that
Brutus’ strong republican sentiments would never overcome his friendship
with and personal loyalty to Caesar).
Now if what one is responsible for is certain of the consequences of one’s

acts (and if, for example, one’s telling a lie and the fact that one has lied do
not count as ‘consequences’ of themselves), then it is doubtful whether
‘Bill is responsible for lying under oath’ makes sense, and it is therefore
doubtful whether the principle of alternative possibilities makes sense,
doubtful whether the sentence that formulates it means anything. Van
Inwagen, as I have said, has long doubted whether the principle of alterna-
tive possibilities does make sense. But he has also been fairly sure that if
he’s wrong about this and the sentence ‘A person is morally responsible for
what he has done only if he could have done otherwise’ does express some
proposition, the proposition it expresses is false, and that Frankfurt’s coun-
terexamples show this.
To recapitulate, van Inwagen thinks that (a) the principle of alternative
possibilities is either nonsensical or false, and that (b) moral responsibility
nevertheless requires free will—that if anyone is morally responsible for
anything, there must be something that person had a free choice about.
He has defended the latter thesis by presenting other principles than the
principle of alternative possibilities that have the consequence that moral
responsibility requires free will, and presenting arguments designed to show
that these other principles seem to be true and cannot be refuted by coun-
terexamples in the style of Frankfurt’s counterexamples to the principle of
alternative possibilities.3 One of these principles is
A person is morally responsible for a certain state of affairs only if
that state of affairs obtains and he could have prevented it from
obtaining.
(Or this was van Inwagen’s original formulation of one of these principles.
He later came to think that the principle needed to be revised. The revised
principle would look something like this:
A person is morally responsible for a certain state of affairs only
if that state of affairs obtains and there was a time at which he could
so have acted that that state of affairs not obtain.
The states of affairs quantified over in this principle are ‘proposition-like’:
the state of affairs ‘Caesar’s having been murdered’ obtains because certain
conspirators stabbed Caesar to death in Rome in 44 B.C.; but it, that very
same state of affairs, could have obtained because Cleopatra poisoned him
in Alexandria in 48—just as the proposition that Caesar was murdered is
222 P. van Inwagen
true because certain conspirators stabbed Caesar to death in Rome in 44 B.C.

and could have been true because Cleopatra poisoned him in Alexandria in
48. This state of affairs may thus be contrasted with the concrete event the
murder of Caesar, which would not have occurred if Cleopatra had poisoned
Caesar in Alexandria in 48, although, in that case, there would have been a
concrete event, which does not in fact exist, that would have been denoted
by the words ‘the murder of Caesar’. (He has also endorsed a principle
about concrete events that corresponds to or parallels the above principle
about states of affairs.)
Van Inwagen has never seen any need to rethink the position he took
concerning this principle when he first formulated it in the late seventies
(with this minor qualification: as I have said, he has come to prefer a
revised version of the principle), to wit that it is extremely plausible that it
entails that moral responsibility requires free will, and that it cannot be re-
futed by any adaptation of the counterexamples Frankfurt brought against
the principle of alternative possibilities. This last point has been disputed,
but van Inwagen has never been able to see any merit in the arguments by
means of which the point has been disputed. He has responded to them
(1997 and 1999), but, as far as he is concerned, his original arguments for
this position are the only answer to these counter-arguments that was
really needed.
This brings us down to fairly recent times. (Or it does if we ignore two
minor essays of the middle eighties. In one of these essays [1985a], van
Inwagen defended the position that the ‘conditional-analysis’ argument for
the compatibility of free will and determinism, and what he has dubbed the
Mind Argument for the incompatiblity of free will and indeterminism—the
latter argument is named for the journal, not for the human intellectual
faculty—are inconsistent; inconsistent in this sense: from the soundness of
either, the unsoundness of the other follows. In the other essay [1985b], he
proposed a conditional-analysis argument for the compatibility of free will
and indeterminism; this argument was, admittedly, a rather absurd argu-
ment; his point was that the unbiased inquirer should see that the condi-
tional analysis argument for the compatibility of free will and determinism
was equally absurd, and absurd for an exactly parallel reason.) Van Inwagen
has thought little about free will in the last ten years, but he has thought
enough about the topic to have changed his mind on one point and to
have become increasingly insistent on another.4
The point on which he has changed his mind is this: he now thinks that
Rule ␤, as he presented it in An Essay on Free Will and other places, is wrong.
This rule of inference was stated as follows:
p, and no one has, or ever had, any choice about that. If p then q, and
no one has, or ever had, any choice about that. Hence, q, and no one
has, or ever had, any choice about that.
Van Inwagen had thought that this rule was obviously valid, had made use
of it in formulating one version of his argument for the incompatibility of
free will and determinism (his favorite version, the version inspired by Carl
Ginet’s “Might We Have No Choice?” [1966]), had said that, despite its
seeming obviousness, it was nevertheless the weakest link in the chain of
reasoning that led from the assumption of determinism to the conclusion
that no one is ever able to do otherwise. And, although this principle is
used in only one of the three formal versions of the argument he has
presented, van Inwagen is on record as saying that, in his opinion, if Rule
␤ should turn out to be invalid, it would almost certainly be the case that
the two versions of the argument that do not involve the concept ‘having
a choice’ would also turn out to be invalid.5
Imagine, then, his embarrassment when Tom McKay and David Johnson
(1996) presented a counterexample to Rule ␤—a ‘beta blocker’, to use the
term that some wag (was it John Martin Fischer?) coined for counterexam-
ples to Rule ␤. (Actually, McKay and Johnson presented a counterexample
to a different rule of inference, but since the invalidity of that other rule
implies the invalidity of ␤, they have for all practical purposes presented a
counterexample to ␤. It is not, by the way, entirely clear why they pro-
ceeded in this indirect fashion, since the idea behind their counterexample
to the other rule of inference can easily be adapted to produce a coun-
terexample to ␤.6)
Van Inwagen (2000, 8–9) has presented a revised rule of inference that is
immune to McKay–Johnson-style counterexamples. Finch and Warfield
(1998) have proposed a different sort of revision that is also immune to
these counterexamples, and McKay and Johnson themselves suggested
some revisions that are, I think, workable. Unfortunately, these revised
rules, although they are far from implausible, lack the ‘luminous evidence’
that was a striking, albeit illusory, property of the original Rule ␤. And this
fact can very properly prompt the compatibilist to ask a pointed question
224 P. van Inwagen
along the following lines: “If the apparent intuitive obviousness of the orig-
inal Rule ␤ turned out to be an illusion—since the rule has in fact turned
out to be invalid—how much confidence should we have in the revised ver-
sions? Nemesis, in the form of a counterexample, was all along lying in
wait for Rule ␤; why should we not take very seriously the epistemic possi-
bility that in some dark corner of logical space, cousins of this Nemesis
patiently await their appointments with the revised rules?” This pointed
question is pointed indeed. Van Inwagen admits that he has no good
answer to this question, and that, in consequence, although he accepts the
revised rule, he assigns its validity a rather lower subjective probability
than the near certainty he once so confidently assigned to the validity
of the original ␤. He does, however, withdraw his assent to a thesis I men-
tioned a moment ago, the thesis that if Rule ␤ should turn out to be invalid,
this would mean that there was almost certainly something wrong with the
other arguments he has given for the incompatibility of free will and
determinism. He withdraws his assent because the counterexamples that
have shown the invalidity of Rule ␤, the McKay–Johnson counterexamples,
depend on some unexpected properties of an English phrase—“has a
choice about”—that played a key role in his formulation of the original
principle. He is inclined to think that the ‘general idea’ behind Rule ␤ was
sound, and that its invalidity stemmed from the fact that certain features of
the English phrase unfit it for the task he assigned to it. When he made the
statement from which he has now withdrawn his assent, he was assuming
that if Rule ␤ were shown to be invalid, this would be because someone had
shown that the general idea behind the principle was fundamentally de-
fective. But the revised versions of the rule appeal to this same general idea;
they are merely (it is hoped) adequate implementations of this idea, imple-
mentations from which a technical defect has been removed.
I have said that in recent times, van Inwagen has changed his mind about
one point and has become increasingly insistent on another. I have dis-
cussed the point on which he has changed his mind. The point on which
he has become increasing insistent is this: free will is a mystery, a ground-
floor, first-water, Colin McGinn–style philosophical mystery. Free will is a
mystery because, although it obviously exists—of course we sometimes
confront a choice between A and B and are, while we are trying to decide
whether to do A or to do B, able to do A and able to do B—it seems to be in-
compatible with both determinism and indeterminism, and thus seems to
be impossible. When he says that free will seems to be incompatible both

with determinism and indeterminism, van Inwagen means that there are
good arguments for the incompatibility of free will and determinism and
good arguments for incompatibility of free will and indeterminism, and
that no one has ever identified a very plausible candidate for the flaw in any
of the arguments in either class. Van Inwagen, of course, believes that the
arguments he has given for the incompatibility of free will and determin-
ism contain no flaws—or, at the worst, contain minor, technical flaws that
could be repaired without altering their essential points—and that there is
some flaw, or are some flaws, in the familiar arguments for the incom-
patibility of free will and indeterminism. But as to the latter class of
arguments—well, he’s damned if he knows what the flaws in them might
be. He simply hasn’t a clue. If you ask him how free will is possible, he can
say only, “That’s a mystery.” This is, of course, a situation that a philoso-
pher must regard as unsatisfactory, but van Inwagen has two consolations.
First, he thinks that if he believed that an undetermined act could not be a
free act, he’d be forced to believe something even more mysterious than
what he now believes. He’d be forced to believe that at least one of the
following three propositions was false:
Free will is incompatible with determinism

Some people are morally responsible for various things, but no one has
free will
People are sometimes morally responsible for various things.
And the falsity of any of these propositions would be, in his view, an even
greater mystery than the falsity of the proposition that free will is compat-
ible with indeterminism. Second, there are lots of philosophical mysteries,
most of which have nothing to do with free will. I might cite one or more
of the great philosophical mysteries, such as ‘temporal passage’ or human
consciousness. But I will mention instead two minor mysteries that are no
less mysterious for being minor. I will mention two mysteries that carry
much less emotional weight than the mystery of free will, or are, as one
might say, unconnected with anything we care deeply about—unlike free
will and time and consciousness, which are connected with many things
that we care deeply about: the mystery of vagueness and the mystery of the
Liar. It is obviously true that a currently living American male who is seven
feet four inches tall is tall, and is tall without qualification (i.e., is not
226 P. van Inwagen
simply tall to degree 0.99972 or something like that). It is obviously true

that the president-elect, falsely believing himself to have lost the election,
might address an audience and say, bitterly and sincerely, “Everything the
president-elect tells anyone today is false.” I do not think that either of
these obvious truths is free from apparent paradox—although, of course,
no paradox can be more than apparent, for no truth can really imply a self-
contradictory statement or even a false statement. And yet there are very
good arguments for the conclusion that each of these obvious truths entails
various statements that are self-contradictory or at least obviously false.
And I cannot find any flaw in these arguments, though, of course, there
must be flaws in them. Van Inwagen’s expectation is that some premise (or
more than one) in the several arguments for the incompatibility of free will
and indeterminism is false but seems to us to be an obvious truth—as was
once the case with the proposition that there are twice as many natural
numbers as there are even numbers, the Galilean law of the addition of
velocities, the argument that, since space is unbounded, it must be infinite,
or the unrestricted comprehension principle in set theory. But if van
Inwagen is wrong about this, the following is certainly the case: some propo-
sition (or maybe there is more than one) about matters relating to free will,
determinism, and moral responsibility that seems to us to be obviously true
is false. Perhaps we shall one day discover what proposition this is, or what
propositions these are, and it, or they, will come to have no more appeal
to philosophers than the proposition that there are twice as many natural
numbers as there are even numbers (which Galileo thought evident, de-
spite his awareness that the two classes of numbers could be put into
one–one correspondence) has for us. Or perhaps, as Chomsky (1988,
151–152) and McGinn (1993) have suggested, some evolutionarily contin-
gent feature of the design of the human intellect renders it biologically
impossible for us to think our way through the free will problem to a satis-
factory conclusion. I will remark, because it makes a nice link with my next
topic, that van Inwagen has strong, I might almost say fraternal, feelings
for Chomsky when he considers Chomsky’s position on free will and
determinism. Chomsky thinks that free will must be compatible with
determinism, and that, nevertheless, it is a mystery how this could be. Van
Inwagen thinks that free will must be compatible with indeterminism, and
that, nevertheless, it is a mystery how this could be. Van Inwagen’s feeling
is that Chomsky, although he is mistaken about a particular point (an
important point, to be sure), appreciates the depth and difficulty of the free
will problem, while the majority of van Inwagen’s fellow incompatibilists,
although they are right on a particular point (and a very important point it
is), do not really appreciate the depth and difficulty of the problem.
I want to close by explaining why van Inwagen thinks one important
group of incompatibilists, those who appeal to what is called agent causa-
tion, do not appreciate the depth and difficulty of the problem of free will.
Many philosophers would agree with this judgment for the simple reason
that they think that the concept of agent causation is incoherent, or
think that agent causation is metaphysically impossible. Van Inwagen is in-
clined to agree with them (although he has no firm opinion on this ques-
tion), but he has lately stressed a different point. It is this: suppose there is
nothing conceptually or metaphysically impossible about agent causation;
suppose in fact that agent causation is a real phenomenon and that an
episode of agent causation figures among the antecedents of every volun-
tary movement of a human hand or limb or vocal apparatus. Van Inwagen’s
position is that even if this is so, and even if (as some have argued) we un-
derstand the concept of agent causation at least as well as we understand the
concept of event causation, all this does nothing to diminish the mystery of
free will. I will try to explain why van Inwagen thinks this by considering a
particular human action. Suppose Marie wants to vote in favor of the pro-
posal before the meeting, and that, for this reason, she raises her right hand
when the chair says, “All in favor . . .?” Suppose that one of the causal an-
tecedents of her hand’s rising was a certain event in her brain that was
undetermined by past events, that the state of her body and her immediate
environment at the moment this brain event occurred was causally suffi-
cient for her hand’s rising, that if this event had not occurred, her hand
would not have risen, and that she, Marie, a particular member of the meta-
physical category ‘substance’ or ‘continuant’, was the cause—that is to say,
the agent cause—of that crucial brain event. The friends of agent causation,
if van Inwagen understands them, believe that these suppositions are suffi-
cient for her having freely raised her hand. If that is so, these suppositions
must entail the following proposition: at some moment shortly before
Marie raised her hand, she was able to raise her hand and she was able not
to raise her hand. But van Inwagen doesn’t see why this entailment should
be supposed to hold. In fact, he thinks he sees a good argument for the con-
clusion that it was not up to her whether her hand rose. Suppose God were
228 P. van Inwagen
miraculously to return the world to precisely the state it was in, say, one
minute before Marie raised her hand, and that he then allowed affairs
once more to proceed, without any further miracles. What would happen?
What would Marie do? Well, if her raising her hand was a free act, and if free
will is incompatible with determinism, then we can’t say. We can say only
that she might have raised her hand and might not have raised her hand. If
God were to cause this episode to be thus “replayed” a very large number of
times, it might turn out that she raised her hand in thirty percent of the
replays and refrained from raising it in seventy percent of the replays. This
much is a simple consequence of incompatibilism, and it brings one of the
main reasons philosophers become compatibilists into stark relief. It seems
to lead us inescapably to the conclusion that on each particular replay, what
Marie does on that occasion is a mere matter of chance. And if there are no
replays, if there is only one occasion on which Marie is in this situation, it
seems to lead us just as inescapably to the conclusion that on that one occa-
sion what Marie does is a mere matter of chance. And if it is a mere matter
of chance whether Marie raised her hand, then it cannot have been true be-
forehand that Marie was both able to raise her hand and able to refrain from
raising her hand, for to have both these abilities would be to be able to de-
termine the outcome of a process whose outcome is due to chance. It is true
that we have, by stipulation, inserted into this process, this process whose
outcome is due to chance, an episode of agent causation. But, if I may so ex-
press myself, so what? That doesn’t change the fact that the outcome of that
process was due to chance. If God caused Marie’s decision to be replayed a
very large number of times, sometimes (in thirty percent of the replays, let
us say) Marie would have agent-caused the crucial brain event and some-
times (in seventy percent of the replays, let us say) she would not have.
Surely, then, whether she agent-caused the brain event was a mere matter of
chance? Whether her deliberations were followed by her agent-causing the
brain event was, it would seem, a matter of chance; Marie, therefore, cannot
have been both able to agent-cause the brain event and able to refrain from
agent-causing the brain event, for to have both these abilities would be to
be able to determine the outcome of a process whose outcome was due to
chance—an impossible ability. I conclude that even if an episode of agent
causation is among the causal antecedents of every voluntary human ac-
tion, these episodes do nothing to undermine the prima facie impossibility
of an undetermined free act. Postulating agent causation, therefore, does
nothing to diminish the mystery of free will. Van Inwagen’s conclusion is

that incompatibilists had better abandon the concept of agent causation,
and seek a resolution of the mystery of free will elsewhere—if, indeed, there
is an ‘elsewhere’.
Notes
1. See van Inwagen (1983, 63–64).
2. See his classic essay, Frankfurt (1969).
3. Van Inwagen (1978). A somewhat condensed version of the arguments of van

Inwagen (1978) can be found in van Inwagen (1983, 161–182).
4. See van Inwagen (2000).
5. At any rate, he is on record as saying things that commit him to this position. In
van Inwagen (1983, 57), he said, “I am quite sure that any specific and detailed
objection to one of the arguments can be fairly easily translated into specific and
detailed objections to the others; and I think that any objection to one of the argu-
ments will be a good objection to that argument if and only if the corresponding
objections to the others are good objections to them.” One of the arguments referred
to as “the arguments” in this quotation is the argument that explicitly appeals to
Rule ␤; and he has said (van Inwagen 1989, 405) that if that argument is unsound,
its unsoundness must be due to the invalidity of Rule ␤. And he has said (van
Inwagen 1994, 95), “My position is that all (logically adequate) arguments for incom-
patibilism must make some sort of implicit or hidden or covert appeal to [Rule ␤].”
6. See van Inwagen (2000, 5–6).
References
Chomksy, N. 1988. Language and Problems of Knowledge. Cambridge, Mass.: MIT

Press.
Davidson, D. 1973. “Freedom to Act.” In T. Honderich, ed., Essays on Freedom of

Action. London: Routledge and Kegan Paul.
Finch, A. and T. A. Warfield. 1998. “The Mind Argument and Libertarianism.” Mind
107: 516–528.
Frankfurt, H. G. 1969. “Alternate Possibilities and Moral Responsibility.” Journal of

Ginet, C. 1966. “Might We Have No Choice?” In K. Lehrer, ed., Freedom and

Determinism. New York: Random House.
230 P. van Inwagen
McGinn, C. 1993. Problems in Philosophy: The Limits of Inquiry. Oxford: Blackwell.
McKay, T. and D. Johnson. 1996. “A Reconsideration of an Argument against

Compatibilism.” Philosophical Topics 24: 113–122.
Montague, R. 1974. “Deterministic Theories.” In R. H. Thomason, ed., Formal Philos-

ophy: Selected Papers of Richard Montague. New Haven, Conn.: Yale University Press.
van Inwagen, P. 1974. “A Formal Approach to the Problem of Free Will and
Determinism.” Theoria 40: 9–22.
. 1975. “The Incompatibility of Free Will and Determinism.” Philosophical

Studies 27: 185–199. Reprinted in G. Watson, ed., Free Will. New York: Oxford
University Press, 1983; in D. Pereboom, ed., Free Will. Indianapolis, Ind.: Hackett,
1997; in L. Waddell Ekstrom, ed., Agency and Responsibility: Essays on the Metaphysics
of Freedom, Boulder, Colo.: Westview Press, 2000.
. 1978. “Ability and Responsibility.” Philosophical Review 87: 201–224. Reprinted

in John Martin Fischer, ed., Moral Responsibility. Ithaca, N.Y.: Cornell University Press.
. 1985a. “On Two Arguments for Compatibilism.” Analysis 45: 161–163.
. 1985b. “Compatibilistic Reflections.” Australasian Journal of Philosophy 63:

349–353.
. 1989. “When is the Will Free?” In J. Tomberlin, ed., Philosophical Perspectives

3: Philosophy of Mind and Action Theory. Atascadero, Calif.: Ridgeview. Reprinted in
T. O’Connor, ed., Agents, Causes, and Events: Essays on Indeterminism and Free Will.
Oxford: Oxford University Press, 1995.
. 1994. “When the Will Is Not Free.” Philosophical Studies 75: 95–113.
. 1997. “Critical Notice: John Martin Fischer’s The Metaphysics of Free Will.”
. 1999. “Moral Responsibility, Determinism, and the Ability to Do Otherwise.”

Journal of Ethics 3: 341–350.
. 2000. “Free Will Remains a Mystery.” The Eighth Philosophical Perspectives

Lecture. In J. Tomberlin, ed., Philosophical Perspectives 14: Freedom and Action. Oxford:
Blackwell. Reprinted in R. Kane, ed., The Oxford Handbook of Free Will. Oxford: Oxford
University Press, 2002.
11 Compatibilist Options
John Perry
. . . those who accept that responsibility for a situation implies an ability to bring it
about and, perhaps, an ability to prevent it, must explain how agents are able to do
other than they are caused to do. Without it, they can give no defense of their coun-
terexamples. With it, they can be confident that the Consequence Argument, by
itself, is no refutation of their position.
—Tomis Kapitan (2002, 154)
1 Introduction
Compatibilism is the thesis that an act may be both free and determined
by previous events and the laws of nature. I assume that in normal cases
a condition of a person’s performing an act freely is that the person is able
to refrain from performing the act. Thus, I accept that if determinism en-
tails that agents do not have this ability, we must give up compatibilism.
In this paper I try to contribute to the rethinking of compatibilism by dis-
tinguishing between strong and weak accounts of laws and strong and
weak accounts of ability. I argue that compatibilism is a tenable position
when combined with either a weak account of laws, or a weak account of
ability, or both. I shall concentrate on influential arguments for incom-
patibilism due to Peter van Inwagen, often called collectively the ‘conse-
quence argument’.
Some versions of the consequence argument seem to rely only on in-
escapable modal principles. In his excellent review and discussion of these
arguments, Tomis Kapitan concludes that these principles are not so logi-
cally inescapable as to completely trap the compatibilist (Kapitan 2002).
This is not to say van Inwagen’s arguments are fallacious, simply that they
rely on certain principles about causation and ability that have not yet
232 J. Perry
been fully articulated and defended. Kapitan says, just before making the
remarks quoted above, that his assessment provides the compatibilist with
“momentary breathing room at best” (2002, 154). I am trying to take ad-
vantage of the momentary breathing room afforded by Kapitan to explore
and to a certain extent defend options available to the compatibilist.
Using terms I explain below, my position is that van Inwagen’s arguments
do show that the combination of compatibilism with a strong account of
laws and a strong account ability (as I define these terms) is not tenable.
The options, then, are a weak account of laws, a weak account of ability,
or both.
2 Basic Argument
As a preliminary to considering a simple argument against compatibilism,

let’s look at an even simpler argument.
(1) ∀t ∀x (␾(x, t) → ␸(x, t 1))

(2) ␾(Elwood, t)
(3) Elwood eats cookies at t 1 → ~␸(Elwood, t 1)
(4) So, ~(Elwood eats cookies at t 1)
Here ␾ is a complex state a person together with a suitably large surround-
ing region the person can be in at a time. If Elwood is in ␾ at t, then at t
states of Elwood and the things around him make it the case that he really
really does not want a cookie at t, that no one is about to persuade him to
change his mind, that there are no forces about to impinge on him that will
change his mind about cookies, or force him to eat one whether he wants
to or not, and so forth. For simplicity we suppose that time is discrete, and
t 1 is the next instant after t. ␸ is a state of Elwood and a suitably large
region that will succeed ␾ according to the laws of nature. It makes it the
case that Elwood does not have any cookies in his mouth.
(1)–(4) constitute a valid argument. If Elwood is in a state that invariably
leads to a state that precludes eating cookies, he will not eat cookies. How-
ever, it seems quite clear that we are not warranted in going further. To add
to our argument:
(5) What’s more, Elwood cannot eat a cookie at t 1
would turn it from a valid to an invalid argument.

Compatibilist Options 233
Prior to being given some quite persuasive argument, we do not take doing
something as a necessary condition of being able to do it, or being able to do
something as a sufficient condition for doing it. We do not take a person’s not
doing something as a sufficient condition for their being unable to do it. We ac-
cept the inference from ‘cannot’ to ‘does not’ and from ‘does’ to ‘can’. But we
do not accept the inference from ‘does not’ to ‘cannot’ or from ‘can’ to ‘does’.
If Elwood did eat a cookie at t 1, that would prove that one of the
premises 1, 2, or 3 is false. But the fact that he can eat a cookie does not
show that one of the premises is false. It merely shows that if he did do
that thing that he is capable of doing, one of the premises would be false.
Hence, the truth of the premises rules out his eating a cookie, but not his
having the ability to eat a cookie.
Now suppose we add another premise, to the effect that (1) is not merely
a true universal generalization, but something that follows from the laws of
nature. We derive (1) from premise (0):
(0) According to the laws of nature, ∀t ∀x (␾(x, t) → ␸(x, t 1))

Now we have the basic argument underlying those used by many incom-
patibilists, although there are many variations on the basic theme.
The intuitive idea is that premise (0) provides enough extra strength to
the premises to get not only to establish (4), that Elwood does not eat a
cookie, but to establish (5), that he cannot eat a cookie. Van Inwagen often
taps our intuitions that we cannot change the past or change the laws of
nature. The idea is that if Elwood can eat a cookie, then he can falsify one
of the premises. He cannot falsify any true statements about the past, so he
cannot do anything at t 1 that changes the fact that he was in state ␾ at
t. And he cannot falsify laws. So Elwood not only will not, but cannot eat a
cookie at t 1, if determinism is true.
Given this way of looking at the argument for incompatibilism, there is
one basic strategy for the compatibilist. This is to deny that the replace-
ment of (1) with (0) adds enough strength to the premises to validate the
step from (4) to (5). There are two basic (and compatible) ways to imple-
ment this strategy: (i) adopt a weak theory of laws, and (ii) adopt a weak
theory of ability, of can, and hence a strong theory of cannot. The compati-
bilist must maintain that it takes more extra power to rule out Elwood’s
being able to eat the cookie than supplementing (1) with (0) provides, by
adopting one or both of these positions.
234 J. Perry
3 Some Preliminaries
Discussions of compatibilism usually employ, at least implicitly, two differ-

ent concepts having to do with propositions and truth. Propositions are
true or false. And propositions are made true and false by actions and other
events. It will be helpful to be clear about these concepts, and how they are
related, before plunging into the main items of business.
Intuitively, many propositions are made true, or rendered true, by events
that occur at some time, or through some interval of time. For example, the
proposition that Nixon won the 1972 election was not made true or rendered
true until the events of election day, 1972, or perhaps not even until inau-
guration day, 1973. What was the status of this proposition before then?
One intuitive option is that propositions are not true or false until they
are made true or false by events. On this view, many of our statements about
the future express propositions that are neither true nor false when they are
made, but become true or false when events make them so in the future.
This option, though intuitive, does not mesh easily with the two-valued
logic that most of us are taught and find easy and convenient to work with.
If we want our familiar logic to mesh smoothly with the concept of
events making propositions true, the simplest way is to think of truth and
falsity themselves as timeless properties of propositions, while the proper-
ties of being rendered or made true or false are properties that occur
at times, or through intervals. ‘Timeless’ here simply means ‘not relative to
times’. So it doesn’t follow from the fact that a proposition is timelessly true
that it is true at every time in any interesting sense. We can say that I am six
feet tall because if I stand erect, one part of my body—the top of my head—
is six feet above the floor. It’s my whole body that is six feet tall. But I’m not
six feet tall at every point on my body, if that means anything more than
that I’m six feet tall.
So we have two properties having to do with truth of propositions: being
true or false, and being made true or false. The first is not relative to times,
the second is. This is the track I shall follow in this essay. All of the points
I make, however, could be made in the more complicated system, in which
some propositions have no truth value until they are made true by events.
How are the properties of being true and being made true related? The
obvious way is that:
if a proposition is ever made true, it is true.

