Michael G. Shafto (1994) What can Insiders Learn From Outsiders?. Psycoloquy: 5(30) Scientific Cognition (4)

Volume: 5 (next, prev) Issue: 30 (next, prev) Article: 4 (next prev first) Alternate versions: ASCII Summary
Topic:
Article:
PSYCOLOQUY (ISSN 1055-0143) is sponsored by the American Psychological Association (APA).
Psycoloquy 5(30): What can Insiders Learn From Outsiders?

WHAT CAN INSIDERS LEARN FROM OUTSIDERS?
Book review of Giere on Scientific Cognition

Michael G. Shafto
Aerospace Human Factors Research Division
Mail Stop 262-1, NASA-Ames Research Center
Moffett Field, CA 94035-1000

shafto@eos.arc.nasa.gov

Abstract

Cognitive models enhance the analysis of the work processes of scientists. The same kinds of models may inform analyses of the cognitive processes of nonexperts learning how to "do" science. If successful, such a program of research could not only enrich cognitive theory, but also guide new developments in science education.

Keywords

Cognitive science, philosophy of science, cognitive models, artificial intelligence, computer science, cognitve neuroscience.
1. Could biologists, chemists, geologists, physicists, and mathematicians learn something useful about scientific work from psychologists, sociologists, historians, and philosophers? Cognitive Models of Science (Giere, 1992) can be read as a collective effort to give an affirmative answer to this question.

2. This brief review will focus on those chapters in which the authors present case studies or other kinds of data on the process of scientific problem solving. The goal of such research is to characterize these cases in terms of cognitive models. Such models are intended to support the analysis of the work processes of real scientists and engineers. The same kinds of models may inform analyses of the cognitive processes of nonexperts trying to learn how to "do" science and mathematics. If successful, such a program of research could not only enrich cognitive theory, but also guide new developments in scientific databases, electronic publishing, data analysis and visualization, and science education.

3. Cognitive neuroscience is an interesting subfield of cognitive science, and one which continues to make important theoretical and applied contributions. The contributions of this subfield to cognitive models of science, however, escapes me. Neurophysiology, elementary sensations, perception, and simple S-R behavior are too far removed from the specific nature of scientific work to be of much help. Commonalities cannot explain variation; thus, it is impossible to explain why science is "a way of knowing that has a uniquely transcendent value for all human beings" (Harris, 1979, p. 27) by appealing to physiological, perceptual, and S-R learning characteristics shared by humans, primates, and other organisms.

4. On the contrary, the characterization of science as a concrete process requires that we somehow focus on an unusual interaction of uniquely human cognitive abilities with "special" social conditions. The social band, however, is hard to study rigorously, because identifiable patterns do not repeat often enough -- that is, the relevant patterns may be "too unique." (See Newell, 1990, on the difficulties of extending cognitive models to social and historical processes.) So whereas commonality cannot explain variation, continual variation cannot be explained. This point of view serves to limit our aspirations when we try to gain insights into science as a cognitive, social, and behavioral process. It seems best to start with the process -- recognized either at a social or at an individual cognitive level -- and then ask which methods from cognitive science seem most likely to illuminate it.

5. Artificial Intelligence (AI), correctly and narrowly construed by Giere as "a branch of computer science," is likely to make increasingly important contributions to the integration, analysis, and interpretation of scientific data. To a limited extent, AI may improve the planning and execution of programs of scientific research. The main contribution of AI to cognitive models of science has been, and probably will continue to be, the development of tools for representing and processing non-numeric data (i.e., tools for symbolic information processing). As such tools become more reliable and more widely available, and as their theoretical underpinnings become better understood (Hoare, 1985; Milner, 1993), they will probably no longer be considered part of AI. For example, is Mathematica considered an AI program?

6. The kinds of computational tools that scientists find most useful may provide clues to how they prefer to represent, summarize, and reason about data. Certain computational methods may prove equally useful, though for different purposes, in developing tools to aid scientists and in developing models of their cognitive processes. Successful cognitive models of science may, in turn, suggest ideas for improved tools, including tools for science education. Tools, however, must pass the test of practical utility, whereas models must pass the test of empirical accuracy. Thus, as a branch of applied mathematics, AI is to cognitive psychology as stochastic models were to the experimental psychology of the 1950's and early 1960's: a source of tools and methods used to formulate and test substantive models.

