The cognitive science of science is not yet here. This can be seen from the volume under review (Giere, 1992) by attending to the diversity and mutual incompatibility of its contents. I consider several of the papers in turn and then discuss reasons why there is not yet a cognitive science of science.
2. This is remarkable, since the notion that the theory of science ought to be an extension of the theory of people is sound and venerable. Theories, explanations, and experimental designs are abstract, formal, objects, but it has always been apparent that they are used by people. Understanding the constraints on human cognition would reveal what science must be such that we can do it. This was not anathema to logical positivism, as Giere (p. xv) suggests -- Rudolf Carnap and Otto Neurath each saw in the psychology of their time a resource for their respective theories of science. Carnap cited Gestalt psychology in support of his choice of the "elementary experiences" in the Aufbau (Section 67; for Neurath's naturalism see Uebel, 1992). Positivist-leaning psychologists formulated theories of science based upon their respective psychologies as well, as Laurence Smith (1986) has shown with regard to Tolman, Hull and Skinner. It is thus wrong to regard the psychological sensitivities of Campbell, Hanson, Kuhn and Quine as evidence of discontinuity with a preceding generation. Giere is right, though, that none of these gestures toward a science of cognition were successfully developed. Giere's view, and the apparent premise of this volume, is that this failure was due to the lack of a sufficiently developed cognitive science, but that now that we have a mature cognitive science, the cognitive science of science is at hand.
3. The book itself tells against that claim. It presents several attempted cognitive theories of science, many of them roughly sketched and mutually inconsistent. The contributors are united only in the view that the philosophy of science in some guise and cognitive science in some guise have something to do with each other. Indeed, that is too strong; Arthur C. Houts and C. Keith Haddock urge radical (i.e., Skinnerian) behaviorism, not the cognitive sciences, as the scientific basis of our understanding of science. Steve Fuller and Michael E. Gorman are also wary of a cognitive theory of science which purports to account for its irreducibly social character. Bracketing these three papers in a separate section, "Between Logic and Sociology," recognizes their opposition to the cognitive project, but Giere does not seem to appreciate the degree to which they fail to "fit [with the cognitive project] in a coherent fashion," as he suggests they should (p. xxvi). It is to Giere's credit that the volume includes these divergent positions, but the net effect is to underscore the very little that is held in common.
4. Cognitive science for Giere divides into the "disciplinary clusters" of cognitive psychology, artificial intelligence, and cognitive neuroscience, and the volume's remaining chapters are arranged accordingly. They reflect a bewildering array of perspectives on the cognitive science of science. I will consider a subset of the remaining eleven chapters to illustrate what I regard as a telling diversity.
5. Nancy Nersessian examines conceptual change via "cognitive- historical analysis." Her starting point is a near identification of scientific theories with the mental models postulated by Johnson-Laird (1983). Nersessian then provides an account of conceptual change in science and illustrates it with a case study involving Maxwell's understanding of electrostatic phenomena by analogy with mechanical phenomena. Mental model research is promising ground, but its application to central problems in the history and philosophy of science is, I believe, more complicated than Nersessian lets on. The proposed identification of a scientific theory with a mental model (an identification affirmed by Giere and presumed by Susan Carey) suggests a confusion between a scientific theory and the neural hardware which believes, rejects, entertains, or is otherwise involved with the theory (Clark Glymour raises essentially the same point against Paul Churchland's views (p. 466)). Indeed, Nersessian sometimes (pp. 9-10) writes as if scientific thinking amounts quite often to inventing mental models which satisfy an antecedently articulated scientific theory. If so, we have less an account of theories than an account of how scientists use them. At any rate, mental models aren't mentioned again, except critically by Houts and Haddock and in passing by Gooding and Gorman.
6. Conceptual change in developmental psychology is addressed in two long papers by Carey and Michelene Chi respectively. Against Spelke (1991), Carey argues that children's cognitive development involves genuine conceptual change (and subsequent incommensurability) rather than merely the enrichment of innate concepts, and Chi offers interesting thoughts about conceptual change as it occurs both within and across ontological categories. It is worth noting, though, that neither essay provides a distinction between a change in the meaning of a concept and a change in the concept itself. Chi and Carey both suggest, and intuition agrees, that a concept can retain its identity while changing its meaning; my concept of homosexuality, for example, probably remains just that after I read up on the latest research on sexuality's biological basis. However, my concept is unlikely to survive other semantic adjustments; were I to come to believe of homosexuality that it was a direct consequence of the proper functioning of the liver, essential to the European financial markets, and inversely correlated with the level of the Loch Ness, it would be best to say that I had traded in my old concept for a new one of the same name. Historians and philosophers of science have puzzled over real-life cases of conceptual change almost as fanciful (e.g., Kitcher, 1978), but none of this work has obviously influenced Carey or Chi. Yet without the capacity to make such a distinction, empirical work here is premature.