It might be better to say it be true, using a tenseless form, and in fact I shall
do so from now on, and say things like
If a proposition is ever made true (or made false), then it be true (or be
false).
The fact that a proposition has not yet been made true (or made false) by
events, does not imply that it be not true (or be not false).
Compare:
The fact that Gore has not yet been chosen as our next President does
not imply that he is not our next president.
If Gore ends up being nominated and elected in 2004, then he is our next
president. If I call him ‘our next president’ now, I’m correct if the future
goes one way, incorrect if it goes the other. It is possible to become rather
puzzled by this. How can Gore be our next president now, if it hasn’t been
decided yet? So he must not be our next president. By parity of reasoning,
no one is our next president. That will be a constitutional crisis. But we can
avoid the crisis. Being our next president is a property Gore has if at some
point between now (summer, 2002) and January 2005 he is elected and in-
augurated and Bush hasn’t been replaced in the meantime. It all works out,
as long as we are careful about the difference between being our next pres-
ident, a property Gore may have, and being elected and inaugurated as our
next president, one that he does not yet have as I write this, in 2002. The
fact that lots of propositions be true that have not yet been made true is
sort of like that. It can be confusing. It may well be that from a metaphysi-
cal point of view our two-valued logic of propositions may not be optimal.
Still, if we are careful, things will work out.
It will be useful to have the following locution available:
Events establish whether P if they make P true or make it false (make ~P

true).
Perhaps we should simply say ‘make whether P’, but that sounds even worse.
It seems like there are lots of important propositions whose truth value is
established not by being made true by events, but in some other way. For
example, consider Pythagoras’s theorem, the proposition that the square of
the hypotenuse of a right triangle is equal to the sum of the squares of the other
two sides. No event has ever made this proposition true, and none ever will.
It’s not at all like the proposition that Nixon won the 1972 election. There is
236 J. Perry
no sequence of events, ending at a certain time, the occurrence of which

makes Pythagoras’s theorem true. So the converse of the principle above
isn’t right; it is not true that if a proposition is true, then some events either
have made, are making, or will make it true.
It would not be correct to say that the truth of the proposition that
Pythagoras’s theorem is independent of events; events do conform to it. But
they don’t make it true. They reflect its truth.
For propositions that report ordinary facts, such as the proposition that
Eisenhower was president in 1954, or that Gore will be president in 2006, it is
natural to use the term ‘because’:
The proposition that Eisenhower was president in June, 1953 be true because
of events that occurred prior to 1954, including his election in 1952 and
inauguration in 1953.
The proposition that Gore will be president in 2006 be true, if it be, partly
because of events that have yet to occur.
With propositions such as that Pythagoras’s theorem, a quite different kind
of explanation of their truth seems appropriate, and of course there is a lot
of philosophical controversy about what the correct explanation might be.
I’ll simply say that such propositions are not made true by events, and leave
it at that for now.
Finally, and importantly, suppose that a true proposition, that is not
made true by events, together with some other propositions, that have
already been made true by events, entail a proposition that has yet to be
made true. What should we say about that proposition? To return to our ex-
ample, suppose that the laws of nature are not made true by events, and that
these laws, together with propositions made true by events that have
already happened, entail that Elwood will not eat cookies at future time t. I
will say that although the proposition that Elwood will not eat the cookie at
t has not yet been made true, it’s truth value has been settled. The proposi-
tion won’t be made true until the events that the laws of nature and the past
determine have actually occurred: the time t has arrived, and Elwood says
“no” to the proffered cookie, keeps his arms at his sides, and walks away. But
these events were already entailed by a combination of propositions some of
which were already made true and the rest of which aren’t made true by
events at all. So the truth value of the proposition that he would not eat
was, in that sense, already settled, before he refrained from eating.
The two issues on which I believe the tenability of compatibilism turns

are:
Is the truth of laws established by the events that confirm them and fail to
disconfirm them, so that laws are laws because events conform to them? Or
is the truth of laws established by something else, so that events conform to
them because they are laws? The first view is a weak theory of laws, the
second a strong theory of laws.
Can one have the ability to perform or refrain from an action A at time t,
even though the issue of whether one will perform A at t or refrain from
doing so has been settled before t? A weak account of ability will allow us to
answer yes to this question; a strong account will force us to answer no.
4 Strong and Weak Laws
One option for the compatibilist is to adopt a weak conception of laws.

On a weak conception of laws, (0) does not add much, if anything, to the
argument of section 2. Laws are basically true generalizations, and true gen-
eralizations are made true at least in part by the events that, as we say, con-
firm them. The laws that determine that Elwood won’t eat a cookie may be
true, but, nevertheless, not be made true until the last human or the last
cookie has passed out of existence. Elwood’s not eating cookies was part of
the sequence that established the law, not something the law settled. The
law and the facts leading up to Elwood’s choice may have determined that
Elwood would pass up the cookie, but they did not settle it, for the law itself
wasn’t made true until long after Elwood made his decision, and in fact his
decision was part of what made it true.
A person can make true generalizations false in the following sense: the
generalization be true, but there is something the person could have done,
or can do, such that if they had done it, or were to do it, the generalization
would not be true. Suppose there were two soccer teams, Manchester
United and Nottingham Sherwood. Suppose that Nottingham Sherwood
existed from 1960 to 2000 and then was disbanded. During that time it
played Manchester United eighty times and never won or tied. So it is a
true empirical generalization that
(G) For all soccer games g, if g is a game between Manchester United and
Nottingham Sherwood, Nottingham Sherwood loses g.
238 J. Perry
While by 1975 or so this may have seemed like a decree of God to the
Nottingham fans, we’ll assume for now that it was really just a sad but
true generalization. Suppose that in the second game between these two
teams in 1978 they were in a 0 to 0 tie at the end of the game, and then
Manchester won in overtime. In the last second of regulation play, Not-
tingham had a clear and easy shot on Manchester’s goal, that their best
player missed. Nottingham’s worst player was watching a plane overhead
skywriting advertisements for Guinness, and ran into their best player just
as he kicked. I think the Nottingham fans at least would think that the
Nottingham team could have won that game, even though they did not.
And that means they could have made the true generalization G false.
I think it is common sense to suppose that laws are not simply true
generalizations. Suppose that one Nottingham fan says ‘We could have won
that game in 1978 . . .’ Another particularly bitter Nottingham fan says,
No, you are wrong. You have not grasped that it is an unshakable law of
nature that Manchester United beats Nottingham Sherwood. It’s a law of
nature because God decreed that Nottingham would always lose. Laws
of nature are universal generalizations that God issued as fiats during
creation week, and other things that follow from them. For reasons finite
mortals can’t be expected to fathom, he often punishes the virtuous and
rewards the wicked. And in this spirit he decreed that Manchester
United would always beat Nottingham Sherwood. It seemed like
Nottingham could win, but in fact it could not.
This fan’s remark embodies the intuition behind the argument of section 2.
Intuitively, laws are more than true generalizations, and (0) adds something
substantial to the argument. You can make a mere generalization false;
even if no one gets around to doing so, it remains true that someone could
have. But laws are laws. You cannot make a generalization false, if it follows
from the laws of nature. This is the strong conception of laws. But of
course, one can have a strong conception of laws without believing in an
unfathomable God or any god at all.
Let’s remind ourselves of (1):
(1) ∀t ∀x (␾(x, t) → ␸(x, t 1))
If (1) is true then there have not been, and will not be, times t and individ-
uals x such that ␾(x, t) and ~␸(x, t 1). And (0) says that (1) is not simply
true, but true according to the laws of nature.
The question is, does (1) make (0) true, or does (0) make (1) true? Is (0)
true because (1) has no disconfirming instances? Or does (1) not have dis-
confirming instances because of the truth of (0)—because (1) follows from
the laws of nature? Is the truth of (0) one of those things, like the truth of
Pythagoras’s theorem, that is established by something other than what
happens? Is it the sort of things to which events conform, but do not make
true? Or is it just a sort of fancy way of saying (1)?
To return to our soccer fans. A third soccer fan, also a fan of Hume’s, may
say:
I don’t think (G) is a decree of God. But I agree with you that it is a
law of nature. A law of nature simply is an exceptionless generalization
that we have grown used to so that it shapes our expectations. And (G)
certainly is that.
This remark would express a weak conception of laws. It isn’t quite enough
for (G) to be a law that it is true. More is required: that we use it to form our
expectations. But that’s all. There is no big metaphysical condition, like a
command from God, which is also required. Being a law is just being a true
generalization that we have internalized so that our expectations about the
future are shaped by it. On this conception, it is (1) being true that ex-
plains, or partly explains, that it is a law. (1) Explains (0), not the other way
round.
On either conception of laws, laws will have no disconfirming instances.
On the strong conception, the fact that L is a law explains why events con-
form to it; the truth of the law is due to something other than the lack of
disconfirming instances. If determinism is also true, laws and propositions
about the past not only entail propositions about the future, but also settle
them.
On the weak conception of laws, however, the incompatibilist argument
does not work. (0) adds nothing to the argument that might push the con-
clusion from (4) to (5). A Humean conception might add something to the
requirement for being a law. (1) not only has to be true, but accepted and
used to guide our expectations. But this doesn’t push us from (4) to (5).
One option for the compatibilist, then, is to insist on this very weak,
Humean conception of laws. What we do is up to us; laws are merely those
descriptions of what we do that will end up being true once human actions
are complete. Laws determine, but do not settle. I’ll call this view ‘existen-
tialist compatibilism’.
240 J. Perry
I find existentialist compatibilism very appealing, but not wholly con-

vincing. Consider the law that for every action there is an opposite and
equal reaction. On the weak conception, this is a law because there never
have been and never will be any exceptions to it, and we are attuned to it:
when we see a reaction, we expect an opposite and equal reaction. There is
nothing about things that make this law true, except that everything con-
forms to it. It seems to me much more plausible that this law gets at some-
thing (or some things) about the universe that explains why things
conform to the law and it has no disconfirming instances. I find it hard to
stick with the Humean conception of laws.
One nontheistic but strong conception of laws holds that they are true
generalizations that derive from the nature of things, and so describe con-
straints that form the structure of the world. These constraints are relations
between types of things and types of situations.
We can look on (1) as telling us that a certain relation holds between two
types of events: coinstantiation. Whenever there is a ␾(x, t) type of event,
there is a ␸(x, t 1) type of event. On the constraints view, there are other
relations between types, such as causing, making happen, and forcing. These
are the relations Hume wanted to reduce to, or eliminate in favor of, coin-
stantiation plus psychology. He called them ‘necessary connections’. I
think that ‘necessary’ is rather confusing, given the uses of the term that
are familiar in current philosophy. Causal relations are not necessary in the
sense of being analytic, or in the sense of being true in all logically possible
worlds, or even all metaphysically possible worlds. Still, causal relations be-
tween types of events are basic to the structure of the actual world. So (0)
explains (1), not the other way around. I’ll call them structural connections.
Structural connections are not necessary, like Pythagoras’s theorem, but
they are not made true by events, either. Events conform to them because
they capture factors about the world that shape events, not because they
report events.
On this conception, if a generalization is true according to the laws of na-
ture, it reflects a constraint that holds among types of things or situations
in virtue of their nature, or a necessary consequence of such constraints.
Laws are rooted in the nature of matter, space, and time, or the nature of
whatever else it is that makes up the universe. When one billiard ball hits
another the direction and velocity of the second is determined by the
direction and velocity of the first in a certain way. What makes this so? Not
some decree, sentence, statement, or proposition that truly describes it.

Nor the facts about what has happened in similar situations in the past and
will hold in the future. There are real connections between types of things
and situations. This seems to be what Hume denied, or at least denied we
could ever understand. Disagreeing with Hume makes me nervous, and I
find it hard to say what else there is about the universe, other than the flow
of events, that constitutes such constraints. Nevertheless, I can’t bring
myself to accept any weaker conception of laws.
On this conception, the states of Elwood that are involved in his being in
␾ cause him to not eat cookies. These states include such things as really
not wanting to eat cookies, and seeing no reason to eat cookies. It seems to
me that in this sort of case we are in touch with properties that cause us to
take or not take certain actions in a pretty direct way. It seems in the nature
of things that someone in such states would not take a cookie. It would be
nicest, from a compatibilist point of view, to have a completely weak con-
ception of laws. Nevertheless, this conception seems to hold some promise
of fitting into a compatibilist picture when combined with a suitably weak
conception of ability.
5 Ability and Action
Consider this principle:
(S) If x can perform A at t, then at no time earlier than t is it settled

whether x performs A at t.
A strong theory of can supports (S), while a weak theory does not. I’ll argue
for a weak theory, and explain why it undercuts van Inwagen’s arguments.
First an analogy.
It’s 1956 and Elwood doesn’t buy a new Edsel. He thinks they are ugly, un-
gainly, and overpriced. He doesn’t want one. So he doesn’t buy one. Now
does it follow that he can’t afford one? Of course not. He may have all the
money he needs, and simply not want one. One question has to do with
what he wants in a car and what he thinks about the Edsel. These facts, what
he thinks about Edsels and what he wants in the way of a car, are pretty
much located in Elwood’s head. At any rate, they are not located down the
street at the bank. But that’s where the facts about how much money he has
in his account, and how much credit the bankers will give him, reside. It
242 J. Perry
may be that Elwood would rather be drawn and quartered or have rats gnaw
out his eyes than buy an Edsel. But those facts about his mind don’t tell us
anything about his bank account. He may be loaded, so he can easily afford
a fleet of Edsels. He can’t buy the car without money or credit, but he can
not buy the car even though he has plenty of money and plenty of credit.
We could put forward a little argument that Elwood won’t buy an Edsel:
(1) Reasonable people don’t buy cars that they think are ugly, ungainly
and overpriced and that they simply don’t want and have no other
reasons to buy (law of nature).
(2) Elwood thinks Edsels are all of those things, and has no other
reason to buy one (fact about Elwood’s mind).
(3) So he won’t buy one (fact about Elwood’s action).

No conclusions about Elwood’s bank account can be validly drawn from
this argument. It would be silly to draw the further conclusion
(4) So he can’t afford one (fact about Elwood’s bank account).

The premises don’t say anything about Elwood’s bank account or his credit.
So no conclusions about his bank account or credit can be validly drawn.
That’s the model for a weak account of ability. Whether Elwood performs
A is one question, having to do with what he wants and believes. What he
can do is something else, having to do with what abilities he has. If Elwood
can’t perform A then he won’t. But it doesn’t follow from the fact that he
won’t that he can’t.
Of course, people’s basic abilities aren’t kept down the street in the bank.
Elwood’s ability to pay for an Edsel may depend in part on his bank account.
But his ability to write a check, or say, “Please, sell me an Edsel,” depend on
facts about him. Still, facts about abilities are quite different than facts
about desires and beliefs. Let us suppose—completely fancifully—in order
to keep this important point vividly in mind, that one part of the brain has
to do with what people actually do, and another has to do with what they
can do. Let’s say the left side contains the desires and beliefs and the other
stuff that actually motivates actions. The right side contains all the basic
abilities, the repertoire of actions.1 The repertoire is tapped when one
decides to do something that requires a certain ability.
When I learn how to do something, to walk or pick up a glass of water or
ride a bicycle or write a check or balance a checkbook or prove a theorem,
things change on the right side of my brain. My repertoire of abilities

increases. As I learn to do these things, the left side of my brain may change
too. I may develop aversions to proving theorems and balancing check
books, while learning to like riding bicycles, walking, writing checks, and
drinking. Then we can predict that I’ll do a lot more riding bikes, walking,
and drinking than theorem proving and checkbook balancing. What I
want to do, and so what I will intentionally try to spend my time doing,
depends on the left side. What I can do, depends on the right side.
Given a weak account of ability, the fact that someone did not do some-
thing in the way that we would describe as ‘intentionally and of his own
free will’ if we were not worried about determinism, and that his not doing
that thing fell under a strong law of nature linking what one thinks and
wants with what one does, could hardly have the implication that he could
not do it. That he does not do it has to do with the lack of reasons he has
for doing it, a fact about the left side of the brain. That he can do it depends
on what is going on in the right side of the brain, a quite different question.
If we put an account of abilities in the context of a theory of intentional
action, the weak conception of ability makes a lot of sense. Here is a sketch
of what is involved in an intentional action; the sketch is no doubt
simpleminded and controversial, but I do not think adding sophistication
and resolving disputes should affect the points I make (see Israel, Perry, and
Tutiya 1993 and Goldman 1970).
First, there is a motivating complex of cognitions. Such a motivating complex
for an action A includes beliefs (broadly construed so beliefs include fleeting
perceptual beliefs, implicit beliefs, and so forth) and desires (broadly con-
strued to include wants, urges, whims, and so forth) that rationalize A-ing. A
set of cognitions C rationalizes an action A if A-ing will promote the satisfac-
tion of the desires in C given the truth of the beliefs in C. In other words,
when a person does something intentionally, there are a bunch of beliefs,
perceptions, wants, desires, preferences, and the like, which for conve-
nience, I’m just calling ‘desires and beliefs’, relative to which it is reasonable
for him to do it. For example, if I intentionally order a vanilla ice-cream cone,
the motivating complex might include the desire for a vanilla ice-cream
cone; the perception of a counter, a server, cones, and vanilla ice cream;
knowledge of English; a belief that I can afford it; a belief that it won’t do me
any harm; a belief that I can get one by ordering it; a belief that I order one by
asking for it; and so on.
244 J. Perry
The motivating complexes cause volitions to perform basic actions,

which will cause the basic action, if it is in the repertoire of actions—that
is, if the person has the ability to do it. I think of the basic actions as bod-
ily movements, and so use the term ‘execute’ for the special case of per-
forming one of these basic actions: we execute movements, and thereby do
lots of other things. I’ll try to order a vanilla ice-cream cone, by trying to
execute coordinated movements of voice box, throat, lips, tongue, and the
like that produce the sounds that will be recognized as the English sen-
tence, ‘May I please have a vanilla ice-cream cone?’ If that is one of the
things I can do, I’ll say it.
These basic actions in turn cause various results, depending on the cir-
cumstances. And these results cause further results, depending on wider
and wider circumstances. My words will cause events in the air between me
and the ice cream server, in his ear, in his brain, and so on, until with a
little luck I get my ice-cream cone.
Here’s another example. I am on an airplane and desire a drink of water,
and a steward comes by and holds out a tray full of water glasses. I believe
that there is a glass of water on the tray in front of me, due to perception
and trust in airlines to fill glasses with water rather than gin or hydrochlo-
ric acid when they intend to offer them to passengers as water. I know that
in these circumstances taking a glass from the tray and drinking from it is
a way of quenching my thirst. I can’t think of any reason not to take a
drink of water. My beliefs and desires rationalize the action of taking a glass
from the tray and drinking it, for this will satisfy my desire for a drink of
water, without leading to any untoward consequences, given the truth of
my beliefs.
This complex will then cause a volition to move in a certain way. Picking
up a glass from a tray is a rather delicate movement, but even a klutz like my-
self can usually do it right. I can suit the movement to the situation based on
perception, so that my hand moves to the glass and then brings it to my lips.
An important piece of evidence that I can do this is that when I intend to get
a glass of water, and see the glass in such and such a relative position, I usu-
ally move my hand in a way that succeeds in grabbing it and getting it to my
lips. This is due to the ability to execute various movements, and know-how
on my part that allows me to execute the right movements in the right situ-
ation based on perception. This is something I’ve gotten reasonably good at,
thanks to years of picking up some things and dropping others.
If there is a glass of water there, and it is reachable in the ordinary ways,

and I have the ability to execute the needed movements in the circum-
stances, then I can take a drink of water. If there is a invisible shield be-
tween me and the glass, or if the steward is a smart aleck who will move the
tray when I get close to it, or if he is an evil airline demon who has brought
around glasses full of hydrochloric acid, then I cannot get a drink of water.
So, part of the question of whether I can do it is a matter of the circum-
stances I am in. The other part has to do with what actions are in my reper-
toire. If I cannot reach as far as I need to, or grab the glass as firmly as
required, then I cannot get a drink.
A person has the ability to bring it about that R in circumstance K if (i) the
person’s repertoire of basic actions includes some movement M such that
(ii) executing M in K will have the result that R. These conditions for being
able to bring it about that R can be met when a person does not in fact
bring it about that R or even try to. Neither of the conditions depends on
the person actually bringing it about that A. Neither of them require that
he want to do so, or have a reason to do so. They do not preclude the person
really really wanting not to A. The person may be willing to die rather than
perform A. Conditions (i) and (ii) clearly can be satisfied even if the per-
son’s not executing M falls under a law of nature to the effect that a person
with his motivating complexes will not execute M.
This weak account of ability does not support (S). On this account of
ability, it does not follow from the fact that the (strong) laws of nature plus
Elwood’s beliefs and desires settle that he will not raise his hand at t, that
he does not have the ability to raise it at t. That this does not follow can
be seen by considering our argument (0)–(5). With a weak theory of abil-
ity, it clearly does not work, even if we assume a strong theory of laws.
Go down the steps. From (0) to (4) nothing is said about abilities. Then, in
step (5), abilities are ruled out. It is a left-brain argument, with an invalid
right-brain conclusion tacked on. It sounds sort of intuitive, but it just
doesn’t follow.
6 Van Inwagen’s Arguments
Changing the Past

Now let’s turn to one version of van Inwagen’s argument (1975, 23).
The issue at hand is whether or not J could have raised his hand. Q is a
246 J. Perry
proposition that rules out J having raised his hand at t. I’ll just take Q to be
the proposition that J did not raise his hand at t. Call the laws of nature L
and call the facts up until t that are relevant PF. Assume that we cannot
change the past. Then,
(1) If determinism is true, then the conjunction of PF and L entails that Q.

(2) If J had raised his hand at t, then Q would be false.
(3) If (2) is true, then if J could have raised his hand at t, J could have
rendered Q false.
(4) If J could have rendered Q false, and if the conjunction of PF and L
entails Q, then J could have rendered the conjunction of PF and L false.
(5) If J could have rendered the conjunction of PF and L false, then J could
have rendered L false (since J cannot change PF).
(6) J could not have rendered L false.
(7) So J could not have raised his hand.
This argument, if valid, clearly generalizes to any action whatever, and so

rules out compatibilism.
Premise (4) is false. The appearance of truth is due to ambiguity in ‘ren-
ders a proposition false’. We have to resolve the ambiguity in favor of the
sense of ‘renders false’ in which step (3) follows from (2). In this sense (4) is
false.
One sense of ‘render a proposition false’ is to do something, which
changes a proposition from being true, to being false. This seems quite im-
possible. Of course one can do something that makes a proposition that
had looked like it was going to turn out to be true turn out to be false
instead. It might seem virtually certain that one team is going to win a
baseball game, but then the other team scores twelve runs with two outs in
the ninth inning to win 12–11. They snatch victory from the jaws of defeat.
But they do not really manage to change a proposition from true to false.2
It is crystal clear that in this sense of ‘render false’, (3) does not follow
from (2). In this sense, (3) says that if J could have raised his hand at t, then
J could have changed the truth value of that J does not raise his hand at t
from true to false. But this does not follow from (2).
The second sense of ‘render a proposition false’ is to do something which
makes the negation of the proposition true, at a point in time at which
nothing has yet made the proposition true or made it false. This concept of
‘render a proposition false’ makes perfectly good sense. By eating a cookie

at t, I render the proposition that I do not eat a cookie at t false. So, the propo-
sition that I do not eat a cookie at t be false, and it be false because at t I (will
refrain/refrain/did refrain) from eating a cookie.
In this sense of ‘render a proposition false’, (3) does follow from (2), as
van Inwagen’s argument requires. But (4) is false. It does not follow from
the fact that one renders a proposition Q false, in this sense, and that some
other proposition R entails Q, that one also renders R false.
Let R be the following proposition. Recall that Q is the proposition that J
does not raise his hand at t. Let t be some day after 1950:
(R) Q & J’s mother ate a carrot in 1944.
This proposition entails that J does not raise his hand at t. J can render the
proposition Q false by raising his hand at t. If he renders Q false, R be false,
too. But R may have already been rendered false by the time J renders Q
false. This will be the case if J’s mother did not eat a carrot in 1944. In this
case, J will not render R false, even though R entails Q and he renders Q
false. It simply does not follow from the fact that J will render Q false that
he renders false every proposition that entails Q. What does follow is that
there is no true proposition that entails Q.
Principles (i) and (ii) are clearly true, given the coherent concept of
‘render true’ and ‘render false’.
(i) Suppose one does something that renders P true. Then no proposition
that entails the falsity of P be true.
(ii) Suppose one does something that renders P false. Then no proposition
that entails P be true.
Principles (iii)* and (iv)* do not follow, however:
(iii)* Suppose one can do something that would render P true. Then no
proposition that entails the falsity of P be true.
(iv)* Suppose one can do something that would render P false. Then no
proposition that entails P be true.
Principles (iii)* and (iv)* simply amount to the principle that there is no
difference between being able to do something and doing it—that can
collapses into does, and does not into cannot.
If I can drink a beer, I can render that I drink a beer true. So, given (iii)*, if
I can drink a beer, no proposition that entails that I don’t drink a beer is true.
248 J. Perry
So if I can drink a beer, that I don’t drink a beer isn’t true (since it entails
itself), so it’s false, so I drink a beer. If I can do it, I do it. Can implies does.
Suppose I don’t drink the beer. Then, that I don’t drink the beer is true.
Then something is true that entails the falsity of that I drink the beer. Then,
by (iv)*, I can’t render it true that I drink the beer. So I can’t drink the beer.
Doesn’t implies can’t.
Such a collapse of ‘can’ into ‘does’ and ‘doesn’t’ into ‘can’t’ is, of course,
just what the incompatibilist wants and the compatibilist needs to avoid. If
we accept (iii)* and (iv)*, the collapse would be completed without any
appeal to determinism at all. But of course there is no reason to accept (iii)*
and (iv)*. On the contrary, it seems quite clear that on the weak conception
of ability, (iii) and (iv) are true instead:
(iii) Suppose one can do something that would render P true. This does not
imply that no proposition that entails the falsity of P is true.
(iv) Suppose one can do something that would render P false. This does not
imply that no proposition that entails P is true.
Suppose J can raise his arm at t, but decides not to. Then that J does not raise
his arm at t is true. This proposition entails itself. So J can raise his arm at t,
even though a proposition that entails that he does not raise his arm at t is
true. So (iii) is correct.
Suppose J can refrain from raising his arm at t, but in fact he raises it.
Then that J raises his arm at t is true. This proposition entails itself. So J can
refrain from raising his arm at t, even though a proposition that entails that
he does raise his arm at t is true. So (iv) is true.
Now let’s return to the crucial part of van Inwagen’s argument:
(1) If determinism is true, then the conjunction of PF and L entails that Q.

(2) If J had raised his hand at t, then Q would be false.
(3) If (2) is true, then if J could have raised his hand at t, J could have
rendered Q false.
(4) If J could have rendered Q false, and if the conjunction of PF and L
entails Q, then J could have rendered the conjunction of PF and L false.
Since van Inwagen’s argument proceeds by very general principles, it

should work for any more concrete example we choose. So let:
PF that at t 1 J really really does not want to raise his hand in the next
instant.
L that no one who at t 1 really really does not want to raise his hand
in the next instant, raises his hand at t.
Q that J does not raise his hand at t.
This example meets the conditions of van Inwagen’s argument. That is,
PF & L entails Q.
We can certainly accept steps (2) and (3), given our understanding of
‘render Q false’. But step (4) does not follow.
PF is the proposition that at t 1 J really really wants to not raise his
hand in the next instant. So (4) says that if J could render Q false (i.e., if he
could raise his hand at t), then he could render false the proposition:
that L & at t 1 J really really wants to not raise his hand in the next instant.
But there is nothing that J can do at t, the doing of which would make it the
case that it was not true at t 1 that he really really wanted not to raise his
arm at t.
If J does raise his hand at t, that will show, given L, that PF is not true.
However, that will not make PF untrue; it will not render PF untrue. If he
raises his hand at t, that will be because he is in some state at t 1 than re-
ally really wanting not to raise his arm, perhaps in the state of wanting
to raise it. In this case, PF be untrue, but it be untrue because the events at
t 1 made it false, not because of what J does at t.
If we go back to our simple picture of what it is to be able to raise your
hand at t, this should be reasonably clear. There are two facts about J and
raising his hand, with these possible combinations:
Does raise his hand Does not raise his hand
Can raise his hand 1. Possible 2. Possible

Cannot raise his hand 3. Not possible 4. Possible
The argument starts with the premises that J does not raise his hand, that is,
he is in cell 2 or cell 4. It then hypothesizes that he can raise his hand,
putting him in cell 2. From this it is supposed to follow that he changes the
past, since the past determines that he will not raise his hand. But it clearly
does not follow, for in cell 2 he does not raise his hand, just as the past
determines will happen.
I conclude, then, that as long as we have a weak, but realistic and com-
monsense concept of ability, we can be determinists and compatibilists,
even if we accept a reasonably strong conception of laws.
250 J. Perry
Van Inwagen’s ␤ Principle

Van Inwagen has produced several arguments for incompatibilism. The
one I have discussed is the one that seemed most intuitive and convincing
to me. Recently more attention has been paid to another argument from
his book An Essay on Free Will (van Inwagen 1983, 93–104; see also van
Inwagen 2002). The key principles are:
(␣) p → Np
(␤) From Np and N(p→q) deduce Nq,
where ‘Np’ means ‘p and no one has, or ever had, any choice about
whether p’.
We are thinking of p and q as propositions, and entailment as a relation
between propositions. It seems we should accept
if p entails q then (p → q).
Then if we also accept (␣) and (␤) we’ll have to accept the rule,
(␹) From Np and p entails q deduce Nq.
Principle (␹) is fatal to compatibilism. If determinism is true, and p is the

conjunction of the laws of nature and the facts up until life evolved on
earth, and q is any proposition entailed by them describing an act, no one
will have any choice whether q.
A holder of a weak theory of action will reject (␤) and so be spared from
(␹). The premises of the rule of inference (␤) imply nothing about the abil-
ity to bring it about that q, while the conclusion does. On a weak theory of
action, (␤) is not valid.
Recall the criterion for a strong theory of ability:
(S) If x can perform A at t, then at no time earlier than t is it settled

whether x performs A at t.
To be settled at t is to follow from some set of propositions, each of which

is either established or made true by time t. A strong theory of laws says
that laws are either established or were made true a long time before
humans began doing things. So, given a strong theory of laws and (S) and
determinism, no one will be able to perform any act A at any time. A weak
theory of ability denies (S). The weak theory holds that, since the question
of whether a person has an ability at a given time need not be affected by
his desires and beliefs, and yet it is his desires and beliefs that, together
with the laws of nature, determine what he does, the fact that he will or
will not do something does not preclude his having the ability to refrain or
not refrain from doing it. The weak theorist thinks that a person can have
a choice about something, in the sense that they have the ability to do it or
refrain from doing it, even if that thing is determined by laws of nature that
are established and facts that are already made true. The weak theorist,
then, having rejected (S), need have no qualms about rejecting (␤).
7 Lewis’s Analysis
In “Are We Free to Break the Laws?” David Lewis distinguishes between the
following claims: ‘I am able to do something such that, if I did it, a law
would be broken’ and ‘I am able to break a law’ (Lewis 2001, 31ff). Suppose
the laws of nature and the history of the world up until time t 2 entail
that I will not take the glass of water at t, but I don’t. Suppose, as Lewis
does, I cannot change the past. There seem two possibilities:
(a) Something happened at t 1 that was contrary to the laws of nature,
that is, a ‘divergence miracle’.
(b) Everything up to and including t 1 was in accord with the laws of
nature, but my action was not.
Lewis thinks the fact that I can take the glass of water implies that I am able
to do something such that, if I did it, a law would have been broken at
some earlier time, but this requires only (a). He does not think I am able to
break a law, which would require (b).
I do not think the compatibilist need suppose that if I were to take the
drink, any laws of nature would ever need to have been broken. There are
auxiliary premises from Lewis’s metaphysics and analyses of causation,
counterfactuals, and the like that lead him to this defense of compatibil-
ism. But compatibilism by itself does not force us to the divergent miracles
defense, and it does not seem to me the most plausible thing to say about
cases in which one has the ability to do differently than one does.
If I had taken the drink, freely and voluntarily, then surely my beliefs and
preferences would have been different than they actually were. The most
likely difference would be that I was thirsty. Assuming determinism, if I
had been thirsty when the drink was offered, then something earlier also
would have been different; perhaps I wouldn’t have taken a drink at the
252 J. Perry
fountain before stepping on the plane, as I did, because the fountain was
broken. And that would mean some earlier state of the fountain and its sur-
roundings had been different. And so on. Tracing the changes back to the
big bang, perhaps it might be a slight difference in the direction in which
one particle began its travels through time. Or perhaps it goes back to a
deistic god creating the initial state of the universe a very little bit differ-
ently. Or perhaps it just goes back, infinitely. Who knows? It’s certainly
amazing and weird and in my opinion somewhat depressing that the trail
of differences that would have led to my being thirsty rather than not
being thirsty should lead back even a couple of thousand years, much less
to the beginning of time, or forever. Still, I can’t see why either (a) or (b) is
required for me to take the glass. Assuming determinism, it follows from
the fact that I can accept the drink and don’t, that I can do something such
that if I did it either the laws of nature or the events up until that time
would have been different than they in fact are. It does not follow that if I
did what I can do any law would thereby be broken, or any divergence
miracle would ever have occurred, or I would have changed the past in any
way. I wouldn’t have had to change the past, because, according to deter-
minism and the laws of nature, if I had been thirsty, the past would have
been different.
8 Conclusion
A compatibilist can evade incompatibilist arguments by adopting a weak

theory of laws, or a weak theory of ability, or both. My own inclination is
in favor of a strong theory of laws and a weak theory of ability.
Although I believe in compatibilism, I am somewhat skeptical about the
truth of determinism. I would be happy if it were not true, for I think that de-
terminism is a doctrine that is not very accommodating to important human
hopes and aspirations. I don’t think the problem is that it rules out freedom,
however. I hope I can address these issues in a helpful way on a future
occasion. I’m sure I want to, but I’m not at all sure I have the ability to do so.
Acknowledgments
Early versions of this paper were presented at the 2001 Inland Northwest
Philosophy Conference and the Philosophy Department Colloquium at the
University of Nottingham. I received helpful criticisms and suggestions

from members of both audiences. I received helpful comments from
Michael Bratman, Joseph Keim Campbell, Eros Corazza and Michael
O’Rourke on later versions. Campbell commented on several versions; he
saved me from bad mistakes, and suggested helpful repairs. I am very grate-
ful. Much of what I understand about these topics is due to John Martin Fis-
cher, through many conversations and his works, especially Fischer (1994)
and Fischer (1996). None of these folks is responsible for the mistakes that
remain, although of course if the whole paper is mistaken, and I’m wrong
about everything, and both determinism and incompatibilism are true, I’m
not either.
Notes
1. Let me emphasize that the right and left sides are chosen completely arbitrarily,
simply as a way of easily visualizing the point. This is not an attempt to fit agency
into what is known, or thought, or imagined, or claimed, about the differences be-
tween the right and left side of the brain.
2. Note that even if we adopted the more complicated account of the issues
discussed in section 3, so that propositions were neither true nor false until events
made them so, making a proposition false would not mean changing its truth value
from true to false. If we did things this way, we would have to say that when a set of
premises entail a proposition about the future, the truth of the premises requires that
the proposition will be true, not that it is true.
References
Ekstrom, L. W., ed. 2001. Agency and Responsibility: Essays on the Metaphysics of
Freedom. Boulder, Colo.: Westview Press.
Fischer, J. M. 1994. The Metaphysics of Free Will: An Essay on Control. Cambridge,

Mass.: Blackwell.
. 1996. A New Compatibilism. Philosophical Topics 24: 49–66. Reprinted in

Ekstrom (2001): 38–56.
Goldman, A. 1970. A Theory of Human Action. Englewood Cliffs, N.J.: Prentice-Hall.
Hume, D. 1748. An Enquiry Concerning Human Understanding. London.
Israel, D., J. Perry, and S. Tutiya. 1993. “Executions, Motivations and Accomplish-
ments.” Philosophical Review (October 1993): 515–540.
254 J. Perry
Kane, R., ed. 2002. The Oxford Handbook of Free Will. New York: Oxford University
Press.
Kapitan, T. 2002. “A Master Argument for Incompatibilism.” In Kane (2002):

127–157.
Lewis, D. 1981. “Are We Free to Break the Laws?” Theoria 47: 113–121. Reprinted in
Ekstrom (2001): 30–37. (Page numbers refer to this latter work.)
Locke, J. 1694. Essay on Human Understanding, 2nd edition. London.
Van Inwagen, P. 1975. “The Incompatibility of Free Will and Determinism.”

Philosophical Studies 27: 185–199. Reprinted in Ekstrom (2001): 17–29. (Page num-
bers refer to this latter work.)
. 1983. An Essay on Free Will. Oxford: Oxford University Press.
. 2002. “Free Will Remains a Mystery.” In Kane (2002): 158–177.

12 Freedom and Contextualism
Richard Feldman
There are striking similarities between the debate about freedom and
determinism and the debate about knowledge and skepticism. In each case,
the issue can sensibly be set up as a puzzle about an inconsistent triad of
propositions, each of which seems to be true. Furthermore, there are posi-
tions one can take about one of these puzzles that have their analogues in
positions one can take about the other. An optimist might think that in-
sights achieved in one area will shed some light on the other. Such a person
is, of course, doubly optimistic, thinking both that there are insights in the
one domain and that these insights might be of value in the other. Some
philosophers have thought that contextualist theories in epistemology
have just this sort of value. John Hawthorne (2001) has recently defended
this line of thought in print. While I agree about the potential value of in-
sights in one domain for the other, I think that contextualism is of limited
value in helping us to think about either freedom and determinism or
about skepticism. I have previously argued for the limited value of contex-
tualism in epistemology (Feldman 1999, 2001). In this paper I will summa-
rize my arguments about contextualism and skepticism and then extend
those arguments to contextualism and freedom.
1 Some Inconsistent Triads
Knowledge and Skepticism

One way to set up skeptical issues is by considering the following three
propositions:
(A1) I know that I have hands.
(A2) I do not know that I am no mere brain in a vat.
(A3) If (A2) is true, then (A1) is false.
256 R. Feldman
Each of these three propositions seems, at first glance to be true. (A1) re-
ports the sort of everyday knowledge we all think we have. (A2) is true since
I have no way to know that I’m no brain in a vat. Things could seem just as
they do if I were a brain in a vat.
(A3) requires a bit more explanation. First, the contained propositions in
(A1) and (A2) should be understood in such a way that they are incompat-
ible. That is, being a mere brain entails not having hands. The idea under-
lying (A3) is a principle of closure, roughly to the effect that if I know a
proposition P, and P entails Q, then I know Q. There are many details here
that need attention if a fully adequate closure principle is to be stated.
For one thing, the closure principle should almost surely be restricted to
known entailments. That is, if I know P, and I know that P entails Q, then
I know Q. These details are important, and they may get in the way of gen-
erating any puzzle about skepticism based on the closure principle. How-
ever, the points I want to raise in this paper do not turn on details about the
closure principle, so I will not go into them here.
Stewart Cohen (1987, 1988, 1999) and Keith DeRose (1995), two leading
defenders of contextualist theories in epistemology, set up their problem in
roughly this way. The burden of their theories is to explain what to make
of the seemingly incompatible intuitions that each of (A1), (A2), and (A3)
is true. Cohen explicitly makes the issue one about an inconsistent triad.
DeRose instead formulates the issue as one about what to say in response to
the argument from (A2) and (A3) to the denial of (A1).
There are other ways to set up the problem of skepticism, also based on
inconsistent triads or their analogous arguments. For example, one could
note that with respect to any belief one has, or at least any belief one has
about the external world, there is some possibility of error. One could also
note that a person who allegedly knows something could be in a situation
that is introspectively indistinguishable from the situation of a person who
lacks knowledge. And considerations along these lines could be used to
construct an inconsistent triad. For example:
(B1) I know that I have hands.

(B2) There are possible situations in which a handless person is in a men-
tal state introspectively indistinguishable from the state I am in.
(B3) If (B2) is true, then (B1) is false.
Roughly, any consideration that you think may be a key premise in a
plausible argument for skepticism can be encapsulated in a triad like this.
Freedom and Contextualism 257
What should we say about these triads? Which elements should be given
up? As contextualists seem to see things, a central part of solving the skep-
tical problem is figuring out how to answer these questions.

The problems about freedom, determinism, and moral responsibility can
also be set up as inconsistent triads. One version of the puzzle involves free
action and determinism:
(C1) People sometimes act freely.
(C2) All human actions are causally determined.
(C3) If (C2) is true, then (C1) is false.
Variants of the puzzles could be developed that did not rely on the univer-
sal generalization in (C2). For example, puzzles could be formulated by
saying of certain apparently free actions that they are causally determined.
Also, the puzzles could be formulated in terms of claims about decisions
rather than actions. Furthermore, analogous puzzles could be set up in-
volving the ability to do otherwise or being responsible for one’s actions
rather than acting freely. I will not consider these variations here. I believe
that remarks similar to those I make below apply to all such puzzles.
Each of (C1), (C2), and (C3) has some intuitive appeal. At least in our
unguarded moments, we all assume that (C1) is true. In what we say, and
what we do, we presuppose that we ourselves, and those with whom we
typically interact, are free agents. And we assume that, at least sometimes,
they could have done other than what they did and that they are respon-
sible for what they have done. (C2) also seems to be true, or at least close
enough to true for present purposes. Human beings are, according to
many, animals that are part of the natural world. Our behavior is deter-
mined, or at least as close to determined as is the behavior of everything
else around us. We are not, according to this common view, exempt from
the causal laws governing the natural world. (C3) reflects the apparent
conflict between (C1) and (C2). It has intuitive appeal, and there are much
discussed arguments, which I will not consider here, that can be given
in support.
So all three propositions seem true. Yet one must be false. Which should
it be? One way to conceive of the problem of freedom and determinism is
as the problem of figuring out what to say about this inconsistent triad of
propositions.
258 R. Feldman
2 Context Sensitivity
The idea of context sensitivity is simple and familiar. It is crucial for present
purposes to distinguish the sort of context sensitivity under discussion here
from a very different sort. One philosopher (Annis 1978) has argued that
for a person to know that some proposition is true requires that the person
be able to answer all the objections to that proposition that are available in
the setting in which the person is located. This view seems to have the im-
plication that someone might know some proposition to be true in one
setting but could lose that knowledge simply by moving to a place where
people have some particularly difficult objections to that proposition. On
this view, the setting, or context, of the putative knower—the subject’s
context—plays a crucial role in determining whether the person has
knowledge. On this view, a person with the same reasons for believing the
same true proposition might in one setting know that proposition, yet not
know it in another. This is not the view under discussion here.
Context sensitivity as it is invoked in recent discussions of skepticism
has to do with the speaker’s context. Allegedly clear examples of this kind
of context sensitivity involve familiar comparative words such as ‘tall’ or
‘wealthy’ or ‘big’. In a conversation about elephants we might say of
Jumbo, the baby elephant, that he is not big, since Jumbo is much smaller
than the average elephant. But when talking about the animals we saw at
the zoo, we might say that Jumbo is quite big, since he is bigger than al-
most all the animals at the zoo. One idea about what goes on here is that
sentences of the form ‘X is big’ are incomplete because they are implicitly
relational. Things are not simply big or not big; they are big or not relative
to some group or class: they are big for an animal at the zoo but not big for
an elephant. But language works in such a way that we do not always have
to mention the reference class explicitly. The prior conversation and other
features of the setting, all making up the context, enable participants to
understand the intended relativizations.
The sort of context sensitivity just described has to do with the speaker’s
context. The point can be illustrated most clearly by noting that two con-
versations about Jumbo might be going on simultaneously. In one, some
elephant experts are discussing the relative size of some elephants. One of
them says that Jumbo is not big. What the expert says is true. At that very
moment, a child and his family are visiting the zoo. The child has tired of
seeing the little animals and asks if there are any big animals in the zoo.
The father says that there are, since Jumbo is big. This is true too. The
expert’s statement and the father’s statement do not conflict, even though
they are spoken at the same time. And there is nothing peculiar about
Jumbo’s size that plays any role in this. Nor is there anything about per-
spective or opinion that plays a crucial role here. Jumbo, we may assume,
is at the moment some fully determinate size. The predicate ‘big’ truly ap-
plies to a thing of Jumbo’s size in the context in which the father speaks. It
does not apply to a thing of that size in the context in which the expert
speaks.
The lesson to learn from this is a familiar one: the same sentence can be
used in different settings to say different things with potentially different
truth values.
3 Context Sensitivity and the Puzzle about Knowledge and Skepticism
As it is applied to epistemology, the heart of the contextualists’ view is that

the word ‘knows’, or sentences containing it, display context sensitiv-
ity. They think that all three elements of the inconsistent triads can express
truths, but not in a single context. In ordinary contexts, ordinary standards
for applying the word ‘knows’ are in place. By those standards, what’s
expressed by saying something of the form ‘S knows that p’ is often true.
But then, once we start talking about skepticism, the standards for apply-
ing ‘knows’ change. People rarely satisfy those higher standards. In those
contexts, the things people say by uttering sentences of the form ‘S knows
that p’ are typically not true. The skeptic’s denials are correct. Thus, in or-
dinary contexts, sentences like (A1) express truths. In those ordinary con-
texts, (A2) expresses a falsehood. When an ordinary speaker in an ordinary
context utters the sentence ‘I know that I have hands’, that speaker says
something true. However, in settings in which the standards for knowledge
have been raised and skeptical issues have become salient, by uttering (A1)
one would express something false. In these contexts, the proposition (A2)
expresses is true. On the best, but not all, accounts, (A3) always expresses a
truth.
Analogous things apply to the second version of the epistemological puz-
zle. In ordinary contexts (B1) is true and (B2) is false. This is in part because
among the possibilities relevant to the assessment of (B2) there are none
260 R. Feldman
that include my being a handless brain in a vat. Skeptical considerations

expand the set of possibilities relevant to the assessment of (B2), enough to
include some in which I am a handless brain in a vat having experiences
just like mine. In such contexts, (B2) is true and (B1) is false.
This description of the second puzzle accords well with David Lewis’s
version of contextualism. He says that:
S knows that P iff S’s evidence eliminates every possibility in which not-P—Psst!—
except for those possibilities that we are properly ignoring. (Lewis 1996, 554)
Lewis’s solution, which fits well with his more general view about lan-
guage, is that the stock of possibilities which must be eliminated if we are
to have knowledge varies with context. In effect, his view pins the context
sensitivity of ‘knows’ on the context sensitivity of ‘all’. ‘All possibilities’ do
not include possibilities that we can properly ignore. In ordinary contexts,
we can eliminate all possibilities. This is in part due to the fact that in
ordinary contexts the odd possibilities envisioned by skeptics are properly
ignored and thus are not among ‘all’ possibilities. But in skeptical contexts
there are additional possibilities, and we cannot eliminate them. His view
is that various things that we say have the effect of enlarging the pool of
possibilities. For example, mentioning a possibility makes it not properly
ignored. The sort of contextualism about freedom that Hawthorne discusses
is modeled on this proposal from Lewis.
There is a detail worth mentioning about contextualist solutions to the
skepticism puzzles. I did not say that in ordinary contexts one would ex-
press a falsehood by uttering (A2) or (B2). The reason for this is slightly
tricky. Contextualists do not simply assert that there are differing standards
for the application of the word ‘knows’. They say something about what
brings about the raising of standards. (Curiously, less is said about what
brings them back to normal.) The leading idea is that talking about skepti-
cism and skeptical hypotheses raises the standards. So uttering sentences
like (A2), by mentioning the skeptical hypothesis that I am a brain in a vat
(or my lack of knowledge that I am not a brain in a vat), induces the higher
standards. So, if I utter (A2), then I have raised the standards and made it
true. But I do meet the ordinary standards for the application of the predi-
cate ‘know that I am not a mere brain in a vat’. So, we can say that relative
to the standards determined by ordinary contexts, (A2) is false. But if we
were to say (A2), it would be true. This illustrates the point David Lewis put
colorfully by saying, “Epistemology robs us of our knowledge” (1996, 550).
Another detail is worth noting. Contextualism is a thesis about the func-

tion of the word ‘knows’. A completely accurate discussion of it would be
conducted almost entirely in a metalanguage. Thus, contextualists do not
hold that the standards for knowledge vary. Rather, they hold that the stan-
dards for application of the word ‘knows’ vary. If I use the words ‘knows’ or
‘knowledge’ in the present context, then the standards for their application
are whatever standards are in place in the present context. However, since
the metalinguistic talk is tedious and cumbersome, I will generally avoid it,
provided there is little danger of misunderstanding.
Contextualism is an explanatory hypothesis. The data to be explained
are the seemingly conflicting things we are apt to say using the word
‘knows’. The proposed explanation appeals crucially to the idea that the
standards for the application of that word vary, so that the things that
appear to conflict do not really conflict. Any alternative explanation will
imply that we are frequently making mistakes: either what we are apt to say
in ordinary contexts is mistaken or what we are apt to say when we are
influenced by skeptical considerations is mistaken. According to contextu-
alists, it is better to think that the truth conditions for applying ‘knows’
have changed than to think we are routinely mistaken in this way.
As I noted above, contextualists do provide some accounts of what makes
the standards change. Exactly how they formulate their theories will de-
pend in part upon what they think is required for knowledge. Presumably,
they will find some element of knowledge, or some condition on knowl-
edge, that can vary in a suitable way. Thus, if knowledge requires justifi-
cation, then one could say that how much justification is required varies
contextually (Cohen 1999). If knowledge requires ruling out relevant alter-
natives, then one could say that which alternatives are relevant varies with
context (Lewis 1996). If knowledge requires sensitivity to the truth value of
the proposition in question through all relevant worlds, then one can say
that the range of relevant worlds varies with context (DeRose 1995). Any
such view about knowledge will be amenable to a contextualist treatment.
To achieve the desired explanations of linguistic variation, contextualists
will have to provide an account of contexts and their affect on the truth
conditions for sentences using ‘knows’ that get the desired outcome,
namely, that our utterances are by and large true. This will require a poten-
tially complex theory about how features of a context alter the amount of
justification needed, or the set of relevant alternatives, or the set of worlds
262 R. Feldman
through which one must track a proposition. Contextualists have proposed

accounts of these matters, but I will not go into those details here. It is pos-
sible to raise doubts about the specifics of their theories. But I think that
there is a perspective from which those details do not matter. If they are
right about the data—that our inclinations to affirm and deny knowledge
sentences varies with context—then it is at least in principle possible to
figure out what features of the context cause those varying inclinations.
And if there are factors that affect us—that is, if the variations are not
random—it is possible to hold that the truth conditions for sentences using
the word ‘knows’ somehow are also affected by those same factors. Thus, if
there is any systematic way to account for the relevant changes in our
linguistic dispositions, it is at least in principle possible to construct an
account of the truth conditions for applications of the word ‘knows’ that
appeals to those very factors (or correlates of them) so that we are system-
atically disposed to affirm truths. Contextualists want to construct a the-
ory that makes our seemingly inconsistent inclinations routinely correct. It
is possible to construct such a theory.
The contextualist solution to the problem of skepticism thus amounts to
the following. In ordinary contexts, when we make claims using the word
‘knows’ we frequently say things that are true. But when the stakes are
raised, as they typically are in discussions of skepticism, the skeptics are
right. I find this to be a disappointing solution to the problem of skepti-
cism. I will explain why after looking at an analogous attempt to deal with
the problems associated with freedom and determinism.
4 Contextualism and Freedom
Given the structural similarity of the puzzles about knowledge and about
freedom, it is not surprising that contextualist responses to the latter puz-
zles are also possible. In broadest outline, the idea is that the truth condi-
tions for sentences involving the key terms—‘free’, ‘morally responsible’,
‘could have done otherwise’—are context sensitive. In a recent paper,
John Hawthorne has taken this approach to sentences using ‘free’ (2001).
(In an older paper Terrence Horgan 1979 applied it to sentences involving
‘could have done otherwise’. I will not discuss his view here.) Although
Hawthorne makes some remarks about ‘moral responsibility’ he does not
claim that that phrase itself displays context sensitivity, and I know of no
one who has. In this section I will spell out the key features of Hawthorne’s
response to the puzzle about freedom.
I should note at the outset of my discussion of Hawthorne’s paper that he
concludes by saying that he does not confidently believe the contextualist
analysis he discusses. Rather, he thinks it deserves to be “taken seriously”
and that it is a “worthy competitor to the more standard” views (2001, 77).
It will be easier to proceed here, as Hawthorne does in his paper, as if he
actually does accept the view he describes.
Hawthorne’s proposal is modeled on David Lewis’s version of contextual-
ism in epistemology. The idea is to make utterances of ‘S did x freely’ true in
ordinary contexts but not true in special philosophical contexts. Hawthorne
writes:
When ordinary speakers utter English claims of the form ‘S did x freely’ (and their
synonyms), they frequently speak the truth. But when our sphere of attention is
widened by philosophical inquiry, we are rarely in a position to truly utter the
English words ‘S did x freely.’ (2001, 68)
He gives an account of the meaning of ‘S did x freely’ that helps to account

for this fact. The account will require a little explanation. It is:
S does x freely only if S’s action is free from causal explainers beyond S’s control—
Psst!—apart from causal explainers that we are properly ignoring. (2001, 68)
The idea is that in any context in which we talk about whether an action is
done freely, we are considering only a certain set of potential causal expla-
nations of the action. We say that an action is done freely when we think
that there is no causal explanation of it within that set, and, if we are right
about this, then what we say is true. In ordinary contexts, he suggests, we
properly ignore as causal explainers hidden facts involving the “neurologi-
cal springs of action” and explanations involving the laws of nature oper-
ating on “the distribution of microparticles in the distant past” (2001, 67).
In ordinary contexts the sorts of causal explanations that would preclude
the correct attribution of ‘free’ apparently include causal explanations
involving macro processes occurring outside the individual. If, for example,
I am caused to move to the left by a sudden gust of wind, then I did not
move to the left freely. Ordinarily, this sort of explanation is not properly
ignored. But if there is no such explanation of my move to the left, if the
only causal explanations involve neurological events that we rarely think
of, or complex processes going back to childhood and before, then there is
264 R. Feldman
no relevant causal explanation, since we can properly ignore those causal

explanations. So in an ordinary situation, an utterance of ‘I did that freely’
would be true.
For reasons that I hope are apparent, I will not go into the details of
Hawthorne’s account of what determines which causal explainers are not
relevant (i.e., which are properly ignored). If our inclinations to make
utterances involving the word ‘free’ vary in the way he says they do, then
there is some reason for this variation. Something about the situation
prompts us to respond that way. He can then build into the account of rel-
evant conditions factors that pick up on just those features in just the way
needed to get the desired outcome. I grant that it is possible to do this,
though I do not grant that it is possible to do that in a way that has any
intuitive plausibility. In any case, I will not contest this point.
It is relatively easy to see how to apply these thoughts to the puzzle about
freedom and determinism.1 Recall the puzzle:
(C1) People sometimes act freely.
(C2) All human actions are causally determined.
(C3) If (C2) is true, then (C1) is false.
The contextualist resolution to the puzzle here is quite similar to the contex-
tualist approach to the skepticism puzzle. It says that (C1) is true relative to
some contexts and false relative to others, that (C2) is false in contexts in
which (C1) is true and vice versa, and that (C3) is true in all contexts.
Consider ordinary contexts first. Here, only some sorts of causal explana-
tions are relevant. Some are properly ignored. In such contexts (C1) expresses
a truth. Some actions do not have ordinarily relevant causal explanations.
But in such contexts (C2) is false. It says that all human actions are causally
determined, but in an ordinary context, this is true only if all human actions
have causal explanations among the sorts of explanations relevant in such
contexts. They do not. So (C2) is false in such contexts.
It might be a little better for Hawthorne’s account if we think of (C2) a lit-
tle differently than we have so far. It is not immediately clear that the phrase
‘causally determined’ has the sort of context sensitivity under discussion.
The key idea is that all human actions have a causal explanation involving
factors the agent cannot control. But the domain of potential explanations
varies with context. In some settings, when only ordinary explanations
count, there is no such explanation for some actions. But when we expand
the domain to include microphysical and ancestral explanations, then every
action does have a causal explanation involving factors beyond the agent’s
control.
Consider next philosophical contexts in which a much larger set of
causal explanations have been made relevant. In such contexts, (C1) ex-
presses a falsehood, on the assumption that there is some causal explana-
tion for all behavior. In such contexts, (C2) is true. Again, (C3) is true.
It may be that uttering a sentence like (C2) has the effect of making all
causal explanations relevant. That is, it may be that merely using a sen-
tence like (C2) puts us into a context in which higher standards for ap-
plying ‘free’ are in place. This, too, is analogous to the epistemic case.
Thinking about knowledge in terms of relevant alternatives makes the
analogy even clearer. Philosophy has the unhappy consequence of ex-
panding alternatives. Attributions of ‘knows’ require that the putative
knower be able to eliminate all relevant alternative possibilities. Thinking
philosophically about knowledge expands the set of alternatives we must
be able to eliminate so much that it includes remote skeptical hypotheses
that we are unable to eliminate. So, in such contexts, attributions of the
word ‘knows’ are typically false. Similarly, attributions of ‘free’ require that
there be no causal explanation of the action in question among the rele-
vant causes. Thinking about determinism can expand the set of relevant
causal explanations to include all the neurological and remote historical
explanations that we typically ignore. There always are causal explana-
tions in this expanded set. So, in such contexts, attributions of the word
‘free’ are typically false. Adapting Lewis’s remark about epistemology, we
can say that the idea is that metaphysics robs us of our freedom.
5 Contextualism as a Response to Skepticism
I find the contextualist solutions to these puzzles disappointing. In dis-

cussing my reasons for this, I want to separate questions about the truth of
the claim that sentences about knowledge and freedom are context sensitive
from questions about whether their context sensitivity is useful in resolving
or clarifying philosophical issues. I will emphasize the second point, after
briefly discussing the first.
266 R. Feldman
Is ‘Knows’ Context Sensitive?

One might doubt that the sentences under discussion really do have
context sensitivity of the sort contextualists say they have. Contextualists
contend that people’s inclination to predicate ‘knows’ or ‘free’ vary in
predictable ways. They offer semantical accounts that in effect track those
changes. Typically, their view is that when confronted with skeptical hy-
potheses, people are unwilling to predicate ‘knows’. It seems to me that the
facts about this are likely to be much more complex and variable than this
suggests. Some people are not moved by skeptical hypotheses. They dismiss
them as silly, frivolous, obviously false, and not worth taking seriously.
They continue to be willing to affirm knowledge sentences. Contextualists
must say that such people are somehow not keeping track of the relevant
standards and are mistaken.
Furthermore, contextualists must say that people lose track of their own
assertions in a surprising way. Suppose that in some ordinary setting you
utter the sentence ‘I know that I ate breakfast this morning’. Suppose that,
given contextualist views, this is true. You then go to philosophy class and
discuss some arguments concerning skepticism, and start to worry. You say,
‘I do not know that I ate breakfast this morning’. You might go on to say
that your earlier claim was mistaken, that you were overlooking important
skeptical considerations when you made your earlier claim. But on con-
textualist views you should say that the utterance you made earlier was
true. You should think at the later time that you were right earlier as well.
You should think that the case is something like one in which you say ‘I am
hungry’, then eat a big meal and say ‘I am not hungry’. There is no contra-
diction here, and, on contextualist views, the change in your claims
involving knowledge sentences is comparable. Or, to take an even more
similar case, if you first say of the baby elephant Jumbo, ‘He is big’, and
then see the adult elephants and say, ‘Jumbo is not big’, any worry that you
have contradicted yourself is easily set aside. You can explain that you
meant something like ‘big for an animal’ and ‘not big for an elephant’.
There is no contradiction. Knowledge attributions do not seem similar. It
seems, at least to me, that when skeptical doubts affect me, I am entertain-
ing doubt about the very thing I confidently asserted at another time.2
There are alternative explanations of our varying inclinations regarding
knowledge sentences. Briefly, one possibility is that changes in circum-
stances make us more or less cautious about asserting one and the same
thing. When what turns on the truth of a sentence increases, there may be
some tendency to be more cautious about asserting it. Thus, it may be that
hearing about skeptical possibilities also has the effect of making us more
cautious about predicating ‘knows’ without a change in the meaning of
the word. In addition, saying that someone knows something suggests
that the person could defend the thing known from objections. (I do not
mean to suggest that knowledge implies having this ability.) By mention-
ing odd scenarios one calls to mind possible objections that may be harder
to respond to. Realization of this could make us less willing to attribute
knowledge. It need not change the truth value of such attributions.
It is also possible that consideration of skeptical hypotheses and argu-
ments affects our willingness to attribute knowledge for a different reason.
Reflecting on those hypotheses and arguments may make salient reasons
for thinking that we lack knowledge, reasons that are temporarily persua-
sive. There are some propositions that at least some of us do not have stable
attitudes about. Perhaps propositions ascribing knowledge are like that, so
we change our minds.
The contextualist’s claim that knowledge sentences are context sensitive
does not strike me as clearly true. There are plausible alternative explana-
tions of our varying inclinations.
Conceding Too Much to Skepticism

Contextualists sometimes describe themselves as presenting a response to
skepticism. It is true that contextualists do provide a way to make our ordi-
nary assertions using the word ‘knows’ true. Still, contextualists concede a
great deal to skeptics. This can be seen by considering a brief statement to
which, as I understand it, contextualists are committed. But before making
the statement, I must properly set the stage. I can do that by explicitly call-
ing attention to the skeptical hypotheses and arguments that have been
mentioned earlier in this paper. These hypotheses involve such possibilities
that I am a brain in a vat or the victim of an evil demon. I will now write
explicitly in the object language. What follows is, according to contextual-
ists, true.
We know hardly anything. Skepticism is the correct solution to the

puzzles about knowledge. (A1) and (B1) are false. We do not know where
we are, who the president is, or many of the things we typically believe.
Moreover, we do not know those things when we are at home with our
268 R. Feldman
families and have left our ordinary philosophical thoughts behind. We

know very little about the world around us.
You might think that contextualists are not committed to the part of this
about ordinary contexts. But they are. Contextualism holds that the truth
conditions for sentences using the word ‘knows’ are determined by the
speaker’s context. By mentioning skeptical hypotheses and arguments prior
to the passage highlighted above, I put us in what contextualists regard as a
high standards context. In that context, when we speak about ourselves and
others we should say that we all lack knowledge. If you thought otherwise,
you were confusing this version of contextualism with the version I men-
tioned and set aside at the outset, the version that said the putative knower’s
context was somehow significant.
Thus, the contextualist solution to skeptical problems is to concede that
the skeptics are right. That is, when we are in a philosophy talk or class and
we are discussing skepticism, the skeptics are right. But I think that is a mis-
take. As a ‘solution’ to the problem of skepticism, this is hardly edifying.
Now, this may be a largely verbal issue. Contextualists think that in the
current context the standards for attributing ‘knows’ are very high. We do
not meet those standards. And I agree that we do not meet the standards
they think are now in place.3 So contextualists might be able to console
me, even here in the current context, by saying that we do often meet the
standards required for sentences using the word ‘knows’ to be true in ordi-
nary contexts. Speaking casually and loosely, and ignoring the need for
the metalanguage, contextualists can say that we should not feel too bad
because we do meet the ordinary standards for knowledge. As a response to
skepticism, then, this is an old solution dressed up in new clothing. The old
solution said that there were two senses of ‘knows’. One was a strong sense,
and in that sense, we know little. The other is a weaker sense, and in that
sense we know much.4 And, oddly, it is their assertion of this latter claim
that leads to my central complaint about contextualism as a response to
skepticism.
Failing to Address Some Central Skeptical Worries

My main complaint is not that contextualism concedes too much to skep-
tics. Just the opposite. I think that the primary interesting problem about
skepticism is not a debate about whether we meet some outrageously high
standards for knowledge. We do not, and there’s not much more to be said
about that. I think that the interesting issue concerning skepticism has to
do with whether we actually do meet the ordinary, lower standards for
knowledge. The most powerful skeptical arguments conclude that we do
not meet those standards. It is thinking through that issue that strikes me
as the fundamental challenge posed by skepticism. And contextualism is
entirely irrelevant to that issue. I should be careful about what I mean by
this. Contextualist philosophers may have something to say about what the
ordinary standards for knowledge are and why it is true that we meet them.
That would be of value in responding to skeptical arguments designed to
show that we do not meet the ordinary standards. The fact that there are
some other extraordinarily high standards that we do not meet is simply
irrelevant to this fact.
Here is another way to put the point. One central part of the challenge of
skepticism arises from the need to respond to arguments alleging that our
reasons for believing much of what we ordinarily believe are not very good,
not good enough for knowledge and perhaps not good enough for some
lower level of justification. These arguments prompt us to consider carefully
the reasons we have for believing the things we typically believe. One such
argument asserts that our ordinary beliefs provide no better explanations of
our basic perceptual data than do rival explanations involving dreams,
hallucinations, evil gods, or computer simulations. Related arguments are
based on the claim, endorsed by Lewis, when he asks the rhetorical ques-
tion, “What (non-circular) argument supports our reliance on perception,
on memory, and on testimony?” (1996, 551).
There are in the philosophical literature a variety of responses to the
arguments I’ve just alluded to, none entirely satisfying. The existing views
seem to amount to stipulation that we do know what we think we know,
avowals of faith in the powers of human cognition, questionable appeals to
inference to the best explanation, the assertion that commonsense views
are ‘intrinsically rational’ and so on.
I think that some accounts of knowledge, those that drop justification or
good reasons as necessary conditions of knowledge, also contribute to less
than edifying responses to skepticism. They may make it true that we have
knowledge, but seem to concede that our reasons for our ordinary beliefs
are not as good as we thought they were. To concede this, but nevertheless
insist that we have knowledge is, as I see it, to win the battle but lose the
war. Skepticism is centrally about whether we have good reasons to believe
270 R. Feldman
what we do. If knowledge is attained by having beliefs that are causally

connected to the facts in the right ways, or by reliable means of belief
acquisition, then we may have knowledge without having good reasons.
But this in no way deals with skeptical questions about the quality of our
reasons.
Contextualist considerations simply do not have much to say about this.
To defend the claim that ordinary claims to knowledge (or good reasons)
are true, contextualists must avail themselves of one of the responses just
mentioned. The distinctive addition made by contextualists is the con-
tention that there are some other contexts in which higher standards are in
place, standards that we do not meet. Whether or not this is true, it is the
merits of the case concerning ordinary standards that matters most as a
response to skepticism.
6 Contextualist Solutions to Puzzles about Freedom
Points analogous to those just discussed apply to the contextualist solution

to the free will puzzle.
Is ‘Free’ Context Sensitive?

I will discuss this point only briefly. We can raise doubts about whether
‘free’ is context sensitive in the way Hawthorne’s contextualist theory
implies.
First, it seems to me that our intuitive judgments about whether we
are free do not vary in just the way Hawthorne’s account would predict.
Some people, at least, are not moved by arguments for incompatibilism or
the mention of deterministic explanations. Contextualists must say that
they are simply misunderstanding the language or the context when they
maintain that they are free even in the light of such considerations.
Second, when people are moved by considerations about determinism to
say that they are not free, I think that they typically regard this conclusion
as incompatible with their ordinary attributions of the word ‘free’. But if
the contextualist account is right, their denials in philosophy class are not
inconsistent with their affirmations outside of class. It’s surprising that it
often does not seem that way to competent speakers of the language.
This point may well be most compelling to incompatibilists. If contextual-
ism is right, then incompatibilists should nevertheless agree that ordinary
claims to be free are true. I doubt that they do. Those who get worried
when they encounter arguments for the lack of freedom, such as the conse-
quence argument, should, if contextualists are right, think that the conclu-
sion that argument leads them to is not the denial of what they were
inclined to assert prior to first hearing the argument. But, in my experience
at least, no one does.
Finally, there are alterative explanations of the varying intuitions and
linguistic inclinations contextualists say we have. It is possible that focus-
ing on certain facts or arguments causes varying reluctance to assert the
same thing. It is possible that varying salient considerations make us some-
times believe and sometimes deny the same thing. There is, at least, room
for debate about these matters. We can wonder what best explains these
varying intuitions.
There is, then, room to doubt that there is context sensitivity here at all.
This issue is much more complex than my treatment here touches on. One
complication is that the word ‘free’ is ambiguous as well. There is the sort
of metaphysical freedom philosophers concern themselves with, auton-
omy that some psychologists focus on, as well as political freedom. This
makes things even messier.
But suppose that ‘free’ is context sensitive. Does that help us see our way
through the issues?
Conceding Too Much to the Incompatibilists

It seems to me that contextualist considerations leave unanswered many of
the central questions at the heart of the controversy about freedom. Notice
that Hawthorne, in spite of describing his view as ‘compatibilist’ at one point,
is committed to almost everything incompatibilists say. Compatibilists—
real compatibilists as opposed to metalevel compatibilists like Hawthorne—
think that some actions are free even if they are causally explained by factors
that they are not ignoring.5 Incompatibilists disagree. Now let’s assume that
determinism is true. I call your attention to that fact, and particularly to the
fact that there are processes in your brain that causally explain all of your
behavior. I have thereby set the stage. According to contextualism, the
following is true:
No one ever acts freely. No one is morally responsible for their behavior.
No one could have ever done anything other than what they actually did.
Moreover, in ordinary situations when people are not thinking about
272 R. Feldman
processes in their brains that cause their behavior, they are not free. No
one is ever free. Freedom is an illusion.
The points here are exactly analogous to the points made about contextual-
ism and skepticism. If philosophical settings make the standards such that
the sentences ascribing ‘free’ are not true, then in philosophical settings
such as this one, the incompatibilists (and those who deny that we are free)
are right. The philosophical debate, which obviously occurs in philosophi-
cal contexts, is won by the incompatibilists. Indeed, the compatibilists lose
without much of an argument in Hawthorne’s account. They capitulate
from the outset.
Failing to Address Some Central Worries

Just as contextualism fails to address some central issues associated with
skepticism, I think that contextualism fails to address some of the central
philosophical questions associated with freedom. Real compatibilists think
that actions are free if and only if they are caused in the right way. So
they think the absence of causal explanations of the sort mentioned by
Hawthorne is not necessary for freedom. Further, and more crucial to my
current concerns, is that they think that the absence of a causal explanation
for an action is not sufficient for that action to be free. The problem here is
that, according to compatibilists and many incompatibilists, random ac-
tions are not free actions. So, a central question in the free will controversy is
how to make sense of the free actions. There is some inclination to think that
causally determined actions cannot be free, but also some inclination to
think that causally undetermined actions cannot be free. So one crucial part
of the free will controversy concerns whether any good sense can be made of
an indeterministic account of freedom.
Recall that Hawthorne’s account states only a necessary condition for
true ascriptions of ‘free’. He is explicit about the failure of this as a sufficient
condition. And this is for good reason. Actions that are completely random
would also satisfy the specified necessary condition. Yet they are not free
actions, in the relevant sense of ‘free’. A question then arises as to what
Hawthorne might add to his account in order to state sufficient conditions
for free actions.
In the epistemological case my contention was that contextualists need
some account of why we do in fact satisfy the ordinary standards for knowl-
edge (or justification). Contextualism by itself did little to help here. It could
work mainly by making use of some independent account, involving best

explanations, intrinsic rationality, or stipulation. The situation with respect
to contextualism and freedom is similar, though perhaps worse.
What might Hawthorne say are sufficient conditions for free action?
Could he make use of existing accounts, in the way contextualists about
knowledge can make use of existing fallibilist accounts of justification? One
standard view holds that actions are free when they are agent caused. But if
free actions are agent caused, then they are agent caused even when all ex-
planations are considered. Contextualism is irrelevant here. Our actions are
correctly described as ‘free’, or else as ‘not free’ depending on whether
or not we are agent causes. Context shifts would make no difference. So
Hawthorne cannot add a condition requiring agent causation to get a suffi-
cient condition for free action.
Hawthorne might add to his stated condition a second condition requir-
ing that the action be caused by a decision of the agent, where this decision
is free from causal explainers outside the agent’s control. But there are ex-
amples of random decisions, not under the agent’s control, that result in
actions that are not free. This condition is not sufficient for free action.
Hawthorne might try adding a condition requiring that the action be
caused by a decision of the agent, where this decision is under the agent’s
control. Given my compatibilist inclinations, this seems fine to me. It is ex-
actly what real compatibilists say. There are, of course, hard questions about
exactly what it means to say that a decision is under an agent’s control. Some
might think that this is a large part of what the free will problem is about
in the first place. It is hard to see that contextualism sheds any light on it.
But, furthermore, it is hard to see why one should not say that this condi-
tion, whatever exactly it is, is satisfied even if all causal explanations are
considered. That is, it is hard to see why this condition is not satisfied in
philosophical contexts in which we do consider ancestral and neural expla-
nations. In other words, if Hawthorne were to accept this account of when
an action is free, he should just accept real compatibilism. Contextualism
itself contributes little, or nothing, to an explanation of how actions can
be free.
As I see it, among the central philosophical questions arising in connec-
tion with freedom and determinism are these: is the (or a) sense of the word
‘free’ such that it can apply to actions even though they are determined?
Is this sense one that properly grounds ascriptions of responsibility and
274 R. Feldman
control? Is it one that properly grounds assessment of blame and the appli-
cation of punishment? Another set of questions arises for incompatibilists,
especially if the answers to the questions just listed are negative. Is the (or a)
sense of the word ‘free’ such that it can apply to actions that are not causally
determined? Is this sense one that properly grounds ascriptions of responsi-
bility and control, that properly grounds attributions of blame and the
imposition of punishment?
I find little of any help in thinking about these matters coming out
of contextualist considerations. Contextualism, as developed here, appar-
ently yields negative judgments about our freedom in philosophical
contexts. But it says little about why this is true, other than to endorse in-
compatibilist intuitions. It fails to engage considerations that support com-
patibilism. Furthermore, it is not really designed to explain or support an
understanding of ‘free’ that would help incompatibilists. Indeed, as I un-
derstand it, it does not help with this at all, since it does not address
the question of how causally undetermined actions could be other than
random.
7 Conclusion
Part of the problem with contextualism, both with respect to freedom and
with respect to skepticism, seems to me to arise because of the way contex-
tualists pose the problems. The problems are said to be explaining our
seemingly varying intuitions. We sometimes say one thing and we some-
times say another. Perhaps our intuitions do vary, and what we are inclined
to say varies accordingly. If so, these facts warrant explanations. But there
is another fact that warrants attention. It is that in philosophical contexts
in which all the relevant causal (or skeptical) alternatives have been made
salient, there are puzzles about exactly what to say or think concerning our
freedom or our knowledge. There are competing considerations all at one
time, and the hard question is to figure out what to make of them. Con-
textualism, at least as I’ve described it here, does not really address this in
any developed way. It just concedes the issues to the incompatibilists and
the skeptics.
The denials of incompatibilism and skepticism—compatibilism and
fallibilism—are in certain ways analogous. One such way is that their
advocates are willing to assert the key sentences in the face of all the alter-
natives. When confronted with skeptical possibilities, fallibilists remain
willing to say ‘I know’. Similarly, when confronted with all the possible
causal explanations of behavior, compatibilists are willing to say, ‘I am
free’. They have things to say in support of those attitudes. Contextualism,
as presented here, simply says that in the contexts in which they speak,
they are wrong. Contextualists add that those very sentences express truths
in ordinary contexts, but they have little to say about why that is the case.
I conclude that contextualism may at best provide ways to explain some
varying linguistic dispositions and may contribute to ways to retain the
truth of some ordinary utterances. But it saves our ordinary utterances
at the expense of addressing some of the central philosophical arguments
that lead people to question those utterances in the first place. Words such
as ‘knows’ and ‘free’ may be context sensitive. The philosophical issues
remain unresolved.
Acknowledgments
Earlier versions of this paper were presented at the Inland Northwest

Philosophy Conference, Saint Cloud State University, and Cornell Univer-
sity. Peter Murphy commented on the first occasion and Michael Fara on
the third. I am grateful to them, as well as to members of the audiences, for
helpful comments.
Notes
1. Hawthorne does not explicitly apply his contextualist solution to the inconsis-
tent triad under discussion here. He does explain the varying truth conditions for
sentences containing the word ‘free’. Some of the details in my account of the
resolution of the puzzle are based on my way of extending his remarks to the puzzle.
2. See Schiffer (1996) for a defense of a somewhat similar view.
3. This depends, of course, on the details of just what the current standards for
knowledge are.
4. See, for example, Malcolm (1952).
5. Metalevel compatibilism is the thesis that ordinary assertions made by saying

‘S did X freely’ are compatible with determinism.
276 R. Feldman
References
Annis, D. 1978. “A Contextualist Theory of Epistemic Justification.” American

Cohen, S. 1987. “Knowledge, Context, and Social Standards.” Synthese 73: 3–26.
. 1988. “How To Be a Fallibilist.” In J. Tomberlin, ed., Philosophical Perspectives

2: Epistemology. Atascadero, Calif.: Ridgeview.
. 1999. “Contextualism, Skepticism, and the Structure of Reasons.” In J.

Tomberlin, ed., Philosophical Perspectives 13: Epistemology. Atascadero, Calif.: Ridgeview.
DeRose, K. 1995. “Solving the Skeptical Problem.” Philosophical Review 104: 1–52.
Feldman, R. 1999. “Contextualism and Skepticism.” In J. Tomberlin, ed., Philosophi-

cal Perspectives 13: Epistemology. Atascadero, Calif.: Ridgeview.
. 2001. “Skeptical Problems, Contextualist Solutions.” Philosophical Studies

103: 61–85.
Hawthorne, J. 2001. “Freedom in Context.” Philosophical Studies 104: 63–79.
Horgan, T. 1979. “‘Could’, Possible Worlds, and Moral Responsibility.” Southern

Journal of Philosophy 17: 345–358.
Lewis, D. 1996. “Elusive Knowledge.” Australasian Journal of Philosophy 74: 549–567.
Malcolm, N. 1952. “Knowledge and Belief.” Mind 51: 178–189.
Schiffer, S. 1996. “Contextualist Solutions to Scepticism.” Proceedings of the

Aristotelian Society 96: 317–333.
13 Buddhism and the Freedom of the Will: Pali and
Mahayanist Responses
Nicholas F. Gier and Paul Kjellberg
Without cognising free will as a philosophical problem, [the Buddha] takes it for
granted that the innate character of each being leaves him the freedom to decide
about the actions that determine his future.
—H. W. Schumann (1973, 53)
There is free action, there is retribution, but I see no agent that passes out of one
set of momentary elements into another one, except the [connection] of those
elements.
—The Buddha1
[Aristotle] did not do too badly without [the will].

—W. F. R. Hardie2
[Nagarjuna’s philosophy] is not an incoherent mysticism, but it is a logical tightrope

act at the very limits of language and metaphysics.
—Jay L. Garfield (1995, 102)
Buddhist philosophy generally divides itself into an early stage in which

most of the texts are written in Pali, a language closely related to Sanskrit,
and a later Mahayanist movement based on writings in Sanskrit. Those who
favor a Mahayanist perspective are sensitive about the adjective ‘early’, be-
cause they believe that the roots of Mahayana are just as old as the Pali
texts, which were not written down until circa 80 B.C.E. Pali Buddhist phi-
losophy is generally empiricist and realist in its epistemology and ontology,
whereas most Mahayanist philosophers embraced various forms of idealism
and even skepticism. Buddhists, along with most all ancient philosophers,
embrace a universal determinism of cause and effect, but they have never
acknowledged the problem of free will, which, as we will argue in the first
section, appears to be a distinctively modernist phenomenon. For purposes
278 N. F. Gier and P. Kjellberg
of this discussion, free will is defined as a power truly our own independent
from causal determinants.
In sections 2–5 we will analyze the issues of this book from the Pali per-
spective. Here we will find a more robust view of the self, best interpreted
in functionalist terms, and also a realist view of causality. We will argue
that these Buddhists would join the compatiblists of this book in their
defense of moral responsibility. When we turn to the Mahayana schools, we
will discover that it is much further removed from the free will–determinism
debate. Mahayanists take the Buddha’s idea of ‘no self’ much more radi-
cally and they also generally reject a realist view of causality.
In the last two sections we will offer two views of Nagarjuna, one from ‘con-
structive postmodernism’ as opposed to ‘deconstructive postmodernism’
with which Nagarjuna is usually associated. (Because of space limitations, we
will just briefly mention Yogacara idealism, the other major Mahayanist
school.) While we would like to support a constructive postmodernist
Nagarjuna, the texts actually do not allow us to do this. This means that while
Pali Buddhists can embrace real agency in the world, Nagarjuna and his
followers appear to argue that any positive view of the self will prevent us
from escaping the cycles of existence. In this view the self is completely
deconstructed and all that we are left with are competing rhetorics of free-
dom and determinism.
We are delighted that we have been invited to contribute to this book
and to offer a Buddhist perspective on these fundamental issues. We believe
that Euro-American philosophers have much to learn from the Asian
philosophical tradition. Many of them are still unaware of the degree of
philosophical sophistication Asian thinkers have attained. Finally, we want
to emphasize that we are only teachers of Buddhism and not experts with
knowledge of Pali or Sanskrit and all their voluminous texts. We rely heav-
ily on the best work in the field and trust that we have used it accurately
and responsibly.
1 Free Will as a Modernist Phenomenon
One possible reason for the failure of Asian philosophers to address the is-
sues of this book is the common observation that Asian thinkers, especially
those of the Indian subcontinent, are far more interested in absolute spiri-
tual freedom than freedom of the will. Rather than a freedom somehow
Buddhism and Freedom of the Will 279
related to the world of cause and effect, the Indian yogis appear mostly
concerned with a freedom that transcends the physical world entirely.
With this vision in mind the casual determination of karma is not a prob-
lem at all, as karmic bondage is a state of all beings until final liberation.
Hindu scripture describes the saints as veritable supermen. For example,
the Taittiriya Upanishad tells us that the yogi “attains . . . independent
sovereignty,” and enjoys a bliss that is a billion times greater than that of
the highest gods (1.6, 2.8). In the Maitri Upanishad the ascetic surpasses
Brahman, the Godhead, and “will go [yet further], he [will surpass] the
gods in the realm of divinity. . . .” (4.4). In the Shvetashvatara Upanishad
yogis gain incredible powers: they “shall roll up space as if it were a piece of
leather” (6.20); and a yogi in the Taittiriya Upanishad boasts that “I am the
first-born of the world-order, earlier than the gods, in the navel of immor-
tality. . . . I have overcome the whole world” (3.5). Such a view has been
called ‘spiritual Titanism’, an extreme form of humanism in which humans
take on divine attributes and prerogatives (Gier 2000).
In the Pali texts the Buddha rejects these incredible claims of the Hindu
and Jain yogis. He was particularly critical of their claims to omniscience.
He did, however, embrace their subordination of the gods and the require-
ment that they had to be reincarnated as humans in order to be liberated.
Therefore, the Buddha and most of his followers are not spiritual Titans,
primarily because they rejected the divinity of the saint and sought
Nirvana in this world rather than in some otherworldly domain. The
Buddha also believed that the body was constitutive of personal identity
(Sankhya-Yoga and Jain dualists rejected this) and that the emotions and
senses were not evil. Finally, the Buddha distinguished between Nirvana in
the body and Nirvana at the end of the cycles of existence. Because the
Buddha was a strict empiricist and because final Nirvana was beyond the
experience of anyone, he declined to say anything about it at all. However,
the Buddha and many of his disciples presumably did live in a constant
state of embodied Nirvana and the best one-word definition of this state is
‘freedom’. We will analyze the meaning of this freedom in compatiblist
terms in the fourth section.
There is a deeper and more philosophically interesting reason why the
Buddha would have found free will a nonissue. This may be the same rea-
son that Greeks and Romans generally did not find it a problem.3 It is also
connected to why American pragmatism did not address the issue, and
why comparisons of Wittgenstein and the Buddha have become a veritable

scholarly industry.4 None of these thinkers divided up the world in the
ways discussed below, specifically a distinctively modernist separation of
the ‘inner’ and the ‘outer’ that produces the conflict between an internal
freedom and external causality.
Intellectual historians are now getting a better idea about why we find
the first discussion of free will, as Euro-Americans now debate it, in the
writings of Augustine of Hippo. This fact is supremely ironic: Augustine be-
lieved in the absolute sovereignty of God and the corollary doctrine of
divine omnicausality. If free will requires a power truly our own, then
Augustine’s God eliminates any possibility of such freedom. In fact, no me-
dieval Christian philosopher, not even Aquinas, solved this basic problem.5
In the Condemnations of 1277, the Church declared that God’s power was
limited only by law of contradiction. The full implications of this view
were elaborated by William of Ockham, who declared that it was logically
possible (although not very probable) that while we assume that every-
thing operates according to the laws of causality, God could in fact be caus-
ing all effects directly out of his potentia absoluta. It is important to note
that, except for some Stoic discussions, divine power was generally not an
issue for Asians nor was it for Greek and Roman philosophers.
Was the challenge posed by divine omnipotence, the most radical idea of
divine power in the history of religion East or West, the catalyst for prob-
lematizing human will power and its freedom? (Commentators have also
speculated that the paucity of psychological terms in Augustine’s Latin
may have hindered his analysis. Sanskrit is much richer in this regard.) This
could very well be the historical clue for the later development of distinc-
tively modernist forms of thought. With Ockham the via moderna was
in full bloom and Aquinas’s natural theology was in serious trouble. The
resultant split in reason and faith gradually led to the other dichotomies of
modernism, the seeds of which were planted in the late medieval period.
Luther’s nominalist teachers would have flinched if they had known that
he called reason “a whore,” but they had opened the floodgates for the rad-
ical fideism of the Reformation. Faith returned to the inner world, while
reason found new triumphs in empirical science. The basic issues of early
modern European philosophy arose out of this intellectual milieu.
Modernism gave new meaning to what it means to be a subject, and
the primary source of this innovation was the ego cogito of Descartes’s
Meditations. The pre-Cartesian meaning of ‘subject’ (Gk. hypokeimenon;

Lat. subiectum) can still be seen in the subjects one takes in school or the
subject of a sentence. In this ancient sense all things are subjects, things
with ‘underlying [essential] kernels’, as the Greek literally says and as Greek
metaphysics proposed. After Cartesian doubt, however, there is only one
subject of experience of which we are certain—namely, the human think-
ing subject. All other things in the world, including persons and other
sentient beings, have now become objects of our thought, not subjects in
their own right. Cartesian subjectivism gave birth simultaneously to mod-
ern objectivism as well; and, with the influence of the new mechanical
cosmology, the stage was set for uniquely modern forms of otherness and
alienation.
2 The Buddha and Modernist Dichotomies
Modernism is a form of thought that loves to dichtomize. It separates sub-

jects from objects, the inner from the outer, the private from the public,
fact from value, religion from science.6 (Making these distinctions has great
advantages but also, as postmodern critics have shown, profound liabilities
as well.) If the freedom of the will is something subjective and causality is
something objective, and if free will happens only in an internal realm and
cannot happen in an outer realm of cause and effect, then free will and
moral responsibility are indeed problems of supreme significance. It is no
accident, then, that modern philosophy generated other related problems
as well. The issue of the freedom of the will was joined with the problem of
the ontological status of the external world, the problem of the knowledge
of other minds, and the rejection of the idea of moral facts—in sum, the
table of contents of an introductory text in modern European philosophy.
None of these problems and none of these modernist distinctions appear
in the voluminous records of the Buddha’s forty-five years of daily philo-
sophical conversations. Was he philosophically naive or was he right in his
assumptions about the nature of reality and knowledge? If one, for exam-
ple, does not make a firm distinction between the inner and the outer, then
there can be no talk about free events inside us and determined events out-
side of us. Neither can there be a problem of the ontological status of
the external world and the skeptical impasses that arise from this. The
Buddha’s empiricism was first compared to Hume’s, but the most accurate
parallel is to James’s radical empiricism. James and the Buddha observed that
basic experience does not divide into inner and outer; rather, the inner flows
into the outer and the outer flows into the inner. (One could perhaps read
Hume in the same way.) It is only by some Cartesian method of systematic
doubt that an inner world of ideas and perceptions is separated from an
outer world of physical things. The Buddha and James also claimed that
basic experience does not divide into facts and values, because as the Buddha
said: “What one feels, one perceives; what one perceives one reasons
about.”7 What one feels is obviously filled with values and emotions.
Critics might agree with James and the Buddha that there are no onto-
logical bifurcations in our immediate experience, but they still might ob-
ject that even though their ‘inner’ flows into their ‘outer’, their ‘inner’ does
not flow into another’s ‘inner’. The Buddhists disagree but for a reason that
many Euro-American philosophers will reject outright. The Buddha and his
disciples claimed to have ESP powers and they said that they could read the
contents of other people’s minds. Specifically, the Buddha claimed to be
able to determine the balance on anyone’s karmic mortgage.
Quite apart from the validity of ESP, one could argue that we do in fact
read other people’s minds through their body language, as many people do
in fact carry their emotions on their sleeves. (Is this what Wittgenstein
meant by “meaning is a physiognomy”?) The Buddhists would say that this
minimal capacity that all of us have is simply the potential for anyone
to use the faculty of mindfulness to a very high and sensitive degree. Al-
though usually referred to as a virtue, mindfulness (sati) is more accurately
called the faculty (indriya) by which Buddhists control the six senses (the
mind is included) and their moral development.
3 Buddhist Compatibilism
Earlier we distinguished between Nirvana at the end of existence and

Nirvana while embodied, which can be described as state of contentment
(sukha as an equivalent to the Greek eudaimonia) and as freedom from all
craving. In the terms of Keith Lehrer’s analysis in this volume, the Buddhist
is free from distracting desires (Lehrer’s example is a craving for chocolate)
and disciplines herself, in a way similar to Lehrer’s writer, to act on her spir-
itual preferences. Lehrer claims that “our preferences are the source of our
freedom,” and Buddhists would agree that we empower them as well as
they empower us. Buddhists would also agree with Lehrer that what we pre-
fer is always our option while what we desire wells up in us involuntarily.
By giving in to her cravings the Buddhist would continue the cycles of
existence forever, something she, or any other clear-minded person, does
not prefer. By choosing meditation and other spiritual disciplines as prefer-
ences, she prepares herself for the ultimate preference: Nirvana and free-
dom from Samsara (the cycle of birth, death, and rebirth, of which Nirvana
is the end). Furthermore, note that this Buddhist is an agent of her prefer-
ences, as Lehrer explains, rather than being just a passive victim of her
desires. Finally, with regard to the problem of preference control, the non-
theist Buddhist does not have to contend with an omnicausal deity, who,
according to Augustine and Luther, empowers us to turn to God as well
as to turn away from divine grace. Most Buddhists also do not have to con-
tend with mad neurophilosophers from Southern California to control
their preferences.
While the issue of free will does not arise in Buddhism, it is indisputable
that it embraces a universal determinism: every effect, without exception,
has a cause. The idea that the will is uncaused or is self-caused violates the
Buddhist principle of interdependent coorigination (prattiyasumutpada):
nothing in the universe can originate itself as substances allegedly do or
the will is said to do. Buddhist causality, however, is seen as a cosmic web
of causal conditions rather than linear and mechanical notions of push-
pull causation. Furthermore, the Buddha claimed that we are morally re-
sponsible only for those actions that we intend. He took strong exception
to the Jain theory that we suffer from accidental karma, such as stepping on
a bug that we do not see. The Jains, another Indian religion contemporary
with Buddhism, charged that the Buddha’s qualified determinism would
lead to antinomianism and ethical subjectivism. Only their strict deter-
minism, they claimed, would maintain objective ethical standards.
The Buddha countered, however, that if the mind is always subject to the
control of the past, then there could never be any liberation, nor could there
be any moral responsibility. The Buddha proposed that the ‘freedom of
the mind’ works in the following way: “If I were to oppose the formation of
the cause of sorrow, by opposing this formation I would become dispassion-
ate. Also, if I were to become evenminded in respect to this cause of sorrow,
if I were to develop this evenmindedness, I would become dispassionate.”8
The Buddha also claimed that personal temperament and circumstances
qualify the effects of past causes such that there is no strict correspondence
between cause and effect. The most famous example is the analogy of a lump
of salt, first placed in a cup of water (representing the corrupt person) and
then placed in a lake (representing the Buddhist saint). The effect in the for-
mer is obviously different from the latter. This example has become contro-
versial, because the salt is usually seen as an evil intention, having disastrous
consequences in the sinful person but virtually no effect on the saint. The
Jains’ fears appear to be vindicated: can the saint commit a sin and not be
harmed or even blamed for it?
Pali Buddhist ethics can be called a contextual pragmatism, which can be
best explained by the famous motto “They who know causation (prattiya-
sumutpada) know the Dharma.”9 This can be read as they who know their
own causal web of existence know the truth (i.e., the true facts of their
lives) and they will know what to do. The truths we discover will be very
personal truths, moral and spiritual truths that are, as Aristotle says,
“relative to us.” This is not simply a cognitive knowing but a practical grasp
of what is appropriate and what is fitting for us and our surroundings.
Being mindful is deliberately forming preferences over the desires that
might lead us out of our own personal means and the virtue we have de-
veloped there. Like phronesis, Buddhist mindfulness is primarily nonsensu-
ous, correct perception. Note also the Buddha’s ethical naturalism,
outlawed in European philosophy after Hume, and the bold fusion of fact
and value, a topic we will not pursue further.
Both Aristotle and the Buddha thought it was always wrong to eat too
much, but each person will find his or her own relative mean between
eating too much and eating too little. Such a view is not relativistic in the
nonnormative sense, because the principal determinants for eating just
right are primarily objective not subjective. If people ignore these objective
factors—for example, temperament, body size, metabolism, and other
physiological factors—then their bodies, sooner or later, will tell them that
they are out of their respective means. If people are unmindful, they allow
unhealthy eating habits to take them in one direction or another, but the
mindful stay within their respective means. Therefore, the mindful ones
are free, while the ones tending to either extreme are not.
We can now define Buddhist free agents as those who are keenly aware
of the effects their actions have on themselves and others. They are free
from ego attachment and craving either for ascetic deficiency or indulgent
excess, representing karmic bondage rather than karmic freedom. Free

and mindful agents know what their needs are and what their preferences
should be; and, on the basis of that knowledge, they can separate de-
sires from cravings, defined as desires that either cannot be fulfilled, or for
things that are simply not needed. Moral freedom lies in the ability of agents
to form desires that are consonant with their needs and personal circum-
stances. In terms of the contemporary free will debate, the Buddhists believe
in ‘free action’ but have no conception of ‘free will’, as a self-determining
power that moral agents somehow possess.
For the Buddha the deepest and subtlest craving is the desire for a self-
determining and independent self—in a word, a soul substance. It is a
supreme irony that what European philosophers assumed is necessary for
true human freedom is actually the cause of its greatest bondage. If we qual-
ify the term properly within the context of Buddhist relationality, the word
‘autonomy’ can be used to describe Buddhist agents who are not controlled
by craving; rather, they live freely in a personal mean between the extremes
of debauchery and self-mortification. Interestingly enough, liberal political
theorists, taking seriously conservative critiques of social atomism, are now
proposing the idea of a ‘situated’ autonomy.10 Although Nomy Arpaly does
not discuss this idea of autonomy in her chapter in this book, it is clear that
Pali Buddhists would affirm her concept of agent autonomy, where agents
learn to control their cravings and act on their preferencecs.
4 Causality, Conditionality, and Moral Responsibility
The Buddha’s qualified determinism has led David Kalupahana to speak of

Buddhist ‘conditionality’ rather than causality.11 For him Buddhist condi-
tionality represents a middle way between strict determinism and the
destruction of freedom on the one hand, and an equally unacceptable
indeterminism on the other. Because the Buddha rejected both material
and mental substances, Buddhist conditionality amounts to causality with
substance metaphysics. We should envision, as we have been forced to do
in contemporary physics, all events conditioning one another rather than
physical and mental causes pushing, pulling, or otherwise interacting with
one another. Incredibly enough, action at a distance has now been con-
firmed between pairs of subatomic particles, although physicists still do not
understand how this can happen. Therefore, Buddhist conditionality could
possibly be used to explain the actions of subatomic particles, which are

currently impossible to understand according to classical science’s view of
causality. In fact, Jay Garfield has recommended Buddhist conditionality as
a unifying theory for philosophy of science.12
Buddhist conditionality is summarized by the following formulas: “When
this is present, that comes to be; from the arising of this, that arises. When
this is absent, that does not come to be; on the cessation of this, that
ceases.”13 Moving from facts to values, the principle of conditionality is
summarized as a twelve-fold chain starting with ignorance, then unmindful
action, a resultant distorted consciousness, and then nine other conditions
that lead to rebirth. If anyone of these conditions is not present, then rebirth
in a next life will not happen. One might see it in terms of Aristotle’s formal
causation, as the following authors do: “As a theory of causation, this ‘de-
pendent coarising’ concerns the formal concomitances among things rather
than their material derivation from one another. It resembles a medical
diagnosis in several ways. By showing that the ailment depends on a series
of conditions, it indicates the point at which the series can be broken and so
facilitates a cure” (Robinson and Johnson 1992, 17).
The language of causality tends to simplify the explanation of an effect,
while the language of conditionality makes it much more complex. The
doctrine of interdependent coorigination compels the Buddhist to take
a much more comprehensive view of causality. The full complement
(samagri) of conditions must be present in order for an event to happen and
for us to fully understand how it happened. Taking an example offered by
Edward Conze, we might say that a bullet was the cause of a man’s death.
If we then think about the conditions of his death, there might be no end
to the search for relevant conditions, including the fact that the sun pro-
vided the light for the killer to shoot. Conze offers another example: while
sunspots cannot be said to cause economic crises, their occurrence does
indeed appear to coincide with and therefore condition them. Scientific
causality is particularly circumscribed, as Conze explains: “. . . the modern
idea of causality is governed by the ideal of prediction. The concrete total-
ity of events is set aside, certain sectors are ‘isolated’ and observed on their
own, with the intention of ‘controlling’ events” (1962, 149).
Conze brings the discussion back to the basic issue of modernism:
“When [modern scientists] speak of a cause they mean the general cause of
this kind of event, taken in the abstract, whereas the Buddhists are inter-
ested in the concrete conditions of this particular concrete event.”14 There-
fore, Conze concludes that “the Buddhist doctrine of the multiplicity of
conditions seems to make a decision on the ‘freedom of the will’ unneces-
sary. If the total number of conditions is unlimited, and most of them are
unknown, it is impossible to say which condition of necessity brings about
which event” (1962, 146 fn.).
We partially disagree with Conze. It is clear that the Buddha did indeed
reject the existence of any self-causing agents, but did affirm that all events
have a multiplicity of conditions. Therefore, Pali Buddhism falls under the
general rubric of compatiblism and ‘soft’ determinism. These Buddhists be-
lieve that we are morally responsible for our own character and intentions,
which although completely conditioned by antecedent events, are nonethe-
less what we truly want and should do. Recall that the Buddha claims
that we are responsible only for those actions that we intend. In Sanskrit
karma means ‘action’, so it follows that Buddhist karma is volitional action
only.
Seeing the law of karma as a psychological law allows us to avoid both
the extravagance and absurdities of the common view of it as a cosmic
law with inscrutable laws of retribution. (It also means that we may also
see, as some Buddhists do, the six realms of existence as a metaphor for
the ‘animal’, ‘angelic’, and ‘demonic’ moments of one single life span.)15
Therefore, the law of karma can be conceived as the rather trivial truth that
all actions have consequences. Returning to causality as conditionality, we
can now state the following conditionals concerning moral responsibility:
“If we act motivated by greed, hatred, or delusion, we are planting the seed
of suffering; but when our acts are motivated by generosity, love, or wis-
dom, then we are creating the karmic conditions for abundance and hap-
piness.”16 For the Buddhists, karma works at two levels—one immediate
and one delayed. In any of our acts we can immediately experience the
results depending on whether they were done, for example, out of love or
hatred (of which only the truly obtuse person will claim to be unaware).
Later on, these seeds of our actions will produce their inevitable fruits and,
following the principle of interdependent coorigination, these fruits will
finally ripen. The ripening of karmic action is a pervasive metaphor in all
Buddhist literature.
5 The Pali Self as Functionalist
The Buddha’s response to the axial age’s discovery of the self was strikingly
unique: he proposed the doctrine of no-self (anatman), which literally
means no atman, the Hindu soul substance. This conceptual innovation
was so provocative that it was bound to invite misinterpretation, and un-
founded charges of Buddhist “nihilism” continue even to this day. The
Buddha anticipated Hume’s view that the self is an ensemble of feelings, per-
ceptions, dispositions, and awareness (the skandhas) that is the center of
agency and moral responsibility.17 The Buddha’s view, however, is different
from Hume’s (1978), primarily because the Buddha appeared to support real
causal efficacy among internally related phenomena. (We believe that Hume
may have been misled by the current scientific model of externally related
atoms.) While Hume deconstructed any theory of causality, the Buddha
reconstructed causal relations with his theory of interdependent coorigina-
tion. The Buddha agrees with Hume about the absence of causal power but
disagrees with him about the absence of causal relations. As the Pali philoso-
pher Buddhaghosa said: “There is no real production; there is only interde-
pendence.”18 This is the same as saying that causality has been replaced by
conditionality.
The Buddha rejected the soul-as-spiritual-substance view of the
Upanishads, Jainism, and Sankhya-Yoga, and he deconstructed the ‘specta-
tor’ self of these philosophies 2,500 years before recent thinkers dismantled
the Cartesian self. As opposed to strict deconstruction, for example, Pali
Buddhists hold that selves, though neither the same nor different through-
out their lives, are nevertheless responsible for their actions. (Pali Buddhism,
therefore, should be aligned with the school of constructive postmod-
ernism.) These selves are also real in the sense that they are constituted by
relations with their bodies, other selves, and all other entities. This is why the
Pali self should be viewed in relational or process terms rather than the
skeptical implications of the no-self doctrine, which many later Buddhists
supported. The Pali self is relational primarily in the sense of its dependence
on the five skandhas and the internal relations this dependence entails.
Another positive way to express nonsubstantiality is to describe the
Buddhist self as ‘functional’. In fact, each of the skandhas should be seen as
functions rather than entities. On this point, Kalupahana makes good use of
James, who while denying a soul substance, maintained that consciousness
is a function. As Kalupahana states: “Rupa or material form accounts for the

function of identification; vedana or feeling and samjna or perception repre-
sent the function of experience, emotive as well as cognitive; sanskara or
disposition stands for the function of individuation; vijnana or consciousness
explains the function of continuity in experience.”19 Both Kalupahana and
Peter Harvey describe the Pali self in the positive terms of psychophysical
unity, process, and interrelation. According to Harvey, the Buddha never
rejected the existence of a life principle (jiva), which “is not a separate part
of a person, but is a process which occurs when certain conditions are
present . . .” (1993, 31). Not only are there significant parallels to James,
but the Buddha’s process self compares favorably to that of today’s process
philosophers, following in the footsteps of Whitehead and Hartshorne, who
are also the founding fathers of constructive postmodernism.
From this analysis we can clearly see that the Pali self is a robust personal
agent fully capable of maintaining its personal integrity and taking full re-
sponsibility for its actions. (This of course assumes the truth of compatib-
lism.) This view of the self is also fully somatic, giving full value to the body
and the emotions. At the same time it is embedded in a social and organic
nexus of cosmic relations. Hindu philosopher Surendra Verma is unduly
puzzled when he asks how it was possible for the Buddha to be filled with
thoughts and emotions and “at the same time preaching . . . the nonexis-
tence of the soul” (1970, 7). Like many other commentators, Verma simply
does not understand the meaning of the Buddha’s Middle Way, in this case
the mean between annihilationism (no self at all—substantial or otherwise)
on the one hand and eternalism (substantial self) on the other. What appears
not only puzzling but impossible is for the Hindu atman or a Stoic soul—
pure spiritual substances all—to have any relation at all with the finite
world, let alone with the emotions and the body.
In his very instructive comparison between Aristotle and the Buddha,
Damien Keown proposes that the function of moral choice is found in
prohairesis and cetana respectively. Both of these terms have been con-
nected with the European will, but no simple identity can be assumed. (The
etymology of cetana gives the root as cit, which means ‘to think’, and
its basic meaning is ‘visible’ or ‘distinguished’, as in that which appears in
the mind.) Both of them are neither emotive nor cognitive; rather, they
operate as a fusion of the two. Aristotle’s description of prohairesis as either
“intelligence motivated by desire or desire operating through thought”20
can also be applied to the Buddhist cetana, which, as Keown states,

“would . . . embrace a continuum that runs from predisposition through
choice to action” (1992, 218). As one of our readers suggested, cetana com-
bines thinking and desire, a project that the Japanese Buddhist Nishida
attempts in An Inquiry Into the Good.21 Both Aristotle and the Buddha refuse
to dichotomize the self and to compartmentalize a flow of experience that
resists such divisions. That is why cetana is sometimes identified with
karma itself (as we have seen, deliberate intentions are the only karmic
actions) or with the skandha of sanskara, a person’s dispositions that carry
karma from one life to another. Therefore, both Aristotelian and Buddhist
philosophy operates very nicely without a concept of the will. Such a strat-
egy does better justice to experience and does not generate unnecessary
philosophical problems, a primary one being the freedom of an alleged
independent, self-determining will.
6 Nagarjuna and the Self
A South Indian philosopher sometimes called the “second Buddha,”

Nagarjuna stands ambivalently at the beginnings of Mahayana philosophy.
His arguments are subtle, sophisticated, and complex. Their interpretation
is made more difficult because they are written as beautifully crafted
quatrains whose density require extra concentration and persistence. As a
transitional thinker, Nagarjuna is taken by some commentators (David
Kalupahana and Jay Garfield) as continuing the Pali tradition and by others
(Frederick Streng and T. Wood) as making a clean break with it. The latter
school of interpretation reads Nagarjuna’s quatrains as thoroughly dialecti-
cal refutations of any positive thesis, even the minimalist claims of Pali
realism. There is general consensus that the Yogacara school of Vasubandhu
and Asanga reject Nagarjuna’s views, whatever they are, in favor of an
idealism instructively similar to Euro-American schools.22 Unfortunately,
we do not have space to deal with Yogacara other than an indirect way.
As we have seen, Pali Buddhists do not deny the appearance of an empir-
ical self (jiva); rather, they deny that, corresponding to this appearance,
there is anything enduring, separate, or independent. These may just be
three different ways of saying the same thing, but since they represent three
different types of Buddhist arguments, they merit separate presentations.
First, there is no self that endures. What we see is constantly changing and
there is nothing that stays the same. The traditional argument here
proceeds by elimination: the physical bodies change; feelings, beliefs, de-
sires, and intentions all change; consciousness is intermittent; and our self-
conceptions change over time. None of the things we can point to as the
self remains the same. Therefore the self does not endure. The argument is
similar to the one given by Hume.
Second, the self is not separate from the causes and conditions that give
rise to it. A standard metaphor for this comes from the Dhammapada, a Pali
text from the third century B.C.E. The appearance of a rainbow arises out of
a certain combination of mist and light. Remove either one of these and
the rainbow no longer exists. Similarly, the appearance of a self arises out
of conditions: oxygen, food, parents, and so forth. Without them, there
would be no self. This argument can be made on a general level, as just
done, or on a particular level. You wouldn’t be the person you are if your
family, friends, and acquaintances all weren’t the people they are, if you
hadn’t had the experiences you’ve had, lived in the society you live in, and
so on. If we define the constellation of these conditions as one’s ‘world’, we
can say that the self cannot be separated, either practically or logically,
from the world in which it exists. Again we can say that causality has been
replaced by conditionality.
Third, the self is not independent, which is a rough approximation of
a Sanskrit term svabhava, meaning ‘self-nature’ or ‘own being’. Nagarjuna
uses reductio ad absurdum arguments to demonstrate the incoherence of
attempts to say anything about an independently existing self. If the self
existed on its own, it could not change or stay the same, could not be
unified or composite, could not know or be known, and so forth. And
yet, if we cannot say that the self really exists, by the same token we can-
not say that the self really does not exist. We cannot say anything at all
about it as an independently existing thing. What we are left with is an
empirical or a thoroughly conventional self, depending on the way we
read Nagarjuna.
The Pali versus Mahayana distinction is now not very helpful, so hence-
forth we will distinguish between ‘constructive postmodernism’ (CPM) and
a skeptical ‘deconstructive postmodernism’ (DPM). Many commentators
have interpreted Nagarjuna as anticipating the deconstruction of French
postmodernism.23 We prefer to interpret the Buddha’s philosophical in-
tentions as anticipating CPM, but we are not certain that Nagarjuna can be
interpreted under this rubric. We have already referred to more detailed

discussions of these two positions in the endnotes, but suffice it to say that
CPM is generally realist and supportive of the canons of logic and evidence,
while DPM rejects realism and any logocentric methodology. Nagarjuna is
a consummate logician and never rejects logic as a standard, so this obvi-
ously causes problems for any DPM reading of him. Nevertheless, in our
conclusions we will assume that Nagarjuna cannot be fully accommodated
within the CPM model.
Let us compare the translations of two quatrains on ‘agent/producer’ and
‘action/product’ to demonstrate not only the different interpretations but
also the difficulty in deciphering the poetry of Nagarjuna’s The Fundamen-
tal Wisdom of the Middle Way. (Hereafter the Karika, an abbreviation of the
Sanskrit title Mulamadhyama-karika.) Here are Jay Garfield’s and Frederick
Streng’s translations of the last two verses of chapter 8:
Action depends upon the agent. The producer proceeds being dependent
The agent itself depends on action. on the product,
One cannot see any way And the product proceeds being depen-
To establish them differently. dent on the producer.
The cause for realization is seen in noth-
ing else.
From this elimination [‘giving up’] of In the same way one should understand
[substantial] agent and action, the ‘acquiring’ on the basis of the ‘giving
One should elucidate appropriation up’ of the producer and the product.
[‘acquiring’] in the same way. By means of [this analysis] the product
Through action and agent All and
remaining things should be understood Producer all other things should be
(Garfield). dissolved (Streng).24
Even without knowing Tibetan (the original Sanskrit text has been lost),
we can see that Garfield’s translation is more elegant. Moreover, we can see
Streng’s skeptical conclusion that distinctions between agents, their actions,
and all other things “should be dissolved.” (The Yogacara idealist would say
“dissolved in the Buddha Mind.”) Garfield’s rendering is very different: he
takes Streng’s “giving up” only as denying agents and their actions any self-
being, not rejecting their existence altogether.
We propose that the phrase ‘phenomenal self’ be used for the Pali tradi-
tion and CPM interpretation while reserving ‘conventional’ as a place-
holder term for the deluded self who thinks it lives in a real world of
interdependent things and events. This conventional self exists only by
analogy to the appearance of the color spectrum as it is refracted through

a prism. If the prism represents ignorance, then upon enlightenment the
phenomenal world proves to be an illusion. If we complete the analogy by
speaking of the white light that is refracted, the Yogacara idealists would
call that the Buddha Mind while the skeptical Nagarjuna would deny its ex-
istence as well. For Mahayana Buddhists following the skeptical Nagarjuna
the conventional self has a practical purpose only. Clinging to a notion of
self that has self-being, its own perceptions and possessions is the origin of
suffering.
The use of conventions is governed not by their truth or reference to real-
ity, but by their effectiveness in diagnosing the human predicament and al-
lowing us to see the ‘emptiness’ (shunyata) of all things that we thought were
permanent and that we thought were somehow ours. From this standpoint,
appeals to determinism in one moment and to personal responsibility in the
next do not represent an inconsistency; rather, it is a recognition that differ-
ent forms of rhetoric will be effective in different situations. (By contrast the
Pali Buddhists have a realist concept of truth and meaning and that is why
they are able to embrace a compatibilist position on free will.) If the self is
thoroughly conventional, then to ask whether it is free or determined is like
asking ‘What is the sound of one hand clapping?’ Zen Buddhists were pro-
foundly influenced by what they took to be Nagarjuna’s skepticism.
Not all quatrains in the Karita, however, give us equally plausible read-
ings. Here are two in Garfield’s translation that summarize nicely the point
of the preceding paragraph:
If there were no self,
Where would the self’s (properties) be?
From the pacification of the self and what belongs to it,
One abstains from grasping onto ‘I’ and ‘mine’.
When views of ‘I’ and ‘mine’ are extinguished,

Whether with respect to the internal or external,
The appropriator ceases.
This having ceased, birth ceases (18:2,4).
This quatrain can be rendered intelligible from either a CPM or DPM posi-
tion, but other quatrains in this chapter appear to give the former approach
no support:
What language expresses is nonexistent.
The sphere of thought is nonexistent.
Unarisen and unceased, like Nirvana

Is the nature of things.
One can salvage the CPM position only by some clever qualifications, and
then only partially. When Nagarjuna speaks of denying existence, he de-
nies only absolute existence, the negation of which is absolute nonexis-
tence. (Nagarjuna agrees with Parmenides in rejecting this as inexpressible
and meaningless.) But from a CPM standpoint, the nonexistence in this
quatrain must be a relative nonbeing that corresponds to the relative being
of interdependent existence that is expressible and meaningful. (A thing’s
relative nonbeing is everything that it is not, which is actually all the other
things that it is dependent on.) The CPM interpretation of this quatrain
runs aground with the conclusion that Nirvana and all things are ‘unarisen
and unceased’. This quatrain appears to undermine the DPM interpreta-
tion as well, because it implies that Nirvana is the only ‘thing’ that has
self-being. Tibetan Buddhists influenced by Yogacara idealism hold that
Nagarjuna intended only to reject the self-being of phenomenal reality, not
the ultimate reality that is the Dharmakaya. This is the cosmic body of the
Buddha into which all selves that reach Nirvana dissolve. Here is one of
their own quatrains:
If emptiness were the method, then
Buddhahood could not be. Since other
Than this cause there would be no other fruit,
The method is not emptiness.25
Not many Mahayana Buddhists, except those in some Zen schools, were
willing to follow Nagarjuna in rejecting all metaphysical views of the
Buddha and his cosmic body. While Nagarjuna’s deconstruction is logically
rigorous, it was obviously not religiously satisfying for most Buddhists.
7 Nagarjuna on Causes and Conditions
In his interpretation of the Karita Garfield maintains that Nagarjuna

preserves a distinction between conditionality and causality. Garfield
maintains that Nagarjuna holds a positive view of conditionality as ex-
plained above, and he demonstrates that Nagarjuna rejects only the view of
causes as having some occult power to produce effects. For this argument
let us look at the first verse of the first chapter of the Karita:
Neither from itself nor from another,

Nor from both,
Nor without cause,
Does anything whatever, anywhere arise.
This is the famous Buddhist ‘neither-nor dialectic’ in action—also called

the ‘four-cornered negation’ in Indian logic.26 It is meant to exhaust any
possibility of ‘causality talk’, and each of the four options actually repre-
sent previous Indian philosophical schools.
The first chapter’s second verse contains a very rare phenomenon in the
Karita, a positive view of conditionality that Nagarjuna does not dialecti-
cally destroy:
There are four conditions: efficient condition;
Percept-object condition; immediate condition;
Dominant condition, just so.
There is no fifth condition.
The fourth verse makes it clear that causality understood as a metaphysical

power is not a part of conditionality:
Power to act does not have conditions.
There is no power to act without conditions.
There are no conditions without power to act.
Nor do any have the power to act.
We will now offer an example (inspired by Garfield) that demonstrates how

one can conceive of conditionality in its four instances without causality. A
person takes a match and strikes it, which is the efficient condition. The
match lights because there is enough oxygen, it is not too wet, and the tip
of the match has not worn to the point of nonignition, which are some of
the immediate conditions. The percept-object conditions are those within
any perceiver that allows the eyes to see. Finally, the person has struck the
match so that she can see whether what she took for a snake in the dark
was perhaps just a piece of rope. This is the dominant condition, the pur-
pose of any action; a Buddhist ‘final cause’ if you will. Please note that our
example has not appealed to any occult causes or powers as a way of
explaining what happens.
There are of course other translations and interpretations of these
quatrains, some of which support the skeptical DPM position. These read-
ings generally tend to conflate causes and conditions, and in alternative
translations of verses three and ten Nagarjuna appears to reject the exis-
tence of relational existence and the impossibility of anything whatsoever
arising. Here are the two quatrains from A. L. Herman’s revision of
Theodore Stcherbatsky’s translation:
In these four conditions we can find
No self-nature (svabhava)
Where there is no self-existence,
There can be no relational existence (parabhava).
If existing things have no self-nature

Then they have no real existence
The [formula] ‘this being, that appears’
Then loses every meaning. (1: 3,10)27
Note here the clear negation of the two principal features of the doctrine
of interdependent coorigination: the denial of relational existence and the
rejection of conditionality. This view of Nagarjuna’s argument anticipates
Bradley’s famous attempt to show that an ontology based exclusively on
internal relations is just as unintelligible as one based on external relations.
We summarize Bradley’s argument as follows: If A is internally related to B
and R is the relation, one would then need R to relate R to A, R that relates
to R to B, R that relates R to R ad infinitum. Bradley’s conclusion is that
either A and B are totally separate or they are identical (1930, 18). Therefore,
a putative realism of distinct but interrelated entities by necessity collapses
into either an absolute monism (the Yogacara position—they would have
loved Bradley) or a denial of any intelligible view of reality at all (the DPM
Nagarjuna). Appealing to Whitehead’s (1978) doctrine of asymmetrical rela-
tions (e.g., that the present is internal to the past but the past remains exter-
nal), one could assume that the Buddha would agree with him, because he did
not explicitly affirm a doctrine of full internal relations as some Mahayanist
Buddhists did. The presence of asymmetrical relations would save Buddhism
from the extremes related above and keep the CPM interpretation alive.
Garfield’s translation of the two passages above allows us to reject the
deconstruction of dependent beings:
The essence of entities is not present in the conditions. If there is no essence, there
can be no otherness-essence. If things did not exist without essence, the phrase
‘When this exists so this will be’, would not be acceptable. (1. 3,10)
We take this to mean that if any being lacks self-existence, then any being
other than it does also. It also follows that if some beings were substances,
we could not affirm the Buddhist law of conditionality. The CPM interpre-
tation of Nagarjuna appears to dissolve his doctrine of two truths, which
Garfield formulates as “the conventional truth of the reality and interde-
pendence of all phenomena and the ultimate truth of their emptiness of
inherent existence” (1995, 225). From the standpoint of CPM these two
truths are the same: real interdependence ontologically implies that no
being can have self-existence.
8 Conclusions
The traditional libertarian, who believes with Kant that, in the absence of
freedom, there cannot exist anything of absolute value in the world, may
wonder if there is any role left for freedom in Buddhism. Normally we
assume that there has to be a self or an agent in order for there to be free-
dom, but this is just the presumption the skeptical Nagarjuna questions.
If we cannot call the karmic web free since it lacks a self, by the same
token we cannot call it determined, since nothing outside of it is causing
it. To the extent that people identify a self, that self is determined by
causes outside of it. The more cultivated they become on the Buddhist
model, the less they think this way. The less who thinks this way? A ques-
tion that the European philosopher might ask. Nagarjuna’s answer is no
one, really. The nonpersonal web of causes and conditions sheds the delu-
sion, or, rather, ceases to give rise to it. Thus you get the seemingly para-
doxical lines from, for instance, The Diamond Sutra, that “even though
infinite beings have been saved, none have been saved.”28 Thus while we
would assume that there has to be a self in order for there to be freedom,
Nagarjuna would say that there is freedom only to the extent that there is
not a self. The familiar correlation in the European tradition of selfhood
and freedom is reversed: rather than correlates, selfhood and freedom are
antitheses.
The reason this matters for the skeptical Nagarjuna is the connection
between the self and suffering. The belief in a self that is enduring, sepa-
rate, and independent, gives rise to desires that are ultimately unsatisfiable
precisely because no such self exists. If all talk of the self or anything else is
conventional, and if no one set of conventions is more right than the
others, what then governs the choice among conventions? The First Noble
Truth of Buddhism, enunciated by the Buddha upon his enlightenment, is
that all egocentric existence suffers. This is not to say that every moment is
agony, which would obviously be false, but that the process of existing,
thinking one is a substantial self, is inherently painful. Even the joyful
moments take their character from the absence of, or absence of awareness
of, discomfort. Our unwillingness to acknowledge this fact and our stub-
born insistence that suffering is avoidable only compound the difficulties.
More particularly, suffering is caused by belief in a self that does not
exist. Thinking in terms of a self that is enduring, separate, and indepen-
dent causes people to search for a happiness they cannot find precisely
because such a self does not exist. The desire to alleviate suffering by
eliminating the belief in a self is the motive force of Buddhism. It is easy to
see how, in the Mahayana tradition, one would not be able simply to alle-
viate one’s own suffering without relying on the notion of self that caused
the suffering in the first place. As a result, motivation must come from the
desire to alleviate the suffering of all sentient beings. This realization led to
the Bodhisattva ideal, enlightened beings, in one sense greater than the
Buddha himself, who vow to delay Nirvana until all other beings have
entered before them.
The silence of all forms of Buddhism on the debate about free will first
stems from its refusal to divide up the world in the modernist ways
discussed above. Second, it is silent because it rejects a substantial self-
determining self in the Pali tradition and any notion of an individual self in
the Mahayana tradition, whether it be Yogacara idealism or Madhyamaka,
the school of philosophers influenced by Nagarjuna. If this is true, then
contemporary interest in the debate must be premised on certain assump-
tions about the self and personhood. We see this clearly in Harry Frankfurt’s
now classic essay where he claims that one of the desiderata of a theory of
free will is that it demarcate the ‘essential difference’ between people and
animals (1971, 6). Buddhists, however, see people on a spectrum of beings,
between animals, who are unable to diagnose the origin of their suffering or
to take steps to alleviate it, and putative deities who, because of their con-
stant state of bliss, lack the incentive to act. Thus, while Frankfurt and the
Buddhists both observe the same phenomenon of first- and second-order
desires, one reason the Buddhists do not assign it the same weight is because
they refuse to distinguish people from animals for moral purposes. Antici-
pating Tom Regan’s approach to animal rights, the Dhammapada assumes
that animals have interests and that “life [is] dear” to them (10:3).
The second requirement for a theory of free will, according to Frankfurt,

is that it should explain why freedom of the will is regarded as desirable.
For the Buddhists, the existence of orders of desire is an observed fact, of
utility in alleviating suffering but of no intrinsic value. Frankfurt’s pre-
sumption, right or wrong, that freedom is a valuable thing, is a European
notion that can be traced back, as we have seen, to Augustine and medieval
Christian discussions about sin, salvation, and divine power. The two
Buddhist traditions respond in significantly different ways: the Pali tradi-
tion affirms the preference for Nirvana over Samsara, while Nagarjuna
proclaims, in Nancy McCagney’s translation, “There is no distinction
whatever between Samsara and Nirvana; there is no distinction whatever
between Nirvana and Samsara” (1997, 209; Karita 25.19). Zen Buddhists
took this to mean that having a preference for Nirvana is a form of craving,
so Nirvana would come to only those without any desire at all, in a mo-
ment not of their choosing, right in the middle of the natural course of
events. Hence, the Zen master’s resort to the koan’s non sequitor, which is
designed to destroy the craving for salvation. Ideally, the monk has neither
preferences nor nonpreferences and cannot be said to be free in the way
that Lehrer has defined it.
We have argued that these different estimations of the significance of the
free will question are predicated on different ideas about the self. These
ideas about the self may be rooted in still deeper suppositions about moral
significance. In the religious traditions of Europe, the moral significance of
creatures is primarily predicated on a will that is truly their own: people are
thought to have earned salvation or damnation to the extent that they
have had a choice. In the Buddhist tradition, moral significance is predi-
cated on the ability to feel suffering: concern is due not just to rational
beings, but all sentient beings. Both of these could be contrasted, for in-
stance, to Native American religions, which accord moral significance to
mountains and rivers, which presumably can neither choose nor feel but
which still play a role in the cycle of life.
Our coverage of Buddhism has been very general and we have been
forced to be selective in what we have discussed. Our purpose here has not
been to argue the truth of compatiblism or incompatiblism, nor has it been
our intent to conclude that Buddhists or Euro-American philosophers are
right or wrong. Rather, our principal purpose has been to draw attention to
background assumptions from which the current debate has arisen and to
offer a different and arguably challenging perspective from Asia. What

we hope we have demonstrated is that presumptions about selfhood and
moral significance influence our intuitions and the questions we regard as
important to ask. And if indeed it should turn out that belief in a deluded
form of self is the origin of suffering, or that animals and even mountains
and rivers are worthy of moral respect, then these are not things we would
want to take for granted as we reflect, as did Frankfurt, on “our most
humane concern with ourselves and the source of what we regard as most
important and most problematical in our lives” (1971, 6).
Acknowledgments
We would like to thank Joseph Keim Campbell and Michael Myers for very
constructive comments and criticisms, and we are grateful to Michael
O’Rourke and P. J. Ivanhoe for their encouragement.
Notes
1. Cited in Stcherbatsky (1958, 133).
2. Cited in Keown (1992, 215).
3. One could say that the issue was first raised by the Stoics, who criticized Epicurus’s
attempt to introduce indeterminism among the atoms, but what is entirely missing
from Greco-Roman philosophy is the concept of a fully self-determining will.
Furthermore, Epicurus’s indeterminism is no more successful in addressing free will
than those who would now want to appeal to Heisenberg’s indeterminacy principle.
Finally, it is clear that modernist concepts such as autonomy and the separation of
fact and value were indeed anticipated and affirmed by some ancient schools of
thought, such as Sankhya-Yoga, Jainism, Indian and Greek atomism, and Chinese
Mohism.
4. For example, see Hudson (1973, 1985), Gudmunsen (1977), and Katz (1983).
5. See Gier (1991).
6. See Gier (2000, chap. 2) and the Series Introduction by David Ray Griffin in any
volume of SUNY Press’s Series on Constructive Postmodern Thought.
7. Mahavedallasutta, I.293.
8. Majjhima Nikaya II.214ff., cited in Gomez (1975, 84–85). We are indebted to

Gomez for both texts and insights.
9. Majjhima Nikaya I.190–191, cited in Kalupahana (1976, 64). The pronoun has
been changed to “they” to avoid the exclusive “he.”
10. See Galston (1991, passim, 230), and Galston (1995) for a revised view of the self.
See Macedo (1990, passim, 220–221), for “situated autonomy.”
11. See Kalupahana (1976, chap. 3).
12. Garfield (2002, 73–76). “All of the standard desiderata of good theories apply—
economy, elegance, predictive power, confirmation, coherence with other theories,
and so on . . . . Then we can pay attention to . . . interdependence, and its multiple,
multidimensional, inter- and intra-level character—and let a thousand entities
bloom, requiring of each only that it genuinely toil and spin, accomplishing some
real explanatory work . . .” (75).
13. Kalupahana (1976, 28). Reading Kapulahana more carefully than her teacher,
Gier’s student Kaylani Merrill shows that Buddhist conditionality has a different
logic. The Buddha used a Pali locative absolute to describe conditionality. The gram-
mar of ‘when that, then this’ is opposed to the ‘if, then’ formula found in European
forms of sentential logic. See Kalupahana (1992, 57).
14. Conze (1962, 149). The difference between abstract and concrete here can be
explained with the distinction between rational or aesthetic order. See Gier (2001).
15. Bhikkhu Buddhadasa recorded a long tradition of a metaphorical interpretation

of the realms of existence. See Buddhadasa (1972).
16. We are indebted to Joseph Goldstein for this statement and this insightful way
of redefining karma. See Goldstein (1999, 291). The definition of karma as volitional
action is not only good Pali Buddhism but it is also the position of the great
Mahayanist philosopher Vasubandhu: “karma is will (cetana) and voluntary action
(cetayita karanam)” (cited in Stcherbatsky 1974, 32).
17. For the best comparative studies of the Buddha and Hume, see Betty (1971) and
Giles (1993).
18. Cited in Conze (1962, 149).
19. Kalupahana (1987, 20–21). Kalupahana’s Pali has been changed to Sanskrit.
20. Aristotle, Nicomachean Ethics 1139b4–1139b5 (1975, Ostwald trans.).
21. See the “Nature” and “Spirit” sections of Nishida (1990).
22. See Garfield (2002) for the best comparative analysis of this issue.
23. See Mabbett (1995, 203). For a constructive postmodern interpretation see
Kalupahana (1992, 1986).
24. Garfield (1995); Streng (1967, 181–220).

25. Cited in Tsong-Karpa (1977, 117).
26. See Gier (1983).
27. See Herman (1983, 287–288).
28. Conze (1975, 25).
References
Aristotle. 1975. Nicomachean Ethics. Translated by M. Ostwald. Old Tappan, N.J.:

Macmillan.
Betty, L. S. 1971. “The Buddhist-Humean Parallels: Postmortem.” Philosophy East and

West 21: 237–254.
Bradley, F. H. 1930. Appearance and Reality. Oxford: Oxford University Press.
Buddhadasa, B. 1972. Toward Truth. D. K. Swearer, ed. Philadelphia, Pa.: Westminster

Press.
Conze, E. 1962. Buddhist Thought in India. London: George Allen and Unwin.
. trans. 1975. Buddhist Wisdom Books. London: George Allen and Unwin.
Frankfurt, H. G. 1971. “Freedom of the Will and the Concept of a Person.” Journal
of Philosophy 68: 5–20.
Galston, W. 1991. Liberal Purposes: Goods, Virtues, and Diversity in the Liberal State.
. 1995. “Liberal Virtues and the Formation of Civic Character.” In M. A.

Glendon and D. Blankenhorn, eds., The Seedbeds of Virtue: Sources of Competence,
Character, and Citizenship in American Society. Lanham, Md.: Madison Books.
Garfield, J. L. 1995. The Fundamental Wisdom of the Middle Way. New York: Oxford
University Press.
. 2002. Empty Words: Buddhist Philosophy and Cross-Cultural Interpretation. New

York: Oxford University Press.
Gier, N. F. 1983. “Dialectic: East and West.” Indian Philosophical Quarterly 10: 207–218.
. 1991. “Three Types of Divine Power.” Process Studies 20: 221–232.
. 2000. Spiritual Titanism: Indian, Chinese, and Western Perspectives. Albany,

N.Y.: SUNY Press.
. 2001. “Synthetic Reason, Aesthetic Order, and the Grammar of Virtue.”

Journal of the Indian Council of Philosophical Research 18: 13–28.
Giles, J. 1993. “The No-Self Theory: Hume, Buddhism, and Personal Identity.”
Philosophy East and West 43: 175–200.
Goldstein, J. 1999. “Cause and Effect.” In J. Smith, ed., Radiant Mind: Essential
Buddhist Teachings and Texts. New York: Riverhead Books.
Gomez, L. O. 1975. “Some Aspects of Free-Will in the Nikayas.” Philosophy East and
West 25: 81–90.
Gudmunsen, C. 1977. Wittgenstein and Buddhism. New York: Barnes and Noble.
Harvey, P. 1993. “The Mind–Body Relationship in Pali Buddhism: A Philosophical

Investigation.” Asian Philosophy 3: 29–41.
Herman, A. L. 1983. An Introduction to Buddhist Thought. Landham, Md.: University

Press of America.
Hudson, H. 1973. “Wittgenstein and Zen Buddhism.” Philosophy East and West
23: 471–482.
. 1985. “Wittgenstein and Nagarjuna.” Philosophy East and West 35: 157–170.
Hume, D. 1978. A Treatise of Human Nature. New York: Oxford University Press.
Katz, N. 1983. “Nagarjuna and Wittgenstein on Error.” In N. Katz, ed., Buddhist and
Western Philosophy. Atlantic Heights, N.J.: Humanities Press.
Kalupahana, D. J. 1976. Buddhist Philosophy: A Historical Analysis. Honolulu, Hawaii:

University of Hawaii Press.
. 1986. Nagarjuna: Philosophy of the Middle Way. Albany, N.Y.: SUNY Press.
. 1987. The Principles of Buddhist Psychology. Albany, N.Y.: SUNY Press.
. 1992. A History of Buddhist Philosophy: Continuities and Discontinuities.

Honolulu, Hawaii: University of Hawaii Press.
Keown, D. 1992. The Nature of Buddhist Ethics. New York: St. Martin’s Press.
Macedo, S. 1990. Liberal Virtues. Oxford: Oxford University Press.
Mabbett, I. W. 1995. “Nagarjuna and Deconstruction.” Philosophy East and West 45:
203–226.
McCagney, N. 1997. Nagarjuna and the Philosophy of Openness. Lanham, Md.:

Rowman and Littlefield.
Nishida, N. 1990. An Inquiry Into the Good. M. Abe and C. Ives, trans. New Haven,
Conn.: Yale University Press.
Robinson, R. H. and W. L. Johnson. 1992. The Buddhist Religion: A Historical Introduc-

tion. Belmont, Calif.: Wadsworth.
Schumann, H. W. 1973. Buddhism: An Outline of its Teachings and Schools. Georg

Feuerstein, trans. London: Rider.
Stcherbatsky, T. 1958. Buddhist Logic. The Hague: Mouton and Company.
. 1974. The Central Conception of Buddhism. New Delhi: Motilal Barnarsidass.
Streng, F. J. 1967. Emptiness: A Study in Religious Meaning. Nashville: Abingdon Press.
Tsong-Karpa. 1977. Tantra in Tibet. London: Allen and Unwin.
Verma, S. 1970. The Metaphysical Foundations of Gandhi’s Thought. New Delhi: Orient
Longmans.
Whitehead, A. N. 1978. Process and Reality: An Essay in Cosmology. New York: Free
Press.
14 After Compatibilism and Incompatibilism
Ted Honderich
1 Determinism and Whether It Is True
You can take determinism to be the family of views, a few of them clear and
otherwise conceptually adequate, that our decisions and the like and also
our actions flowing from them are the effects of certain ordinary causal
sequences. These personal events are necessitated by initial and also by sub-
sequent causal circumstances that make up the causal sequences. Has this
determinism been shown to be false?
Whether we actually experience its being false, which is to say become
aware of its falsehood in the course of our deciding and acting, has been dis-
puted. From the seventeenth century right up to an issue of the Journal of
Consciousness Studies the other year (Searle 2000), some philosophers have
said that we do indeed experience the falsehood of determinism. We learn
or somehow get the truth that at least sometimes our decisions and actions
are not ordinary effects.
That is, in deciding and acting we ordinarily have a certain idea of free-
dom, and we also somehow see that we do actually have this freedom—
that we are having it in the course of the deciding and acting. It is part of
our consciousness of pieces of deciding and acting that they could go the
other way instead, given things exactly as they are and were. Rather than
being ordinary effects, the decisions and actions are originated.
To say they are originated is to refer to some special mental or neural
generative activity that takes the explanatory place of ordinary and clear
causation. Often an attempt is made by philosophers to describe this
generative activity itself. It is described as being a matter of extraordinary
or funny causes, even self-causes, or creative endeavors, or teleology, or or-
dinary English verbs oddly understood, or something sui generis, or reasons
306 T. Honderich
in a particular sense of the term where they are not only terms of logical
relations, premises for conclusions, but events—although of course not
ordinary causal events. Events of some kind are needed, of course, for the
explanation of other events—the decisions and actions.1
All of these descriptions of origination are found unclear or factitious by
other philosophers, unsurprisingly. But that is not the end of origination.
To say decisions and actions are sometimes originated can be to say no
more than that our decisions and actions come about in such a way, what-
ever it is, that certain attitudes on our part are in order. These attitudes
include certain hopes for the future and the particular moral approval that
credits us with a particular kind of moral responsibility, and so on. To say
the lesser thing is to say something perfectly intelligible.2 Nor does it
become unintelligible if you go further and say that our decisions are
somehow within our control or up to us, whatever that comes to, and
hence that certain attitudes are in place.3
Against the idea that in deciding and acting we actually experience the
falsehood of determinism, other philosophers have said or contemplated4
that what we ordinarily experience in deciding and acting is something
quite different from an absence of effects. We have the idea that decisions
and the like are effects but not effects of a particular kind—and we can
know this idea fits the facts.
That is, decisions and acts are not effects of compelling, constraining, or
inhibiting causal circumstances—causal circumstances in conflict with our
desires, wills, personalities, or characters. As we discern in these typical
cases, there is nothing that is forcing our decisions on us. There is an ab-
sence of compulsion or the like. We know that in this clear sense the thing
can go the other way. There is the freedom of voluntariness, as distinct from
origination. Determinism is left perfectly possible by it.
Whatever is to be said of this dispute about deciding and acting with
respect to determinism, of which more in a minute, there is a truth about
the rest of our common experience, including our actual experience in
science—and in neuroscience most relevantly. It is that this wider experi-
ence leads us towards the view, a second determinism, that all experienced
events other than decisions and actions are effects of ordinary causal se-
quences. If this is so, we surely have good reason, a strong inferential base,
for taking the same to be true of the decisions and actions.
After Compatibilism and Incompatibilism 307
Certainly with nature as we encounter it in our lives, and also with

machines, and most importantly with our bodies, we do know of sequences
of causal circumstances for the events in them. We have at least some evi-
dence. At breakfast, no spoon levitates. At work, no keyboard or lever fails
inexplicably or truly randomly. No event in my central nervous system is a
real mystery, something of which there is no ordinary causal explanation
to be had. We know enough of ordinary events outside of our deciding and
acting to know—what indeed our very language of causation expresses—
that they are effects of ordinary causal sequences.
Do you say that our experience of other ordinary events than our deci-
sions and actions is one thing and our theory, particularly our quantum
theory, is another? You may indeed, but without great effect on some of
your listeners, particularly some who were on hand for Professor Earman’s
admirable tour of the subject at the 2001 Inland Northwest Philosophy
Conference.5 We are not all overwhelmed by science put to philosophical
purposes, or science when it becomes philosophy.
After nearly a century during which the indeterministic interpretation
of quantum theory might have been proved, corroborated, supported by
direct and univocal experimental evidence along these definite lines, it has
not been. This shouldn’t be overlooked with deference. Nor has anyone
dispatched the relevant proposition, annoying to practical-minded physi-
cists, that two theories can both work even if they are inconsistent and
hence one of them is false.
It has not even been made clear, of the items in quantum theory as in-
terpreted that are said not to be effects, that they are indubitably in the cat-
egory of things said by determinists of the several kinds to be effects. That
is, it has not been made clear these items are events, individuals in a stretch
of space and time as distinct from abstract entities. No determinist with his
head screwed on frontwards has ever said one end of an equation is the
effect of the other, or that a figure’s having angles adding up to 180 degrees
is the effect of its being a triangle.
It has become apparent that the indeterministic interpretation of quan-
tum theory has a nature akin to that of a philosophical theory, or indeed
just is a philosophical theory. I have in mind a theory that is at a certain dis-
tance from experience and experimental data, and also mathematics, and
that aspires instead to the essence of philosophy. That is a kind of logic,
308 T. Honderich
which is to say greater conceptual adequacy than that found in science—

higher standards of clarity, consistency, completeness, and so on. That a
theory may fail dismally in this aspiration, first of all on account of being
self-contradictory, as the indeterminist interpretation of quantum theory
has so often been and admitted to be, does not remove it from the class of
philosophical or would-be philosophical theories.
About the truth of a determinism of decisions and actions, it is worth
remarking, too, that the supposed microindeterminism based on quantum
theory would almost certainly not touch the matter if it really were a fact.
It would almost certainly be consistent with the determinism of decisions
and actions, a macrodeterminism related to the macrodeterminism of neu-
roscience. It would be as consistent as it would be with that determinism of
the ordinary world remarked on earlier—the absence of levitating spoons
and so on.6
For the purposes of this paper, a determinism of decisions and actions
will be taken as a reasonable assumption. This paper does not assume, with
the author of the piece in the earlier issue of the Journal of Consciousness
Studies (Searle 2000), that causation works upwards or simultaneously with
brain and mind in deciding and acting, but does not work sideways or
across time with respect to these activities. That is, this paper does not
assume that quantum theory as interpreted does not relate numerically dif-
ferent if simultaneous macroevents in choosing and acting, but does relate
earlier and later ones. This paper does not assume either, with respect to
sideways relations in general, that our brains and minds jump back and
forth between quantum theory as interpreted, and causation, depending
on whether we are deciding something or, on the other hand, just seeing
or thinking something. This stuff, to my mind, is a dog’s breakfast
(Honderich 2001a).
What most moves me to this un-American intemperateness, perhaps, is
that there can be no doubt whatever, as we shall soon be seeing, that certain
desires are in play with respect to determinism and freedom. They are the
stuff of attitudes of hope, moral approval, confidence in belief, personal
pride, and so on. Something like this is rightly allowed by such open liber-
tarians as Kane (1996). They are deep desires, sometimes with religion in
them. They are more in play than desires in some other parts of philosophy.
To speak generally, we want a certain freedom and the dignity it gives
us, and it seems we want them more as retirement approaches. When
suddenly it turns out that an interpretation of quantum theory, until now

supposed to be a perfectly general theory of reality, can be understood as
operating in so friendly and local and inconsistent a fashion as to satisfy a
philosopher’s deep desire, it is possible to wonder, to vary the metaphor,
which is the dog and which is the tail.
And let me remark on one other reason for my assumption of determin-
ism and my intemperateness. Some philosophers, as you have heard, look
into their experience of deciding and acting and see that it is uncompelled.
Whatever introspection comes to, that is something they can actually do,
as implied above. I am indeed aware that what is making me hand over
my wallet is the man with the gun. But can I look into my experience and
see that it is uncaused? Of course not. That fact, if it is one, is something
outside the experience, presumably prior to it. Does more need to be said?
Some other time.
2 Incompatibilism and Compatibilism, Their Falsehood, and Attitudinism
To turn now from the truth of determinism to its consequences or upshot

for us, we have one regiment of philosophers saying that in our conscious-
ness of deciding and acting we have our idea of freedom and it is inconsis-
tent with determinism. And we have the other regiment of philosophers
agreeing that in our deciding and acting we have our idea of freedom, any-
way typically, and it is consistent with determinism. The first regiment
assigns to us the idea of origination, along with voluntariness of course, and
the second regiment assigns to us only the idea of voluntariness. Each side
says we have only their preferred conception. Or, both regiments hedge
their bets, and say that their preferred conception is the only one somehow
important to us (Kane 2002).
These contentions can come to seem very remarkable. In moving and
after moving my finger to the left rather than the right, or in and after vot-
ing for the Left rather than the Right, I can have either idea of freedom with
respect to my experience. I can have it if I haven’t read or heard a word of
philosophy. I can have either idea this very moment about moving my
finger. So can you and anybody else.
What could the incompatibilist philosopher conceivably say to stop me
having the compatibilist idea of voluntariness? It’s true of my moving my
finger at the moment, certainly. What could the compatibilist philosopher
310 T. Honderich
say to stop me having the incompatibilist idea of orgination? It’s intelligible

if it doesn’t go too far, and it’s no contradiction, and it can come to me even
if I take determinism to be true. For these ideas it’s a free country, isn’t it?
Is the incompatibilist’s contention, or the compatibilist’s, actually the
thriller that we have his idea more often? Does the incompatibilist’s con-
tention reduce to the seeming philosophical scandal that his idea about us
accords to us something he wants as a person, some ascendancy or eleva-
tion above nature or the rest of nature? Is the compatibilist’s contention
really that the claim of our voluntariness is the only claim as to freedom
that he in his conventional clear-mindedness finds reassuring?
That we all have both ideas of freedom and, so to speak, make use of
them and are supported by them to about the same extent, has seemed to
me patently true. That is to say that both incompatibilism and compatibil-
ism are false. They are and always have been doctrines to the effect that we
all have a single, settled idea of freedom. They have always agreed in that.
Remember Hume, for a start, about all mankind agreeing about liberty
(1963 [1748], 95).
That they are false is better confirmed, indeed demonstrated, by reflec-
tion on other things than our experience of deciding and acting. There are
whole structures of our culture and social life that not merely give evidence
of but are informed by either one or the other of the two ideas of freedom
we all have. They provide a kind of large-scale behavioral proof that we
have the two ideas.
We have retributive punishment, based on an assumption of the origina-
tion of offences by offenders. Some of our societies kill offenders by way of
this idea. But we also have preventive punishment, which requires no such
assumption. What lies behind the two practices or impulses—and also
various practices or impulses of reward—are other things that are as confir-
matory in themselves of our all having both ideas of freedom.
These are two kinds of moral attitudes toward holding people responsi-
ble and crediting them with responsibility. In a sentence, I can disapprove
of you morally for not having done otherwise given the past and present as
they were, and I can disapprove of you, differently, for being the person
who willingly did that awful thing out of your very own character and
personality and nothing else.
It is as true, to revert to the external rather than the attitudinal world,
that the social and legal structures of individual rights ordinarily have to do
with ensuring only the voluntariness of actions by individuals. What the

framers of bills of rights have in mind, obviously, is to secure that we
are preserved from certain constraints and compulsions—that we have the
freedom of voluntariness.
On the other hand, while we have no significant structures for the
preservation of origination, since this is not within our power, we do in our
lives give about as important a place to origination. It is one large assump-
tion about ourselves that enables us to separate ourselves from the rest of
conscious life, what used to be called the animal kingdom. Our assumption
is a part of what explained, recently, the willingness of England to cure
cows of an ailment bearable to them and harmless to us by killing and
burning endless numbers of them. We do not allow that we can kill people
in order to keep up market prices.
It is indeed a second assumption of this rapid paper, then, that both
incompatibilism and compatibilism are false. The supposed problem of con-
sistency dividing the two regiments of philosophers collapses into nothing
as soon as it is seen there is no single concept of ours in this neighborhood
about which a question of its consistency with determinism arises, but
rather two concepts, one of which is patently inconsistent with determin-
ism and one of which is patently not. Incompatibilism and compatibilism
are answers to a question with a false presupposition, that we have but one
conception of freedom, or one important conception, and they themselves
assert or presuppose that falsehood.
Having been impressed by some of the previous pieces in this book, are
you still as unpersuaded as an editor of this volume is or was?7 Do you say
something along the following lines? “Compatibilism is the thesis that
determinism is consistent with the claim that persons often have moral
freedom or free will. Incompatibilism is the denial of compatibilism. How
can they both be false? Doesn’t this conflict with the law of the excluded
middle?”
The answer is that it doesn’t if there is no such thing as the claim that
persons often have moral freedom or free will—which there isn’t—and one
of two claims here is inconsistent with determinism and one isn’t.
Being as stubborn as philosophers ought to be, do you now say something
of the following sort? “Philosophers who work on moral freedom are con-
cerned to provide an account of what it is for somebody’s actions to be up to
him or something of the sort. Incompatibilists say one thing, compatibilists
312 T. Honderich
another, about our shared belief of what it is for something to be up to him.

It isn’t as if there is a prephilosophical or pretheoretical notion of origina-
tion and a different prephilosophical notion of voluntariness.”
Well, your third proposition there is the conclusion you’re supposed to
be arguing for. And your first two propositions also beg the question. What
you need is an argument for the proposition that we have only one idea, or
one important idea, of what it is for somebody’s action to be up to him.
And most certainly, by the way, you get no such thing in concentrating
on what is called the consequence argument of the incompatibilists—the
thought that something can’t be up to somebody if it is the consequence of
something in the remote past that wasn’t up to him. Or by concentrating
on the compatibilist thought that somebody can be morally responsible for
something even if they couldn’t do otherwise—what is called the argument
for alternative possibilities. But for my reasons for saying so, if you can’t
now guess them, you will have to look elsewhere (2002a, 115–121; 2002b,
469–473).
3 One Remaining Problem, and the Response of Affirmation
So exit a question about consistency with a false presupposition, the pre-

supposition that we have one conception of freedom. And exit incompati-
bilism and compatibilism. What is right is attitudinism—that we all have
two families of attitudes, one family containing one of the two conceptions
of freedom. In that case, what is our real philosophical situation?
Well, the two conceptions do indeed enter into and are indeed bound up
with attitudes and practices with fundamental places in our lives. So there
is a remaining problem of determinism and freedom. It has seemed to me
to be the problem that results from taking determinism to be true, or from
contemplating its truth, the problem of giving up or contemplating giving
up what is inconsistent with it, that the attitudes and practices bound up
with the idea of origination, at bottom certain desires.
The possibility can come to mind, incidentally, that this real problem has
in fact been the concern of at least some compatibilists who have pur-
ported to be proving to us, one more time, that voluntariness by itself is our
only idea of freedom or our only significant idea. One way in which you
can try to give up an idea or belief and what goes with it, is to try to per-
suade yourself that you don’t actually have the idea or belief. You are more
likely to fall into the strategy out of an excessively intellectual orientation

to our existence.
In any case, the real problem of determinism has seemed to me the given
kind of practical problem. It is the problem of how to do something, the
problem of how to accept the frustration of deep desires, those bound up
with origination. We want the content of certain hopes, as already re-
marked, which is to say futures in a sense open as distinct from being only
a matter of agreeable necessity. We want the reassurance of certain moral
attitudes. Also, we want a confidence that our inquiries of various kinds,
our pursuits of knowledge, are in a way unlimited as well as unfettered. So
it is with our nonmoral attitudes to others—personal attitudes. By atten-
tion to these latter attitudes, Peter Strawson (1962) moved the philosophy
of determinism and freedom towards an awareness of the various attitudes
that seem to be its actual subject matter.
The practical problem is not to be solved by a kind of collapse, a response
of dismay to determinism. This, in short, is a kind of concentration on the
particular kind of hopes, confidence in beliefs, personal feelings and so on
that are inconsistent with determinism. For a start, there is no rationality
in persisting in dismay when something else may be possible. Also this
response will not be a settled one. Nor is the practical problem solved
by another response to determinism, an intransigence. This is a kind of
concentration on the kind of hopes and so on that is consistent with
determinism. For a start this response is also unsettled, vulnerable to the
other one.
It has been my own proposal until recently that the best response to
determinism is something called affirmation (2002a, ch. 10; 1988, ch. 9).
This consists in perceiving and valuing the life consistent with determin-
ism, perceiving and valuing certain attitudes, relationships, and structures
of culture, and thereby giving up the life inconsistent with determinism.
The response is usefully parodied as a case of looking on the bright side,
and is indeed a part of a philosophy of life.
This affirmation recommends itself, but it can be supposed that particu-
lar means or strategies can also be of use in connection with it. One such
strategy is a kind of satisfying naturalism in other than the current philo-
sophical sense—the satisfying naturalism that can be called taking up mem-
bership in nature, maybe Nature. Another means, requiring no sensibility
or poetry, is contemplating escape from certain of the feelings inconsistent
314 T. Honderich
with determinism, notably the special guilt and failure associated with our
images of origination in connection with our own actions.
None of this is likely to be successful, essentially because of the strength of
our desires. It has seemed to me that success in the project of affirmation will
be owed, in the end, to nothing other than plain and settled belief in
determinism. That remains rare, a lot rarer than contemplating the truth of
determinism—taking it as something in need of being considered. It has
seemed to me that we or most of us, and perhaps more of our successors, will
succeed in the response of affirmation only when we really believe of the
things we dearly want that we do not have them and cannot ever have them.
4 Another Problem
Thomas Nagel in an account of determinism and freedom (1986) remarks

that he changes his mind about the subject every time he thinks about it,
and others may at least be tempted to carry on like this, as lately I have
been. A question can be raised in your mind, indeed something like a con-
viction, about such an attitudinal view as the one just outlined and the
project of affirmation—or for that matter about compatibilism. But my
concern is attitudinism and the project.
The principal point is that a persistent question can be raised in your
mind about attitudinism—mainly what is said of the role of the idea of
origination—and about affirmation. A question can be raised about them
by way of reflective attention to your own life. A question can be raised,
that is, by indulgence in autobiographical thinking or writing, at any rate
if it has had in it actions and so on such that you would think better of your
life and yourself now if they were missing (see 2001b, 395–399).
It can seem impossible not to feel responsible in a certain way for what
you have done. This is to disapprove morally of yourself in a certain way,
at bottom to have certain desires of several kinds. This disapproval is akin
to the disapproval of which we know, the kind that carries in it an idea
or image of the initiation of your actions that is inconsistent with
determinism—origination.
You are not saved from this self-doubt, self-accusation, or guilt by the
thought that you seem to be wanting your past to have been inconsistent
with determinism. Rather, you are stuck with the thought, aren’t you, that
there is some fact about your past that enters into the disapproval? It is
certainly relevant that the disapproval is unhappy and self-diminishing,

not something agreeable.
That is not the end of the story about autobiographical reflection, but
only half of it. Very likely your dealings with your past are not all of them
judgemental, not all of them concerned with moral or other disapproval or
approval. Often, if you get started on this reflection, what you want is just
to understand. The aim is explanation of what happened, not judgement on
it (see Richardson and Bishop 2002).
And, to come toward the point, the terrible fact is that you can deal with
your past life in this way and fall into no doubt whatever about determinism.
No doubt whatever that everything that happened did have an explanation
in the ordinary and indeed the only real sense. That stuff wasn’t random. It
was ordinary effects. Indeed you can increase your conviction of the truth of
determinism. That, at any rate, has been my experience. This is owed, pre-
sumably, to nothing arcane, but just to your coming to more knowledge
about a subject matter, maybe reading your diary and putting together facts.
So, here is a seeming contradiction. A kind of disapproval exists and
seems to be based on some fact. The natural candidate is origination—but
since determinism is the case, there is no origination (2002a, 2004a, ch. 12).
The seeming contradiction is something like the different and absolutely
intractable one announced by Kant, also as a result of something about
morality (1950 [1781]). Kant rightly did not respond in the compatibilist
way of course, by just giving up the proposition of indeterminism and
going on about freedom as being only voluntariness. Rather, he announced
that he would have both of the determinism and the indeterminism, by
putting them in different places. Determinism in or for the phenomenal
world, indeterminism in the noumenal world.
This higher compatibilism, entirely at odds with ordinary or mundane
compatibilism, seems hopeless. A distinction between two worlds is of
course possible, and has a number of philosophical versions, several of them
less metaphysical than Kant’s. But there seems no hope whatsoever of locat-
ing indeterminism and freedom significantly in only one of them, and cer-
tainly no hope for taking it out of the experienced world entirely. In any
case, since what is undetermined and free must in some sense turn up in
both worlds, it is impossible to see that the contradiction is actually escaped.
Is it conceivable that some philosophical idea as radical as Kant’s can
have a better hope of dealing with the seeming contradiction? In particular,
316 T. Honderich
to come to the crux, is it conceivable that we can by some idea or other

persist in certain attitudes—close to the attitudes tied up with origination—
persist in these attitudes without recourse to origination and consistently with
determinism?
5 Perceptual Consciousness, Causal Lines
Let me lay out a couple of lines of thought. Both of them do indicate more
of how radical it seems to me we need to be—it isn’t a matter of anything
like more tinkering with origination. Both of them just might be of use,
too, with the seeming contradiction. The first has to do with the nature of
consciousness, the second with causation and explanation.
What is it for you to be aware of this room now? More generally, on
the assumption that consciousness divides into perceptual, reflective, and
affective parts, what is it for you to be perceptually conscious? Two sorts of
general answer are given, and also given with respect to the other two parts
of consciousness.
One sort of answer, if you will allow a quick but enlightening parody, is
that perceptual consciousness is cells. It is neural activity. This is the old
physicalism about the mind of the seventeenth century, a physicalism
that now includes functionalism and cognitive science with philosophical
ambition. The other answer is only more disgraceful in terms of the physi-
calist conventionality of our current philosophy of mind. It is not often
given openly, but is implied by the increasing resistance to the idea that
consciousness is cells. This other answer, not much parodied, is that con-
sciousness is nonphysical stuff in heads.
It is possible to be attracted, as I am, to a quite general physicalism—the
view that all that exists is something close to physical. It is possible to be
attracted too, likely by way of the fact of the subjectivity of consciousness,
to the intolerable idea that perceptual consciousnessness is indeed funny
stuff in heads. You might think this situation is rather like what we have
been considering, attraction to both determinism and to something that so
far has been sunk in indeterminism. It seems to me possible that the situa-
tion with perceptual consciousness will be resolved by a radical view of this
consciousness.
What is it, really, for you now to be aware of this room? It is for the room
in a way to exist. That answer can be shown not to be merely a rhetorical
way of saying no more than that you are aware of the room. It is not a non-
analysis. Rather, the claim that your perceptual consciousness consists in a
kind of existence of a world consists in the claim that there is a certain state
of affairs, certainly not in your head. It is things, reasonably called chairs
and the like, being in space and time and dependent both on another
world, roughly speaking the relevant atoms, and also on your neurons in
particular. The world in question is anterior to the physical world, the one
dependent both on atoms and crucially on all of us, above all our shared
perceptual apparatus.
A few words delivered on the wing are unlikely to persuade you of this
doctrine, ‘consciousness as existence’ (2000, 2001c, 2004b). My first and
lesser aim, as you have heard, is to indicate something of the order of dif-
ferentness of thought that seems needed if we are to make a further and
better escape from three centuries of impasse in the philosophy of deter-
minism and freedom. In the doctrine about perceptual consciousness, a
general physicalism is in a way held onto, and the mystery of nonphysical
stuff in the head absolutely abandoned. But what attracted us to the stuff
in the head, its recommendation, is delivered to us by other means.
Above all, we are offered a real subjectivity, something clear on this
subject. This is your world of perceptual consciousness, different from the
shared physical world. Given the history and state of the philosophy of
determinism and freedom, is it not clear that only so significant a departure
from the cart tracks, and maybe one’s own recent tracks, has a chance?
Might it be that we have a chance of dealing with the seeming contradic-
tion involving determinism and certain attitudes by way of reflection on the
nature of consciousness—and in particular on consciousness as existence?
Searle thinks something like the first thing, and so, after all, if I may be per-
mitted another moment of un-American philosophical activity, there may
be one thing in his paper that is right. Surely no question of freedom could
arise about just exactly a physical world—a world in which we are present
only as conceived in the current physicalism of the philosophy of mind.
To say a word now of a particular idea about determinism and con-
sciousness, the autobiographical feelings of responsibility that persist de-
spite determinism are a matter of a certain individualism. If my feelings of
this kind cannot rest on my having originated my life, so to speak, can they
rest on the fact that my consciousness consists in what can be called the
private ownership of some reality?
318 T. Honderich
Consciousness as existence does not leave me, so to speak, outside the

world and merely its product. Rather, my being aware of things is my having
a standing that in a certain sense is creative, at any rate constructive. A world
of perceptual consciousness, a world anterior to the physical world and no
more mental, actually depends not only on atoms but on me. Do I thus have
a role or station that makes sense of feelings like those that have hitherto
been assigned to origination? (2002a, 147–151; cf. 2001b, 395–399).
To finish up here, let me add, for all of us and particularly for the one or
two open-minded graduate students of our age, another indication of the
extent to which we should think of abandoning the philosophy of deter-
minism and freedom as we have it and making a new start in this new
millenium. This has to do with causation and explanation.
A causal circumstance, as you know, is a set of events that necessitated an
effect. We typically isolate one of those events and say it caused the effect,
or indeed was the cause of the effect—as against another mere condition
of the effect, another event in the causal circumstance. This could be the
human action in the set, and will hardly ever be the presence of oxygen. In
general it is the event that most interests us or the event that it is in our
interests to isolate.
Suppose you now set about explaining something in a life, perhaps a pat-
tern of it or a culmination of it, and you take that pattern or culmination
to be the effect of a causal sequence, this being a past array of causal cir-
cumstances. You can now do the further thing of isolating a cause in each
of the causal circumstances or maybe just some of them. This gives you
what can be called a causal line within and from the beginning of the
sequence to the pattern or culmination. It may be that this is much of what
is had in mind by Alasdair MacIntyre (1981) and other philosophers who
speak of a narrative in connection with a life.
There is a problem about isolating a single condition in a causal circum-
stance and dignifying it as the cause. The problem, a paradox if you will, is
that in a clear sense this cause is no more explanatory of the effect than any
other condition in the causal circumstance. All are required or necessary
conditions. But the cause seems to be exactly that—more explanatory. That
is exactly what is conveyed by calling it the cause. Evidently there is the
very same problem about a causal line. In a clear sense it cannot be more
explanatory than any other chosen succession of items or states, say pres-
ences of oxygen. But it is more explanatory, isn’t it?
What this comes to is that the culmination of a life, say, is a matter of

plain determinism, but there seems also to be the possibility of some kind
of explanation of it that is different in kind. Some kind of departure from
determinism, or unexpected addition to it. At any rate there is a problem or
paradox here. The putative explanation would be consistent with deter-
minism, indeed within it, but different in kind. I wonder if the thing
is worth reflection in connection with determinism and the attitudes in
which we can find ourselves persisting, but I mainly offer it here as another
indication of the extent to which we should start out anew with determin-
ism and freedom (2002a, 151–153; 2001b, 399–415).
It may be that we shall get nowhere. If so, I myself shall go beyond com-
patibilism and incompatibilism only to the attitudinism and the project of
affirmation. Sticking to this will have to involve something like seeing that
what presents itself as a certain moral attitude to oneself and one’s past is
in fact moralism, indeed a kind of moralized self-abuse. But this affirmation
will not be perfect contentment.
Notes
1. There would be no hope, of course, in replacing ordinary causation with reasons

conceived as just premises in deductive, inductive, or otherwise normative relations.
Reasons in this sense are indeed just propositions standing in relations with other
propositions. That there is a good reason in this sense for my doing something, even
a reason that overwhelms all others, or even a reason that conflicts with nothing, is
perfectly consistent with my not doing it at all. Reasons of this ordinary kind explain
no actions and cannot conceivably replace ordinary causes in their explanatory role.
To speak only of such reasons is to give no explanation of action in the relevant
sense. As for reasons in what is perhaps the most ordinary sense, where they are ac-
tual causes taken by their owner as rational or the like, sometimes called psycholog-
ical causes, these deterministic items are of course ruled out for the philosopher of
origination from the start.
2. Peter Strawson (1962) spoke for many philosophers before and after him when he
spoke of “the obscure and panicky metaphysics” of origination. Galen Strawson
(1986) takes the view, in brief, that we do not need to think about determinism since
there is nothing intelligible in conflict with it in the talk of origination. Father right,
son not so right.
3. There is more on origination as I see it, and on other matters considered in this
paper, from causation onward, in Honderich (2002a) and (2002b). There is a lot more
on almost all relevant matters, for resolute readers, in (1988).
320 T. Honderich
4. Honderich (2001a). My own contemplation of the proposition about what we

ordinarily experience, however, certainly does not issue in support for compatibil-
ism, the doctrine usually associated with it and considered below.
5. For the final version of Prof. Earman’s lecture, see the first paper in this book, “De-
terminism: What We Have Learned and What We Still Don’t Know”. Cf. the various
contributions in Atmanspacher & Bishop (2002). You need not rush to agree, how-
ever, with what may be assumed, that the only conceptually adequate conception of
determinism is in terms of physics.
6. These various skeptical thoughts about the indeterminist interpretation of quan-

tum theory and its use in defense of freedom of origination are set out more fully,
along with other thoughts of the same tendency, in my (2002a, 71–80) and (1988,
304–334).
7. Joseph Keim Campbell.
References
Atmanspacher, H. and R. Bishop. 2002. Between Choice and Chance. Thorverton and
Charlottesville: Imprint Academic.
Honderich. T. 1988. A Theory of Determinism: The Mind, Neuroscience, and Life-Hopes.

Oxford: Clarendon Press. Reprinted in 1990 as the two paperbacks Mind and Brain
and The Consequences of Determinism.
. 2000. “Consciousness as Existence Again.” In B. Elevitch, ed., Philosophy of

Mind: The Proceedings of the Twentieth World Congress of Philosophy, volume 9. Philos-
ophy Documentation Center. Bowling Green: Bowling Green State University. Also
in Theoria 95.
. 2001a. “Mind the Guff: A Response to John Searle.” Journal of Consciousness

Studies 8: 62–78.
. 2001b. Philosopher: A Kind of Life. London and New York: Routledge.
. 2001c. “Consciousness as Existence and the End of Intentionality.” In A.

O’Hear, ed., Philosophy at the New Millenium, Royal Institute of Philosophy Lectures
for 2001–2002. Cambridge: Cambridge University Press.
. 2002. How Free Are You? Second edition. Oxford and New York: Oxford
University Press.
. 2002a. “Determinism as True, Compatibilism and Incompatibilism as False,

and the Real Problem.” In Kane (2002).
. 2004a. On Determinism and Freedom (collected papers). Edinburgh: Edinburgh

University Press.
. 2004b. On Consciousness (collected papers). Edinburgh: Edinburgh University

Press.
Hume, D. 1963/1748. An Enquiry Concerning Human Understanding. L. A. Selby-Bigge,

ed. Oxford: Clarendon.
Kane, R. 1996. The Significance of Free Will. New York: Oxford University Press.
, ed. 2002. The Oxford Handbook of Free Will. Oxford and New York: Oxford
University Press.
Kant, I. 1950/1781. Critique of Pure Reason. N. Kemp Smith, trans. London:

Macmillan.
MacIntyre, A. 1981. After Virtue. London: Duckworth.
Nagel, T. 1986. The View From Nowhere. New York: Oxford University Press.
Richardson, F. and R. Bishop. 2002. “Rethinking Determinism in Social Science.” In

Atmanspacher and Bishop (2002).
Searle, J. R. 2000. “Consciousness, Free Action, and the Brain.” Journal of Conscious-
ness Studies 7: 3–22 (“Mind the Gap” issue).
Strawson, G. 1986. Freedom and Belief. Oxford, Clarendon Press.
Strawson, P. F. 1962. “Freedom and Resentment.” Proceedings of the British Academy

48: 1–25. Reprinted in Studies in the Philosophy of Thought and Action. Oxford and
New York: Oxford University Press, 1968.
Index
Ability, 216–218, 231, 233, 237, Agency/agent, 61–62, 76–77, 87, 174,
241–245, 252 205, 283–285, 288, 292
basic, 242 Alienation, 178, 180
strong account of, 231–232, 237, Alternative possibilities (also,
241, 250 alternatives), 3–6, 111–112, 114,
weak account of, 231–233, 237, 126–129, 143, 152, 189, 217
241–243, 245, 248–252 argument for, 312
Accountability, 114–115, 117–118, condition of (AP), 70–72, 74
123–124 principle of (PAP), 4, 55, 126–127,
Action, 7, ch. 4 passim, 108–111, 113, 219–222
116–117, 119–124, 126–127, 130, Annihilationism, 289
135–137, 174–175, 181, 206, 217, Approval, moral, 306, 308, 310,
220, 234, 241, 246, 277, 287, 292, 314–315
305–306, 308–310 Atman, 88–289
akratic, 174–175, 179 Attitudes
autonomous, 174–175, 181 personal, 313
basic, 90–91, 244 reactive, 152
consequences of, 217, 220–221 Attitudinism, 312, 314, 319
courses of, 106 Austin-style examples, 71–72, 79
free, 217–218, 228, 257, 273, 277, 285 Autobiographical reflection, 314–315
future, 128 Autonomy, 285, ch. 8 passim
intentional, 243 as agency, 174–175
nonakratic, 174 agent, ch. 8 passim
nonautonomous, 175–176 as authenticity, 176–178, 186
repertoire of, 242, 244 heroic, 179, 184
self-forming, 73–74, 77–82, 85–87 normative moral, 176, 183
theory of, 174–175 as psychological independence,
volitional, 287 175–176, 179, 183–184
weak theory of, 250 as reasons-responsiveness, 180–181,
Affirmation, 313–314, 319 184, 186
324 Index
Autonomy (cont.) Closure principle, 256

as self-control, 173–176, 179–180, 184 Cognition, 269
as self-efficacy, 175 Coinstantiation, 240
Compatibilism, 4, 7, 53, 55–56, 60,
Belief, 120–123, 127, 130, 174, 63–64, 70, 113, 128–129, 140, 152,
242–245, 250–251 190, 194–195, 197, 199, 203,
manifesting, 110 215–216, 222–223, 225, 228, ch. 11
Beta blocker, 223 passim, 271–275, 278, 282, 287, 293,
-rule, 5–6, 198, 223–224, 229, 250. See 299, 305, 309–312, 315, 319–320
also Unavoidability operator conditional-analysis argument for, 222
Blame, 274 existentialist, 239–240
future, 136, 138, 140–141, 144, 146 of free will and indeterminism, 222
past, 136, 138, 140, 144–145 logic of, 53
Blameworthiness, ch. 6 passim, 152, question of, 70–76
162, 164, 166, 168, 173, 178, semi-, 6, 190
186–187 Conditionality (i.e., dependent
Body, 279, 289, 294 coarising), 285–288, 294–297, 301
Brain, 308 Consciousness, 286, 288–289, 291,
in a vat, 255–256, 260 316–317
Braino, 50 as existence, doctrine of, 317–338
Buddha mind, 293 perceptual, 316–318
Buddhism Consequence argument, 5–6, 74,
Mahayanist, 277–278, 290–294, 299 231, 271
Pali, 277–279, 281–291, 299 Constraint, 2
Zen, 293–294, 299 Context sensitivity, 258–262, 264–267,
270–271
‘Can’, 47, 64, 66, 71, 74 Contextualism
‘Causally determined’, 264 epistemological, 255–257, 259–263,
Cause/causation, 47, 56–59, 61–63, 266–275
68, 76–77, 82, 108, 127, 130, 156, free will, 255, 258–259, 262–266,
174, 231, 240, 251, 278, 280–281, 270–275
283–288, 294–295, 305–307, Control, 78, 82, 85
316, 318 failure of, 181
agent, 114, 227–229 guidance, 151–153, 162, 164, 207
efficacy, 109 plural voluntary, 81
event, 227 regulative, 151
line, 318 self, 174–177, 179, 181, 184,
self-, 305 186–187
Cetana, 289–290 ‘Could have done otherwise’, 4–5, 262
Chaos, 27–28, 78–79 Counterfactuals, 251
Choice, 70, 73, 81, 84 Cravings, 283
free, 217–219, 221 Crowding out, 200
having a, 223 Cycles of existence, 278
Index 325
Decision, 90, 108–109, 111–112, 125, Establish whether P, 235–236

127, 135, 139, 305–306, 308–310 Eternalism, 289
autonomous, 176 Events, 234–236, 240, 307
Deliberation, 181, 201, 217–218 concrete, 220, 222
mechanism of, ch. 7 passim mental, 174
purpose of, 118–119, 121–122 Executions, 244
rational, ch. 5 passim Existence question, 70, 75
Desert, 135 Explanation, 316, 318–319
Desert base, 135 action, 110, 123
Desire, 48, 131, 174, 242–245, causal, 263–265, 270–273
250–251
external, 178 Facts, 220, 281–282, 284, 286
first-order, 298 Fallibilism, 273–274
second-order, 174, 298 Field theory, 27–28
Determinism, 7, ch. 1 passim, 63–65, relativistic quantum, 29–34, 41
76, 109–112, 127–128, 213–215, 217, First Noble Truth of Buddhism, 297–298
222–225, 231, 233, 237, 239, 243, Forcing, 240
248–250, 252, 255, 257, 262, 264, Four-cornered negation, 295
270–275, 277–278, 283, 285, 293, Frankfurt examples, 5, 55, 136, 138,
ch. 14 passim 139–140, 142, 147, 151–152,
causal, 2, 3, 140–143, 189, 193–198, 155–157, 163–164, 167–169
200–206, 208 actual sequence, 152, 156–159, 161
fragility of, 22–23 alternative sequence, 155–164
hard, 4, 7, 58 ‘Free’, 114–115, 262–266, 270–275
Laplacian, ch. 1 passim ‘Freedom’, 113–114
logical, 2 Freedom, 51, 64, 67–68, 105–106,
ontological, 21, 27 113–116, 123, 125, 151, 231, 251,
soft, 4, 7, 287 255, 257, 262, 264–265, 270–274,
temporal, 2 277–280, 283, 285, 290, 293, 297,
theological, 2 308–315, 317–319
Dharma, 284 absolute spiritual, 278–279
Dialectical stalemate, 196–199, of action, 53, 55, 60
202–204, 208 concept of, 113–115, 123, 128
Distinction thesis, 154–155, 157 conception-neutral reading of,
Divergent miracles defense, 251 115–116, 118, 123
Domain of dependence, 28–29 conceptions of, 113–114
Doubt, Cartesian, 281 metaphysical, 2, 271
of the mind, 283
Emotions, 279, 282 minimal notion of, 47, 49
Empiricism, 281–282 moral, 1–7
Emptiness, 293 political, 2, 271
Endorsement, 181 religious, 2
Epistemology, 255–256, 260, 263, 265 sense of, ch. 5 passim
326 Index
Free will, 3–4, ch. 3 passim, 105, 113, Interdependent coorigination

115, 128, 175, ch. 10 passim, 243, doctrine of, 286, 288, 296
277–281, 285, 298–299, 311 principle of, 283, 287
as a mystery, 224–227 Intuitions, 166
plurality conditions for, 72, 74 Involuntariness, 283
F-type case, 189–191 Irrationality, 130–131, 169, 175
Functionalism, 278, 288–290, 316
Jainism, 279, 283–284,
Goals, 131 288, 300
Gravitation, Newtonian, 25 Judgment, 86–87
quantum, 41–43 moral, 205
valuational, 114
Hinduism, 279, 288–289 Justification, 117–118, 123, 261, 269,
272–273
Idealism, 277
Yogacara, 278, 290, 293–296, 298 Karma, 283, 287, 290
Imperatives, 130 Karmic bondage, 279, 285
Incompatibilism, 4, 6–7, 58, 63, 65, ‘Know’, 259–262, 265–268, 275
67–68, 70, 72, 113, 128, ch. 10 Knowledge, 114, 255, 257–262,
passim, 231, 233, 239, 248, 250, 265–270, 281
270–272, 274, 299, 305,
309–312, 319 Laws
of causal determinism and alternative causal, 257
possibilities, 189–191 of nature, 65–68, 109, 213–214, 231,
of causal determinism and moral 233, 236–240, 246–252
responsibility, 189, 193–205, 208 physical, 216
of free will and indeterminism, psychological, 216
225–226 strong accounts of, 231–232, 237–240,
Mind argument for, 222 243, 245, 249–250, 252
Indeterminism, 28, 61, 63, 71, 74, weak accounts of, 231–233, 237,
77–86, 106, 109, 112–114, 126–127, 239–241, 252
139, 141–143, 197, 200, 217, 222, Liar, Paradox of, 225–226
224–225, 272, 285, 300, 307–308, Libertarianism, 4, 6–7, 61, 76, 86,
315–316, 320 297, 308
hard, 6 agent-causal theories of, 77
Individualism, 317 paradox of, 62–63
Inevitability, 108–109 teleological intelligibility
Inference to the best explanation, theories of, 77
269, 273 Logic, two-valued, 234–235
Intelligibility question, 70, 75–77 Luck/chance, 79, 82–83, 228
Intention, ch. 4 passim, 111,
130–131 Madhyamaka, 298
Interdependence, 288 Making happen, 240
Index 327
Manipulation principle, 155–156, Passivity, 178

160–164, 167, 169–170 Perception, 243–244, 269, 284
Measurement problem in QM, Personal identity, 177,
31–33 184–185, 279
Meditation, 283 Phronesis, 284
Memory, 269 Physicalism, 316–317
Metamind, 48, 60 antecedent, 208
Middle Way of Buddha, 289 Physics, general relativistic, 34–44
Mindfulness, 282 Newtonian, 23–28, 36
Modernism, 277, 280–281, 298 special relativistic, 28–29
‘Moral responsibility’, 262 Plenitude, principle of, 75
Motion, 21–27, 32 Possible worlds, 215, 240
Motivating complex of cognitions, Postmodernism
243–244 constructive, 278, 289, 291–294,
Multiplicity of conditions, doctrine 296–297, 301
of, 287 deconstructive, 278, 291–295
Power, 71, 74
Narrative, 318 active, 2
Naturalism, 313 to do otherwise, 219
ethical, 284 Pragmatism
Neural processes, 78, 80–81, 83, 87, American, 279
263, 316 contextual, 284
Neuroscience, 306, 308 Praiseworthiness, 152, 162, 164, 166,
Nihilism, 1, 6–7, 130–131, 288 168, 173, 178, 186–187
Nirvana, 279, 281, 283, 294, Predictability, 21
298–299 Preference, ch. 2 passim, 243,
No-responsibility path, 199–200, 282–283
202, 206 first-level, 50–51
Normativity, 114 freedom of, 52–56, 60–61
No-self, doctrine of, 288 power, 51–54, 56–58, 60,
62–63, 65
Objectivism, 281 second-level, 50–51
Obligation, moral, 124, 144–145 Preference structure, 51–57, 60,
Observables, 36 62, 65
Omissions, intentional, 135 Prepunishment, 137–138, 145
Omnipotence, 280 Primacy condition, 52
Openness, libertarian (L-openness), Prohairesis, 289
138–141 Propositions, 234–236, 239–240,
of future, 138 250, 253
Origination, 6, 305–306, 309–316, Psychology, moral, 2, 174
318–319 Punishment, 128, 274, 310
Overdetermination, 108, 191, Purposes, 70
193–195, 200 Pythagoras’s theorem, 240
328 Index
Quantum gravity, 41–43 Samsara, 283, 299

Quantum mechanics, nonrelativistic, Sankhya-Yoga, 279, 288, 300
29–34, 41 Sanskara, 289–290
Quantum theory, 307–309, 320 Science, “no laws” view of, 21
Self, 131, 174, 278, 288–293, 297–300
Rationality, 61, 72, 78, 86, 180–181, 273 Cartesian, 288
Rationalization, 110–112, 127, 243–244 empirical (i.e., life principle, jiva),
Realism, 277–278, 290, 292–293, 296 289–290
Reasoning false, 177
practical, 129, 155, 199–201, 204, 206 functional, 288
theoretical, 131 governed, 174
Reasons, 47, 56–58, 62, 112, 117–119, governing, 174
122–124, ch. 7 passim, 180–181, 184, phenomenal, 292–294
217, 243, 245, 269–270, 319 real, 177–179
moral, 144, 180 -reflection, 122–123
Reasons-responsiveness, 151–152, 155, spectator, 288
166, 169, 205, 207 ‘Sense’, 115–116
moderate, 151, 153–154, 158–162, Senses, 279
164–171 Sensibility, moral, 186
Reference class, 258 Sentiment, moral, 144
Relations Significance question, 70
asymmetrical, doctrine of, 296 Singularity, spacetime, 37–40, 44
causal, 288 Skepticism, 105, 125, 131, 255–260,
full internal, doctrine of, 296 262, 264, 266–270, 272, 274, 277,
Render true/false, 216, 234–236, 281, 293, 297
246–249 Sovereignty, 279
Responsibility, 72, 78–79, 86–87, Space, container view of, 23–24
114–115, 122 Spacetime, anti–de Sitter, 35
creative, 114 classical, 23, 25
moral, 1–4, 114, 122, 124, 128, Gödel, 35
135–136, 139, 143–147, ch. 7 passim, in GTR, 34–41
173, 176, 182, 184–187, ch. 9 passim, Minkowski, 28–29, 37, 40
217–222, 225–226, 231, 257, 273, neo-Newtonian, 25
278, 281, 287–288, 293, 306, 310, 312 Spiritual Titanism, 279
moral self-disclosure view, 145–146 State of affairs, 220–222
prospective, 135–137 Structural connections, 240
retrospective, 135–136 Subjectivism, Cartesian, 281
Rights, 311 Substance, 61, 227, 285
R principle, ch. 5 passim
belief-concept reading of, 106, Teleology, 305
113–124 Temptation, 78
commitment-concept reading of, 123 Testimony, 269
indeterministic reading of, 106–113, Transfer of blamelessness, principle of,
116, 124, 126–128 189–190
Index 329
Transfer of nonresponsibility, principle

of, ch. 9 passim
NR formulation, 190–193, 196–199,
201–205, 207
NR formulation, 191–192, 196
NR formulation, 193–199,
202–203, 208
NRC formulation, 201–203
one-path case of, 192–198, 200–206
two-path case of, 193, 199–200,
204–206
Transfer of powerlessness,
principle of, 190
Truth, 234–236, 253
‘Try’, 91
Trying to act, ch. 4 passim
Two truths, doctrine of, 297
Ultimate responsibility condition, 72–75

Ultrachoice, 58
Ultrapreference, 56–61
Unavoidability operator, 5–6, 198,
223–224, 229, 250. See also -rule
Unwilling addict, 177
Upanishads, 279, 288
Vagueness, mystery of, 225–226

Value, 177, 281–282, 284, 286
Value experiment, 86
Virtue, 282
Volitional necessity, 176–177
Volitions, 62, 89, 244
Voluntariness, 72, 78, 82, 86, 306,
309–312, 315
Wanton, 184
Will, 70–71, 84–85, 90, 277, 283, 290
weakness of, 169, 177
Will-setting, 72, 74
World
phenomenal, 315
state of, 109
Zeno, 22–23, 26

Freedom and Determinism

Uploaded by

Copyright:

Available Formats

Freedom and Determinism

Uploaded by

Document Information

Original Description:

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Freedom and Determinism

Uploaded by

Copyright:

Available Formats

Freedom and Determinism

Topics in Contemporary Philosophy

Editorial Board Members

edited by Joseph Keim Campbell, Michael O’Rourke, and David Shier

Library of Congress Cataloging-in-Publication Data

Freedom and Determinism: A Framework 1

Determinism, Freedom, and Agency 19

1 Determinism: What We Have Learned and What We Still

2 Freedom and the Power of Preference 47

3 Agency, Responsibility, and Indeterminism: Reﬂections on Libertarian

The Metaphysics of Moral Responsibility 103

5 The Sense of Freedom 105

7 Moderate Reasons-responsiveness, Moral Responsibility, and

8 Which Autonomy? 173

The Compatibility Problem 211

10 Van Inwagen on Free Will 213

13 Buddhism and the Freedom of the Will: Pali and Mahayanist

14 After Compatibilism and Incompatibilism 305

Joseph Keim Campbell, Michael O’Rourke, and David Shier

1 The Traditional Problem of Freedom and Determinism

Thoughts about freedom and determinism have engaged philosophers

thrown by an adult with normal cognitive capacities. In trying to uncover

Note that there are no temporal restrictions on p and q. Still if p is a propo-

depending on how they answer the remaining questions above. Compati-

Van Inwagen (1983) offers three formal versions of the Consequence

(␣) From p deduce Np

2 New Work on Freedom and Determinism

Determinism, Freedom, and Agency

the time evolution of the system is indeterministic. Earman’s review covers

The Metaphysics of Moral Responsibility

(R): Rational deliberators, in virtue of their nature as rational

In “Libertarian Openness, Blameworthiness, and Time,” Ishtiyaque Haji

The Compatibility Problem

the argument by providing a counterexample to one of its key premises,

This latter principle plays an important role in arguments in support of

The volume ends with an original and controversial contribution by Ted

2. For discussions of the problems of freedom and logical/theological determinism

(1) ((P0 & L) → P) assumption of determinism

9. For contemporary discussions of libertarianism, see O’Connor (1995) and Kane

14. This is based on an example from Ravizza (1994, 78).

Aristotle. 1963. Categories and De Interpretatione. J. L. Ackrill, trans. Oxford:

Fischer, J. M. 1982. “Responsibility and Control.” Journal of Philosophy 89: 24–40.

. ed. 1986. Moral Responsibility. Ithaca, N.Y.: Cornell University Press.

. 1987. “Responsiveness and Moral Responsibility.” In F. Schoeman, ed.,

. 1994. The Metaphysics of Free Will. Oxford: Blackwell.

. 1998. Responsibility and Control: A Theory of Moral Responsibility. Cambridge:

Frankfurt, H. G. 1969. “Alternate Possibilities and Moral Responsibility.” Journal of

. 1990. On Action. Cambridge: Cambridge University Press.

Hawthorne, J. 2001. “Freedom in Context.” Philosophical Studies 104: 63–79.

Lehrer, K. 1976. “‘Can’ in Theory and Practice: A Possible Worlds Analysis.” In M.

. 1980. “Preferences, Conditionals and Freedom.” In P. van Inwagen, ed., Time

. 1990. Metamind. Oxford: Clarendon Press.

Moore, G. E. 1912. Ethics. New York: Oxford University Press.

Pereboom, D. 1997. Free Will. Indianapolis, Ind.: Hackett.

. 2001. Living without Free Will. Cambridge: Cambridge University Press.

Ravizza, M. 1994. “Semi-Compatibilism and the Transfer of Nonresponsibility.”

Strawson, G. 1986. Freedom and Belief. Oxford: Clarendon Press.

Strawson, P. F. 1962. “Freedom and Resentment.” Proceedings of the British Academy

Taylor, R. 1963. Metaphysics. Englewood Cliffs, N.J.: Prentice-Hall.