7. Giere is right in noting that "the models being developed in cognitive psychology are, at least for the moment, the most useful for a cognitive approach to the philosophy of science" (Giere, 1993) -- granting the fact that psychological models have benefited from modeling tools borrowed from AI.

8. Nersessian's work is an excellent example of the application of mainstream methods to solve the problem of analyzing science as a concrete process. The characterization of scientific reasoning in terms of Johnson-Laird's theory of mental models shows the commonality between mundane reasoning and scientific reasoning. Thus, Nersessian -- like Langley et al. (1987) -- attempts to characterize scientific reasoning as an exceptional case of ordinary cognition. Giere's comment on this point seems to suggest that there is a significant contrast between the traditional view of theories as linguistic entities and Nersessian's account in terms of mental models. I would say, rather, that Johnson-Laird has provided a theory of how humans typically deal with theories and other linguistic entities: the cognitive view, in this case at least, is an extension of the traditional view. (This is convenient, since the traditional, linguistic view is probably correct!)

9. More generally, there is no conflict between Nersessian's approach and the traditional use of "abstractions like 'theories,' 'methods,' or 'research traditions'." It is necessary to focus on specific cognitive, behavioral, and communicative acts to gain a clearer picture of the details of scientific progress. Whether such a detailed focus is appropriate or not depends on the goals of the researcher. Cognitive scientists would be the last to object to using different levels of analysis for different purposes.

10. This point is made quite clear by Gooding's chapter on the various types of reconstruction used in the analysis of the path from experimentation to theory. Neither Gooding nor Nersessian, however, address the question of the SPECIFIC characteristics of scientific reasoning. Thus, Johnson-Laird's theory of mental models applies equally well to a broad range of reasoning tasks, including scientific reasoning, syllogistic reasoning, and sentence-picture comparison. Furthermore, it applies to (and is intended to explain) incorrect, as well as correct, reasoning. Many human performance tasks, such as piloting aircraft, playing chess, and troubleshooting automotive ignition systems, could be analyzed in terms of Gooding's levels of reconstruction. All of these are dependent upon complex procedural knowledge, much of which is rarely articulated or documented. By combining something like Nersessian's analysis with Gooding's, we could obtain a useful framework for looking at a wide variety of reasoning and problem solving tasks. Such a framework, perhaps further enriched by the kinds of semantic analyses seen in Carey and Chi's work, would be useful, for example, in designing instructional systems [see pars. 12 and 13 below] -- but EQUALLY useful for scientific and for nonscientific domains. This is good from the point of view of cognitive science, and especially from the point of view of the human factors of science; but it may leave the philosopher of science (Glymour, perhaps?) wondering what SPECIFIC insight has been gained into scientific practice. I believe there is a response to this issue, but one which is not developed in Giere's book. I will return to this point in the concluding section of this review [pars. 16-20 below].

11. Carey and Chi represent the application of cognitive science to the analysis and improvement of science education. This is the area where the contribution of cognitive science to scientific practice is the clearest and most impressive. One point which Carey and Chi illustrate is that cognitive structures and processes, and particularly cognitive change, are far more complex than common sense would suggest (see also Chomsky, 1959, 1984). Anyone who still doubts this claim, or who doubts that the effort of careful cognitive analysis is repaid in terms of practical benefits, should examine Schools For Thought (Bruer, 1993).

12. As Chomsky (1959) pointed out, complex issues such as incommensurability between children's and adults' concepts, novice- expert comparisons, and conceptual change in science cannot be approached without careful attention to cognitive theory. Discussion of ontological categories of material substance vs. constraint-based events, weight/density concepts, relational vs. substantial properties, and so forth, cannot make contact with scientific practice unless the analyst is armed with the kind of cognitive theory that has developed out of the work of Bruner, Chomsky, and Newell and Simon.

