Graham Priest Thinking The Impossible PDF
Graham Priest Thinking The Impossible PDF
Graham Priest Thinking The Impossible PDF
Graham Priest
City University of New York
University of Melbourne
Abstract
The article looks at the structure of impossible worlds, and their deployment in
the analysis of some intentional notions. In particular, it is argued that one can, in
fact, conceive anything, whether or not it is impossible. Thus a semantics of con-
ceivability requires impossible worlds.
2. Possible Worlds
2.1 Their Structure
Possible-world semantics are familiar to contemporary logicians and philoso-
phers, and need little explanation;6 but let me set things up in a slightly unusual
way, for reasons that will become clear later. Take a propositional language
with the connectives: ∧, ∨, ¬, ◊, .7 (⊃ can be defined in the usual way.) An
interpretation for the language has four components: ⟨X, R, @, ν⟩. X is a set of
(possible) worlds. (It would be more normal to write this as W; but I will hold
this letter in reserve till later.) R is a binary relation on X: relative possibility. @
Î X is the actual world. And ν assigns every propositional parameter a pair of
subsets of X, ν+(p) and ν−(p), subject to the constraints of exclusivity and ex-
haustivity:
Exc: ν+(p) Ç ν−(p) = Æ
Exh: ν+(p) È ν−(p) = X
Intuitively, ν+(p) is the set of worlds where p is true; ν−(p) is the set of worlds
where p is false.
We now define what it is for a formula to be true (⊩+) and false (⊩−) at a
world w Î X:
• w ⊩+ p iff w Î ν+(p)
• w ⊩− p iff w Î ν−(p)
• w ⊩+ ¬A iff w⊩− A
• w ⊩− ¬A iff w⊩+ A
• w ⊩+ A ∧ B iff w
⊩+ A and w ⊩+ B
• w ⊩− A ∧ B iff w⊩− A or w
⊩− B
• w ⊩+ A ∨ B iff w⊩+ A or w ⊩+ B
• w ⊩− A ∨ B iff w⊩− A and w ⊩− B
• w ⊩+ àA iff for some w' such that wRw', w ⊩+ A
• w
⊩− àA iff for all w' such that wRw', w ⊩− A
• w ⊩+ A iff for all w' such that wRw', w ⊩+ A
• w ⊩− A iff for some w' such that wRw', w ⊩− A
Validity (⊨) is defined as preservation of truth at @ in every interpretation.8
A simple induction shows that for every formula, A, and world, w, w ⊩+ A
or w ⊩− A, but not both. Hence, given that no constraints are placed on R, the
logic delivered is simply the modal logic K.
It should be noted that an interpretation is simply a piece of mathematical
machinery. In particular, X is any old set of objects. These are not to be con-
fused with possible worlds themselves. We may naturally suppose, however,
6
See Priest 2008: Chs. 2, 3.
7
What follows applies equally to first-order languages, but their specificities are not rele-
vant to the following considerations.
8
In some standard presentations, there is no designated world, @, and validity is defined
as truth preservation over all possible worlds. As long as no special constraints are put on
@, this is, of course, equivalent. However, it will be useful in what follows to have @ at
our disposal.
Thinking the Impossible 183
that there is one interpretation of the language which is in accord with the real.
(In this, X is the set of real possible worlds, @ is the real actual world, R is the
real relation of relative possibility, and ν+(p)/ν−(p) are the sets of worlds in
which p—understood as some meaningful sentence—is really true/false.) That
is why we can reason using modal logic about reality (not just actuality: actuali-
ty is just one world of the plurality of worlds). Why do we require a plurality of
interpretations to define validity? For the same reason that we do in the case of
propositional non-modal logic. We want our inference relation to be applicable
whatever reality is, in fact, like.
Of course, none of this tells us what kind of thing possible worlds are. Phys-
ical objects or abstract objects, existent objects or non-existent objects? There are
many well-known accounts of this matter;9 and which is right need not concern
us in this essay. Of more concern here is possibility itself.
simply to hold at some world. How to describe this kind of modality, one might
argue about. I shall simply call it logical, λ. (Though it is worth noting that logi-
cal necessity in this sense will include things that are not formally logically nec-
essary, including analytic truths such as ‘all red things are coloured’ and math-
ematical truths such as ‘there is an infinitude of prime numbers’.) Anything that
is possible in any more restricted sense is possible in this sense; and Rλ is simply
the universal relation: every possible world accesses every other. Hence, for any
𝜅 Î K, we have:
• ⟨𝜅⟩
A ⊨ ⟨λ⟩
A
• [λ]A ⊨ [𝜅]A
and the modal logic of λ is S5.
Here we see the beginning of a problem. Some things that are epistemically
possible would seem to be logically impossible. Thus, before Wiles’ proof of the
truth of Fermat’s last theorem, its negation was epistemically possible, though
logically impossible.
3. Impossible Worlds
3.1 The Primary Directive
The rediscovery of modal logic in the 20th century was in the work of C.I. Lew-
is between the two World Wars. Possible-world semantics came to prominence
in the 1960s and 70s. At first, under the influence of Quine’s attack on things
modal, possible worlds and their machinations were considered creatures of
darkness. But the clarity of the mathematics involved, and their usefulness in an
analysis of many things other than modality—such as conditionals, meaning,
knowledge and belief—meant that they soon became part of the intellectual
landscape. The philosophical debate around worlds changed from whether one
can make sense of them to how best to make sense of them, given the slew of
theories about their nature.
Impossible worlds rose to prominence some 20 years later. Under a very
different ideology (that of the unintelligibility of inconsistency), they, too, were
often taken to be creatures of darkness. Current debates may still concern
whether one can make sense of them. However, their mathematics is clear, and
their applicability to many philosophical areas—including some of those in
which possible worlds were clearly problematic—have, I think, ensured, that
they will soon be as much part of the landscape as possible worlds.11
Of course, for certain notions of possibility, impossible worlds can be ac-
commodated in possible-world semantics. Thus, physically impossible worlds,
where, say, a particle accelerates through the speed of light, are logically possi-
ble; and so can be accommodated in a model which allows for all logical possi-
bilities.
The main problem is with logical impossibilities themselves. On standard
possible-world semantics there are no worlds which realise these. But if there are
worlds which do so, there must be worlds where logical impossibilities hold;
and dually, worlds where logical truths fail. There appears to be no reason to
distinguish between different kinds of logical truths and falsehoods in this re-
11
On these issues, see Priest 1997a and Berto 2013.
Thinking the Impossible 185
12
I note that if we make all the propositional parameters true and false at w, then every
formula is true there; and if we make all the propositional parameters neither true nor
false at w, all formulas are neither true nor false there. Hence, the primary directive can,
in fact, be satisfied with just these two worlds. However, the two, on their own, hardly do
justice to the diversity of impossible situations.
13
Any such world is obviously closed under the consequence relation. Conversely, if w
does not satisfy Exh and Exc, it is clearly not closed under the relation. If it accesses a
world, w', that is not so closed, then for some, p, p is either both true and false at w' or
neither true nor false there. In the first case, à(p ∧ ¬p) is true at w; but in S5, à(p ∧ ¬p) ⊨
A, so the world is either not closed under the consequence relation, or is trivial. In the
second case, it is not true that (p ∨ ¬p) at w, so the world is not closed under Necessita-
tion.
186 Graham Priest
S5, except the trivial world.14 I note that the Primary Directive may no longer be
satisfied for modal formulas involving λ. Thus, in LP (p ∨ ¬p) will hold at all
worlds; and in K3 à(p ∧¬p) will fail at all worlds. This will be rectified with the
Secondary Directive, as we shall see in a moment.
It is sometimes touted as a virtue of possible-world semantics that they pro-
vide a reductive account of (logical) possibility. To be possible is simply to hold
in some world. Exactly the same is true if P is X; but of course, it is not true if P
is a proper subset of X. Reduction has alway struck me as a dubious virtue,
however. Why should one expect such a reduction? We obviously don’t have it
for all the other kinds of possibility; why just this one? Or better, we have to give
a non-reductive account for the other notions of possibility, so we ought to be
able to do it for this one too. We have just seen how.
More importantly in the present context: a natural thought is that if, at a
possible world, something is possible in any sense, it is logically possible. For
some notions of possibility this seems right. We would expect any physical pos-
sibility to be a logical possibility. And if X = P then everything is logically possi-
ble.15 Indeed, for any 𝜅 Î K, if w Rkw' then wRλw'.
