Dialogue

http://journals.cambridge.org/DIA

Additional services for Dialogue:

Email alerts: Click here

Subscriptions: Click here
Commercial reprints: Click here
Terms of use : Click here

Modelling Conceptual Revolutions

Paul Thagard

Dialogue / Volume 35 / Issue 01 / December 1996, pp 155 - 159

DOI: 10.1017/S0012217300008131, Published online: 13 April 2010

Link to this article: http://journals.cambridge.org/

abstract_S0012217300008131

How to cite this article:

Paul Thagard (1996). Modelling Conceptual Revolutions. Dialogue, 35, pp
155-159 doi:10.1017/S0012217300008131

Request Permissions : Click here

Downloaded from http://journals.cambridge.org/DIA, IP address: 130.132.123.28 on 17 May 2015

Modelling Conceptual Revolutions

PAUL THAGARD University of Waterloo

Like Clark Glymour, Don Ross does not appreciate what computer mod-
els can contribute to cognitive psychology and naturalistic philosophy
of science.1 This is not surprising, since the cognitive-computational
approach differs dramatically from the conceptual systems and method-
ologies familiar to most philosophers. I will therefore begin by reviewing
the use of computational models in cognitive science and then explain why
the computational model ECHO (Explanatory Coherence by Harmany
[sic] Optimization) is an essential part of my account of explanatory coher-
ence and conceptual revolutions. This should make it obvious that my
approach is not a search for anything like "computational foundations"
as Ross suggests. I will also reply to some subsidiary objections and point
to some recent work that expands the ideas in Conceptual Revolutions.
Many philosophers' view of reasoning divide it into two kinds, deduc-
tive and inductive, with first-order logic providing the standard way of
thinking about deduction, and other formal accounts such as probability
theory and confirmation theory providing the standard way of thinking
about induction. In contrast, those of us studying scientific thought
within cognitive science have a broader view of thinking, which empha-
sizes two components: structures and processes. In opposition to the
standard philosophical assumption that sentence-like structures are the
sole consituents of knowledge, in my book I described the importance of
concepts and conceptual systems. Cognitive science also has a much
richer view of processes than the traditional philosophical view of deduc-
tion and induction allow, making possible rigorous discussion of the
nature of hypothesis formation, concept formation, analogy, problem
solving and other important aspects of scientific thought. The interesting

Dialogue XXXV (1996), 155-59

© 1996 Canadian Philosophical Association/Association canadienne de philosophic
156 Dialogue

question is: what would it take to shift those who have not seen the advan-
tages of a cognitive approach away from the narrow logic-based view of
scientific reasoning? Naturally, my answer is explanatory coherence:
much more about the structure and development of science can be
explained within the broader framework.
For investigation of mental processes, computer models are indispen-
sable. Comprehension of their use requires noting the distinctions and the
connections among four crucial aspects: theory, model, program and
platform. A cognitive theory postulates a set of representational struc-
tures and a set of processes that operate on these structures. A computa-
tional model makes these structures and processes more precise by
interpreting them by analogy with computer programs that consist of
data structures and algorithms. Vague ideas about representations can be
supplemented by precise computational ideas about data structures, and
mental processes can be defined algorithmically. For the model to be
tested, it must be implemented in a software program in a programming
language such as LISP or C. This program may run on a variety of hard-
ware platforms such as Sun Workstations or IBM™ personal computers
(PCs), or it may be specially designed for a specific kind of hardware.
Many kinds of structures and processes can be investigated in this way,
from the rules and search strategies of some traditional sorts of artificial
intelligence to the distributed representations and spreading activation
processes of newer connectionist views.
The analogy between mind and computer is useful at all three stages of
the development of cognitive theories: discovery, development and evalu-
ation. Computational ideas about different kinds of programs often sug-
gest new kinds of mental structures and processes. Developing a theory
and a model often goes hand in hand with developing a program, since
writing the program may lead to the invention of new kinds of data struc-
tures and algorithms that become part of the model and have analogs in
the theory. Similarly, we often evaluate theory, model and program simul-
taneously, since one's confidence in the theory depends upon the validity
of the model as shown by the performance of the program. The program
can contribute to evaluation of the model and theory in three ways. First,
it helps to show that the postulated representations and processes are com-
putationally realizable. This is non-trivial, since many algorithms that
seem at first glance reasonable are intractable and could not be applied to
large problems on real computers.2 Second, in order to show not only the
computational realizability of a theory but also its psychological plausi-
bility, the program can be applied qualitatively to various examples of
thinking. Third, to show a much more detailed fit between the theory and
human thinking, the program can be used quantitatively to generate
detailed predictions about human thinking that can be compared with the
results of psychological experiments. Cognitive theories by themselves are
 Modelling Conceptual Revolutions 157

normally not precise enough to generate such quantitative predictions, but

a model and program mayfillthe gap between theory and observation.
Now it is possible to answer Ross's question about the purpose of the
computational model ECHO. He is puzzled about whether it adds any-
thing to the theory of explanatory coherence (TEC), which I used to
account for belief change in the major scientific revolutions. As suggested
in the last paragraph, ECHO is crucial to the evaluation of TEC in three
ways. First, the general problem of choosing the best explanation of a set
of observations is known to be computationally intractable (Bylander,
Allemang, Tanner and Josephson 1991). Any practical account of theory
choice needs a computational implementation to demonstrate that it is
realizable. For example, a crude Bayesian theory that requires a full joint
distribution of probabilities is not realizable for large examples, since the
provision of 2" conditional probabilities for n propositions will quickly
exhaust the capacity of any physical system.3 ECHO has the desirable
property that the number of cycles of activation updating required to
determine the best explanation does not increase with the number of
propositions to be evaluated (Thagard, forthcoming).
The second way in which ECHO is crucial to the evaluation of TEC is
that it made possible detailed and effective application to numerous cases
in the history of science. The largest of these, Newton versus Descartes
and Copernicus versus Ptolemy, each involved more than 100 proposi-
tions each and are, I believe, the most detailed analyses of cases of scien-
tific reasoning ever done (Nowak and Thagard 1992a, 1992b). ECHO
shows that TEC can be applied in great detail to real historical cases,
directly based on the writings of the relevant scientists. Contrary to Ross's
assertion that computer simulation leads to gross simplification, this
methodology encourages much more serious analysis of cases than is nor-
mally done in historical philosophy of science.
Finally, the third way in which ECHO has served to evaluate TEC is in
the quantitative fit to psychological experiments. The theory and model
have led to a series of psychological experiments designed to test it as an
account of human thinking involving competing hypotheses. Using
ECHO to generate predictions about human reasoning, psychologists
have designed experiments that have shown a good fit between how the
theory of explanatory coherence describes reasoning and how people
actually do reason, in the social domain and in examples relevant to sci-
ence education (Miller and Read 1991; Read and Marcus-Newhall 1993;
Read and Miller 1993; Schank and Ranney 1991, 1992). These studies
provide evidence that inference to the best explanation based on explan-
atory coherence is part of the cognitive apparatus of people in general, as
well as of the scientists who use it more explicitly. TEC and ECHO are
intended to be models of how scientists actually think, as well as of how
they should think.
158 Dialogue

Ross erroneously implies that ECHO and computational philosophy of

science in general depend on a kind of computational internalism that
ignores physical and social environments. An expanded version of Con-
ceptual Revolutions would indeed deal more extensively with the physical
environments that had a major effect on the observations and experiments
of scientists such as Lavoisier and Darwin, and it would also describe
more fully their social environments. But these additions would have to
mesh with the account of the conceptual changes that took place in indi-
vidual scientists. Much of my work in the past couple of years has been
concerned with overcoming the acknowledged omission of social factors
in the discussion of scientific revolutions. Cognitive and social explana-
tions of the development of science are best seen as complementary rather
than competitive. What is needed is an account of how cognitive and
social accounts of science can be integrated to provide unified naturalistic
explanations of a wide variety of features of scientific development. I
recently proposed that concepts from distributed artificial intelligence can
be used to tie together cognitive and social accounts of science (Thagard
1993b), and have classified the explanation schemas used by cognitive and
social theorists in order to show how unified schemas can be produced
and used (Thagard 1994). I look forward to the expansion of current cog-
nitive models of science and their further integration with social models.
But taking physical and social environments into account supplements
rather than replaces the kind of cognitive-computational explanation of
scientific thinking that I and others have been offering.
In sum, Ross has seriously underestimated what computational ideas
and models can contribute to naturalistic philosophy of science.
Notes
1 This note is a response to Don Ross's critical notice of my book, Conceptual
Revolutions. My intervention incorporates some responses I made to comments
by Clark Glymour and Ron Giere at the 1993 Eastern Division meeting of the
American Philosophical Association in a symposium on the book. A few para-
graphs are drawn from the English text that was translated for the Preface to the
Italian edition of the book published by Guerini e Associati. This research is
supported by the Social Sciences and Humanities Research Council of Canada
and the Natural Sciences and Engineering Research Council of Canada.
2 For a philosophical discussion of intractability, see Thagard 1993b.
3 Much more sophisticated Bayesian approaches have been developed. See, for
example, Pearl 1988, and see Thagard (forthcoming) for a comparison of
explanatory coherence theory with Bayesian networks.

References
Bylander, T, D. Allemang, M. Tanner and J. Josephson
1991 "The Computational Complexity of Abduction." Artificial Intelli-
gence, 49: 25-60.
 Modelling Conceptual Revolutions 159
Miller, L., and S. Read
1991 "On the Coherence of Mental Models of Persons and Relation-
ships." In Cognition in Close Relationships. Edited by F. Fincham
and G. Fletcher. Hillsdale, NJ: Erlbaum, pp. 69-99.
Nowak, G., and P. Thagard
1992a "Copernicus, Ptolemy, and Explanatory Coherence." In Cognitive
Models of Science. Edited by R. Giere. Minneapolis: University of
Minnesota Press, pp. 274-309.
1992b "Newton, Descartes, and Explanatory Coherence." In Philosophy
of Science, Cognitive Psychology and Educational Theory and Prac-
tice. Edited by R. Duschl and H. R. Hamilton. Albany, NY:
SUNY Press, pp. 69-115.
Pearl, J.
1988 Probabilistic Reasoning in Intelligent Systems. San Mateo, CA:
Morgan Kaufman.
Read, S., and A. Marcus-Newhall
1993 "The Role of Explanatory Coherence in the Construction of Social
Explanations." Journal of Personality and Social Psychology, 65:
429-47.
Read, S. J., and L. C. Miller
1993 "Explanatory Coherence in the Construction of Mental Models of
Others." In Proceedings of the Fifteenth Annual Conference of the
Cognitive Science Society. Hillsdale, NJ: Erlbaum, pp. 836-41.
Schank, P., and M. Ranney
1991 "Modeling an Experimental Study of Explanatory Coherence." In
Proceedings of the Thirteenth Annual Conference of the Cognitive
Science Society. Hillsdale, NJ: Erlbaum, pp. 892-97.
1992 "Assessing Explanatory Coherence: A New Method for Integrating
Verbal Data with Models of On-line Belief Revision." In Proceed-
ings of the Fourteenth Annual Conference of the Cognitive Science
Society. Hillsdale, NJ: Erlbaum, pp. 599-604.
Thagard, P.
1992 Conceptual Revolutions. Princeton, NJ: Princeton University
Press.
1993a "Computational Tractability and Conceptual Coherence: Why Do
Computer Scientists believe that P # NP?" Canadian Journal of
Philosophy, 23: 349-64.
1993b "Societies of Minds: Science as Distributed Computing."Studies in
History and Philosophy of Science, 24: 49-67.
1994 "Mind, Society, and the Growth of Knowledge." Philosophy of
Science, 61,4: 629-45.
Forthcoming "Probabilistic Networks and Explanatory Coherence. In Auto-
mated Abduction: Inference to the Best Explanation. Edited by
P. O'Rorke and J. Josephson. Menlo Park, CA: AAAI Press.