13. Gorman seems to make too much of the contrast between computational modeling and experimental simulation. The computational approach to cognitive modeling does not stand in opposition to empirical studies of the sort that Gorman advocates. Rather it can be argued that many strands of cognitive theory arose out of an increasing level of attention to detailed facts about cognition. For example, the work of Broadbent and of Estes grew out of their careful attention to the details of human and animal behavior. Bruner and his colleagues (1956) looked carefully at the details of reasoning and problem-solving performance, as did Newell and Simon. Even Chomsky, who addresses competence rather than performance, uses theory to direct attention to difficult FACTS about language and cognition. As Chomsky noted in 1959 and has continued to emphasize (1984, 1988), laboratory studies of cognition and learning do not score particularly high on "ecological validity," because they tend to eliminate significant phenomena. A much more promising approach would seem to be a combination of theoretical analysis and careful observation. (The work of Carey and of Chi can trace its roots back to Piaget's studies of child development -- again, theoretical analysis and careful observation.) From a Chomskyan perspective, it might seem that cognitive science has not yet gone far enough from the psychological laboratory toward the real world.

14. Gorman and Fuller suggest a reduced emphasis on individual cognition and an increased emphasis on science as a social activity conducted by a network of cognitive agents. This might lead us to ask about the necessary and sufficient characteristics of the individual cognitive agents, as well as to ask about the required ethical, linguistic, and technological support for the network itself. Fuller's historical argument, I think, provides an answer to the questions raised above [pars. 3, 4, 10] about the SPECIFIC insights provided by cognitive science into the nature of scientific practice, as opposed to a view of scientific practice as an instance of (more or less mundane) cognitive processes. Gorman and Fuller suggest that the missing ingredient may be the analysis of the scientific community as a SPECIFIC KIND OF network of cognitive agents. The specific characteristics of this community pick out science as a special type of cognitive activity. It may prove awkward to specify these characteristics at the level of the individual scientist, particularly at the level of individual cognitive abilities and processes. Rather, a fair argument could be based on the work discussed by Nersessian, Gooding, Carey, Chi, Bruer (1993), and Langley et al. (1987), to the effect that mainstream cognitive theory is committed to the notion that there is nothing special about science at the level of individual cognition.

15. To summarize: Cognitive science has made a great deal of progress in understanding and modeling individual cognition, including both knowledge structures and reasoning processes. The current state-of-the-art is well represented by Nersessian, Gooding, Carey, and Chi, who explain how scientific reasoning can be analyzed within the framework of current cognitive theory. However, Gorman and Fuller raise questions about the SUFFICIENCY of models of individual cognition for characterizing scientific reasoning, as opposed to any other kind of reasoning. The distribution of cognitive abilities among individuals remains approximately constant, while the "scientific abilities" of societies vary widely. Thus, by the argument given above [par. 3], science as a human enterprise cannot be fully characterized or explained by a reduction to individual human cognitive processes.

16. In conclusion, I would like to suggest a response to this issue. The response is intended to be consistent with Giere's cognitive perspective: theories, models, and scientific arguments are of secondary importance as units of analysis. We begin with a concept of individual human cognitive abilities in which language, broadly construed (Chomsky, 1984, 1988) plays a central role. Language is certainly tightly linked to the uniquely human cognitive capacities which make scientific practice possible. But language has multiple aspects: process/product, performance/competence, overt/covert, and -- most important in this context -- individual/social. Thus, in understanding science as a linguistically mediated enterprise, it is important to emphasize that the same language used as a medium for scientific communication also controls individual scientific thinking.

17. The focus on language is just the first step in trying to link individual cognition to the distributed cognition of the scientific community. The properties of language mentioned above [par. 16] are reflected in all uses of all languages. They therefore fail to capture the unique aspects of scientific thought and communication. These aspects include such features as the general concern for falsifiability and correctness, for broad theoretical coverage, unambiguous definition of basic vocabulary, refinable models, precise measurement, accurate fit of data to models, and definite measures of lack-of-fit. These sorts of concerns are what distinguish scientific discourse from other forms of discourse.