But at least if P is a proper subset of X, there are good reasons why this
should not hold for all 𝜅 Î K. Consider epistemic possibility. Take a logical un-
truth, A, of enormous complexity: one which it would take longer than the his-
tory of the cosmos to decide. As far as is known, A could be true. So there must
be some w Î X − P, such that @Rεw.
Or again, suppose, for the sake of illustration, that the Law of Excluded
Middle is a logical truth. Let us suppose that intuitionist critiques have been so
fierce that we are now no longer sure whether A is true or not, where, this time,
A is: either there are or there are not 17 consecutive 0s in the decimal expansion
of π. It could be false for all we know; but A is false at no logically possible
world, so there must be a w Î X − P, such that @Rεw. Or a more realistic exam-
ple. Let G be a statement of Goldbach’s Conjecture. (Every even number greater
than 2 is the sum of two primes.) The conjecture is currently undecided. It is
true for all we know; it is false for all we know. Hence there are worlds w1 and
w2, such that G is true at w1, false at w2, and @Rεw1 and@Rεw2. Either w1 Î X−P
or w2 Î X−P.
Or consider deontic modality. I may promise to do something logically im-
possible (such as prove some mathematical statement which is, as a matter of
fact, false). Or I may make promises to do incompatible things, such as to be in
two different places at the same time (assuming such to be impossible). I am
then morally obliged to do impossible things;16 that is, the worlds that realize my
obligations are impossible. Hence, for any w such that @Rδw,
14
For these non-classical modal logics, see Priest 2008: Ch. 11a.
15
As argued by Mortensen 1989. Even if X is not P, this may still be the case. In FDE and
LP everything is logically possible, because of the trivial possible world. The thought that
every situation is logically possible may initially seem an odd one. But it should be re-
membered that logical possibility is a very weak constraint. Even if one is of a classical
persuasion, it is a logical possibility that I can jump a kilometre into the air, that the
moon is made of blue cheese, etc. Usually, when we are concerned with possibility, we
are concerned with much more restricted notions, especially physical possibility.
16
See Priest 1987, Ch. 13.
Thinking the Impossible 187
w Î X − P.
Sometimes, at this point, people will say things like ‘Of course we are as-
suming an ideal agent’. A poor move. It helps us not one iota to understand
what it means for us to know or be obliged to do something, if all we understand
is what it is for God to know or be obliged to do it. We are interested in notions
that apply to us: not God.
17
See Priest 2008, Ch. 5.
18
One notable exception: logics where the inference A ⊢ A fails.
188 Graham Priest
3.4 Generalising
By taking advantage of the fact that the set of truths/falsehoods in any world in
X can be imitated by a world in Z, we can, in fact, make matters more uniform
(and cut out the first stage in the construction of impossible worlds). An inter-
pretation is now a tuple ⟨W, @, {Rk : 𝜅 Î K}, ν⟩. W is a set of worlds; @ Î W;
for 𝜅 Î K, Rk is a binary relation on W;21 and for any formula, A, ν±(A) are sub-
sets of W. The truth conditions for @ are as in 2.1/2.2, and the truth conditions
for any other world, w, are as in 3.3. What are the possible worlds? We may
simply take these to be those closed under the S5 version of whichever of our
four logics we hold to be correct (or if this is explosive, all such worlds except
the trivial world).22
The techniques employed here are also generalisable in natural ways to
most other standard propositional logics—not just the four we have met: FDE,
19
In truth, it is only ν+ that is required to deliver the Secondary Directive. We cannot
give up ν−, though, since it may be involved in the falsity conditions of modal formulas at
@; though this leaves us free to impose constraints on ν− if required for any reason.
20
Suggested to me by Hartry Field.
21
In fact, we do not need to consider what is accessed by non-@ worlds under Rk, since
the truth/falsity of modal sentences at such worlds is taken care of by ν. We may there-
fore take Rk simply to be of the form {⟨@,y⟩
: y Î Y} for some Y ⊆W.
22
And the new context may suggest some new members of K, such as ‘It is intuitionisti-
cally possible that ...’.
Thinking the Impossible 189
LP, K3, and classical logic. For example, if we take as our basic propositional
logic an LFI,23 possible worlds have this as their underlying logic, and the nega-
tions of modal statements are given non-deterministic truth conditions. Since
LFIs are paraconsistent logics, all worlds obtained in this way are possible. Or
we may take a propositional logic that itself has world semantics. Thus, the
Kripke semantics for intuitionistic logic adds a binary accessibility relation to
give the truth conditions of the conditional.24 All the worlds delivered are possi-
ble. Or the Routley/Meyer world semantics of relevant logics add a ternary
accessibility relation to give the truth conditions of the conditional. 25 The se-
mantics themselves specify possible (normal) and impossible (non-normal)
worlds. In all cases, open worlds may be added to the models, in order to satisfy
the Secondary Directive (and so the Primary Directive if it is not already satis-
fied).
’Tis an establish’d maxim in metaphysics, That whatever the mind clearly conceives
includes the idea of possible existence, or in other words, that nothing we imagine is abso-
lutely impossible (Hume 1739-40: 32).
23
See Carnielli, Coniglio and Marcos 2007, and Bueno-Soler 2012.
24
See Priest 2008, Ch. 6.
25
See Priest 2008, Ch. 10.
26
OED, to conceive: ‘to take or admit into the mind, to form in the mind, to grasp with the
mind’.
27
OED, to imagine: ‘to form a mental image of, to represent to oneself in imagination, to
create as a mental conception, to conceive’. There is one sense of the word according to
which what is imagined ‘should not be known with certainty’ (OED, again). This is not
the sense at issue here.
28
For Hume, for something to be absolutley impossible is for it to imply a contradiction.
(See Lightner 1997: 115.) I take it that he holds that the negation of any “relation of ide-
as” would do this.
190 Graham Priest
29
See Yablo 1993. Yablo’s own account of conceivability (in Section 10) is that A is con-
ceivable if one can imagine a world that verifies A. In fact, I agree with this, since I take
everything to be conceivable/imaginable. This is not what Yablo intends, however. For,
by ‘world’, he means ‘(classically) possible world’. Yablo tells us (30) that one cannot
imagine, e.g., tigers that lick all and only those tigers that do not lick themselves. I find
this no harder to imagine than a set that contains all those sets which are not members of
themselves. (And I could imagine this even before I became a dialetheist.)
30
For discussion and references, see Priest 2006: 3.3.
31
Priest 1997b.
32
There is a somewhat thorny issue here about what it is, exactly, to understand. Can a
congenitally blind person understand the predicate ‘is red’, for example? I am inclined to
the view that they can, if they can use the word—by whatever means—in a roughly nor-
mal way. When they imagine something red, the phenomenological content may, how-
ever, be quite different from that of a sighted person who imagines something red.
Thinking the Impossible 191
@Rγ w for all w Î W. Given the Primary Directive, for any A, ⟨γ⟩A is true at @,
and [γ]A is not true at @. We may take γ to be the modality of concep-
tion/imagination: ⟨γ⟩A is ‘A is conceiveable/imaginable’.33 I should note that,
strictly speaking, conceivability is agent-relative (as is knowability). In particu-
lar, the As in question have to come from a language that the agent in question
understands (in a way that, say, a medieval monk could not understand the
language of quantum mechanics).34
Three objections. One. It might be suggested that if I seem to conceive of
(imagine) something that is impossible, I am, in fact, conceiving something else.
Thus (assuming that identities are necessary), when I conceive that water is not
H2O, what I am actually conceiving is that some substance that is a colourless,
odourless, potable liquid—even called ‘water’—is not H2O.35 Of course, I can
imagine that too; but that is not what I am imagining when I imagine that water
is not H2O: I am imagining something about water. The imagination is de re. In
the same way, when I imagine that Sarah Palin was the US Vice President after
the 2012 US election, I am imagining something about Palin. When I imagine
that Routley found a box that was empty and not empty, it is him that I imagine.
And when I imagine that 361 is a prime number (it isn’t) I am imagining some-
thing about that very number.
Two. It might be suggested that this is not the notion of conceivability oper-
ative in Hume’s dictum, since one who imagines impossibilities is not clearly
conceiving. If one takes it that one can clearly conceive only what is logically
possible, this turns Hume’s dictum into an empty tautology—and a useless one,
since we may not know what is impossible in this sense. If one is using the word
in a more common-sense way, it is something of an insult to say that a logician
or mathematician who conceives of impossibilities is not conceiving these things
clearly, since it is tantamount to an accusation of confusion. Perhaps, there is
some other notion of conceivability that satisfies Hume’s dictum, and which can
serve as a test for possibility. If so, I leave it to others to articulate it. I know of
no satisfactory such articulation.36
33
Semantics for a logic of imagination can be found in Niiniluoto 1987, Costa-Leite
2010, and Wansing 201+. These are all variations on possible-world semantics, and hence
do not allow for imagining the impossible. Even worse, they all require imagination to
satisfy certain logical closure conditions. Thus, they all validate the principle that if A is
imagined, and A is logically equivalent to B, then B is imagined. This is clearly incorrect.
A is logically equivalent to (A ∧ C) ∨ A, but I can imagine that Sherlock Holmes lived in
Baker St without imagining that (Sherlock Holmes lived in Baker St and E = mc2, or
Holmes lived in Baker St). Nothing about Special Relativity need have crossed my mind
at all. It is precisely this to which the Secondary Directive caters. Berto (2012: Ch. 7) has
a semantics for conception/representation which uses impossible worlds. He does not
require that everything be conceivable, but the semantics does allow for that possibility.
34
One might also doubt that a person understands indefinitely long sentences of such a
language. By the same token, one might doubt that such sentences are really grammati-
cal. One might therefore be inclinded to put the same bounds of finitude on both both.
35
See Berto 2012: 6.3.2.1 for references and discussion.
36
Chalmers 2002 constructs an eightfold taxonomy of notions of conceivability, and
argues that at least one of these entails possibility: ideal primary positive conceivability.
This may well be different from the notion of conceivability I am discussing here—
though the circularity in his glosses of these notions make me less than certain. But in
any case, one thing is clear: the ideality involved is that of some infinite and infallible a
192 Graham Priest
5. Conclusion
The Norwegian explorer Fridtjof Nansen said: “The difficult is what takes a
little time; the impossible is what takes a little longer”.41 Philosophy plays the
long game. The impossible has always been a marginalised character in Western
philosophy. The infinite had always been a marginalised character in mathemat-
ics until the time of Cantor. But just as Cantor provided an understanding of the
mathematical structure of the infinite, modern logic—especially paraconsistent
logic—has provided an understanding of the mathematical structure of the im-
possible. One can hardly pretend that this is an achievement on the scale of
priori reasoner—not a very useful notion for mere mortals.
37
OED, to suppose: ‘to think or assume that something is true or probable but lack proof or
certain knowledge’, ‘used to introduce a hypothesis and imagine its development’.
38
See Priest 1995: 4.8. Again, I am assuming that the agent understands the term ‘t’.
39
Here, ε is the indefinite description operator: a (particular) object such that.
40
As the old saying goes: be careful of what you wish for; you might just get it.
41
Cohen and Cohen 1992: 291.
Thinking the Impossible 193
Cantor’s (at least so far). However, I think that it has the potential to open peo-
ple’s eyes in the same way. Maybe even wider.42
References
42
Versions of this paper were given at the Fordham University Metaphysics and Mind
Group, the Departments of Philosophy at the University of Amsterdam and the Australi-
an National University, and the conference Thinking the Impossible, at the University of
Turin. Thanks go to those present for their helpful comments. Thanks, too, go to Hartry
Field for comments on an earlier draft.
This paper has already appeared on Philosophical Studies, 173 (10), 2016, 2649-62: many
thanks to the editors of Philosophical Studies for having agreed to republish the paper in
this journal.
194 Graham Priest
Priest, G. 1997b, “Sylvan’s Box”, in Priest (1997a), 573-82; repr. in Priest 2005, Ch. 6.6.
Priest, G. 2005, Towards Non-Being: The Logic and Metaphysics of Intentionality, Ox-
ford: Oxford University Press.
Priest, G. 2006, Doubt Truth to Be a Liar, Oxford: Oxford University Press.
Priest, G. 2008, Introduction to Non-Classical Logic: From If to Is, 2nd ed., Cambridge:
Cambridge University Press.
Smith, R. 2011, “Aristotle’s Logic”, in Zalta, E. (ed.), Stanford Encyclopedia of Philos-
ophy, http://plato.stanford.edu/entries/aristotle-logic.
Wansing, H. 201+, “Remarks on the Logic of Imagination”, Synthese, to appear.
Yablo, S. 1993, “Is Conceivability a Guide to Possibility?”, Philosophy and Phenome-
nological Research, 53, 1-42.