7. Input from AI is dominated by discussions of ECHO, the PDP implementation of Paul Thagard's theory of explanatory coherence for theory-choice. ECHO has been widely discussed, and amidst recognition of its merits there are recurrent sources of concern. As Thagard and Greg Nowak recognize, ECHO, in being set up to address a particular dispute, is fed explanatory relations as these are antecedently determined by the programmer. Moreover, ECHO is offered (at least by Thagard and Nowak) as the reconstruction of a scientific controversy from only one side, usually the winning side. (Also, it is it not clear (at least from descriptions in this volume) that ECHO's optimizations avoid merely local maxima in system coherence). Given these considerations, it is hardly surprising that Nowak and Thagard are able to show that "from Copernicus' perspective, his astronomical system gave a more coherent account of the observable features of the heavens" (p. 78). Eric Freedman's application of ECHO to the latent learning debate between Clark Hull and E.C. Tolman is more informative, in that here ECHO gives the wrong answer initially -- it picks Tolman's theory over Hull's, while Hull in fact triumphed historically. Explaining this result leads Freedman to some interesting considerations about tolerance for competing theories, although it is not clear that the conclusions he draws -- for example, that changing the importance of evidence for latent learning undermines the acceptance of Tolman's theories -- are in need of support from ECHO.
8. What is striking about this computational work is its a priori character. Nowak and Thagard are deeply involved in a historical case and take pains to capture it accurately, but they are not clearly committed (despite Thagard's comment, p. 485) to the very different notion that the scientists modeled by ECHO were in fact computing like ECHO. Freedman initially claims psychological accuracy, but later denies that any change of view was "based upon" ECHO-like computations (p. 331). Lindley Darden, describing her own (non-ECHO) simulation, explicitly denies that it models actual cognition (p. 252). Cognitive science has a formal side, and this is it. But it does not mesh well with the overt naturalism of the other chapters. Again, Giere is sensitive to the disparity, but I believe he errs in identifying computational approaches like Thagard's as empirical (e.g., p. xxiv) and thus he overestimates the contribution computational philosophy of science makes to a cognitive theory of science.
9. Last, Paul Churchland represents the field of cognitive neuroscience. Churchland's approach is both most at odds with others in the volume and most deserving of recognition as a naturalized philosophy of science. The premises of Churchland's arguments have been widely disseminated; here they are used to vindicate claims argued by Feyerabend over twenty years ago about theory-ladenness, incommensurability, eliminativism, and plurality. Questions and comments pose themselves (issues about the identification of theories with brain states and the role of truth in this realist philosophy strike me as most pressing), but my aim is again to note its disconnectedness with the other approaches. Indeed, Churchland's eliminativism would presumably undermine many of the conclusions elsewhere in the volume.
10. As of Oakland, California in 1937, so too it seems of the cognitive science of science in 1992: there is no THERE there. This is not to criticize the papers in the volume, nearly all of which are substantial and represent worthwhile research projects. Nor is it to condemn the volume, for I would argue that no better sense of the cognitive science of science emerges when recent efforts not represented in the volume are taken into account. The question then is: why do we not yet have an identifiable and compelling cognitive science of science in hand?
11. The question is taken up by Clark Glymour in a notorious commentary on Churchland, Thagard, and Giere's work that aspires to a kind of "emperor has no clothes" dismissiveness but conveys hastiness instead. Glymour's answer is that certain parties have let their intellectual guard down (pp. 470-471). More charitable readers will fault the great diversity within cognitive science, and recommend that the cognitive science of science be delayed until the science of mind suffers its Newtonian Revolution and subsequent synthesis. I sympathize with both answers, but there is a deeper one worth considering. Cognitive science, it has been urged, is profoundly interdisciplinary and perhaps irreducibly pluralistic as well. Taking this interdisciplinarity seriously may mean nothing less than giving up the hope, inherited from positivism, of a single, unified cognitive science of science. Such a change is not to be underestimated; it would make theory choice, conceptual change, and incommensurability live issues in daily practice. We may indeed find, in the end, that there is no "THERE there." Subsequent work in the cognitive theory of science would then help us see this as natural and appropriate.
Giere, R.N. (1993) Precis of Cognitive Models of Science. PSYCOLOQUY 4(56) scientific-cognition.1.giere.
Giere, R.N. (1992) Cognitive Models of Science. Minnesota Studies in the Philosophy of Science, volume 15. Minneapolis: University of Minnesota Press.
Johnson-Laird, P.N. (1983) Mental Models (Cambridge, MA: Harvard University Press).
Kitcher, P.S. (1978) Theory, theorists, and theory change, Philosophical Review LXXXVII (4), pp. 519-547.
Smith, Laurence D. (1986) Behaviorism and Logical Positivism: A Reassessment of the Alliance (Stanford, CA: Stanford University Press).
Spelke, E.S. (1991) Physical knowledge in infancy: reflections on Piaget's theory, in S. Carey and R. Gelman (eds.) The Epigenesis of Mind: Essays in Biology and Cognition (Hillsdale, NJ: Lawrence Erlbaum) (pp. 133-169).
Uebel, T. (1992) Overcoming Logical Positivism from Within: The Emergence of Neurath's Naturalism in the Vienna Circle's Protocol Sentence Debate (Amsterdam: Editions Rodopi).