18. Underlying these concerns is a single concept or perspective. We might characterize the concept as "procedural objectivity," even though this may sound uncomfortably close to the (discredited?) notion of the "scientific method." By procedural objectivity, I mean an ethical attitude which affirms science as a unique "way of knowing... based on a set of rules designed to transcend the prior belief systems of mutually antagonistic tribes, nations, classes, and ethnic and religious communities in order to arrive at knowledge that is equally probable for any rational human mind" (Harris, 1979, p. 28). The perspective which accompanies this concept, and which must be spelled out more completely before the concept becomes clear, is what Harris (1968, pp. 568-604; Harris, 1979, pp. 32-41 et passim; and Headland et al., 1990) has called the i"etic" point of view. The "emic/etic" distinction can be glossed, for present purposes, as the distinction between insiders' and outsiders' points of view, with the caveat that the "outsiders" are scientists. Thus, a chemical engineer explaining how to control a refinery is using an emic (insider's, expert's) perspective; and a cognitive scientist discussing the declarative and procedural knowledge underlying refinery-control is using an etic (outsider's) perspective.

19. There are two points to be made about the emic/etic distinction in the context of this review. First, a chapter on the emic/etic distinction, and its application to the cognitive analysis of scientific reasoning, would have been a valuable addition to the book. Fuller touches on related themes at several points (pp. 436-440, 445-448). Cognitive scientists explaining to scientists what scientific practice "really is" are in the same position as anthropologists determining for the Aranda what their kinship structure "really is." That is, the cognitive scientists, like the anthropologists, first need to specify whether they want to assimilate Aranda kinship terminology to a global theory of kinship semantics (an etic perspective), or whether they want to describe how the Aranda think (from an emic perspective) about their own kinship, marriage, ancestry, and related concepts. Though equally valid, the etic and emic perspectives are quite different. They correspond to two different languages for describing the same phenomena. The constraints on the etic language, which have NO relevance to the emic language, are precisely those mentioned above [par. 17] as characteristic of scientific thought and communication.

20. Second, the emic/etic distinction arises as an important issue only in the social and behavioral sciences. Biologists, chemists, geologists, and physicists always operate from an etic perspective, and (I would argue) they have an unconscious and unquestioning belief in procedural objectivity. It is the foundational character of this belief that has been the focus of traditional philosophy of science. Cognitive models of science, and the innovative educational approaches based on them, may need to give more attention to modeling this belief and the etic perspective as central features of scientific thought. The trick is not to lose sight of the philosophical "what" of science as we continue to deepen our understanding of the cognitive "how."

REFERENCES

Bruer, J.T. (1993) Schools for thought. Cambridge, MA: MIT Press.

Bruner, J.S., Goodnow, J. & Austin, G.A. (1956) A study of thinking. New York: Wiley.

Chomsky, N. (1959) Review of Skinners Verbal Behavior. Language, 35, 26-58.

Chomsky, N. (1984) Modular approaches to the study of the mind. San Diego, CA: San Diego State University Press.

Chomsky, N. (1988) Language and problems of knowledge. Cambridge, MA: MIT Press.

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.

Harris, M. (1968) The rise of anthropological theory. New York: Harper Collins Publishers.

Harris, M. (1979) Cultural materialism: The struggle for a science of culture. New York: Vintage Books.

Headland, T.N., Pike, K.L. & Harris, M. (Eds.) (1990) Emics and etics: The insider/outsider debate. Newbury Park, CA: Sage Publications.

Hoare, C.A.R. (1985) Communicating sequential processes. Englewood Cliffs, NJ: Prentice-Hall.

Langley, P., Simon, H.A., Bradshaw, G.L. & Zytkow, J.M. (1987) Scientific discovery: Computational explorations of the creative processes. Cambridge, MA: MIT Press.

Milner, R. (1993) Elements of interaction: Turing Award Lecture. Communications of the Association for Computing Machinery, 36; 78-89.

Newell, A. (1990) Unified theories of cognition. Cambridge: Harvard University Press.


Volume: 5 (next, prev) Issue: 30 (next, prev) Article: 4 (next prev first) Alternate versions: ASCII Summary
Topic:
Article: