My review of Giere's Cognitive Models of Science (1992) is written from the standpoint of contemporary behavior analysis. Science is an activity, so what we know about behavior should be relevant to it. Yet cognitive approaches do not take facts about behavior into account. The themes I consider include the limitations of language, the ambiguity of knowledge structures, and some implications of information processing. For example, cognitive theories of information processing fail to deal with the finding that organisms do not work to produce information per se; they only work to produce information correlated with reinforcing events. Cognitive models of science have little to say about those major scientific innovations, obvious in retrospect, that were slow to be accepted because their content was not what people wanted to hear.
2. The most informative chapter for me was Bradshaw's chapter on the Wright brothers' invention of the airplane. Maybe that is because of all of the chapters it was the one most concerned with what scientists actually do (Tweney's chapter was also of interest, in its account of how Faraday maintained his notebooks). Certainly it is not bad advice to suggest that variables be studied one at a time before their interactions are tackled. That is what experimental control is all about. The chapter also called to my attention how precious little of the rest of the book is concerned with scientific methodology. I doubt that working scientists will come away from this volume with many good ideas about how to do science better.
3. Although I was glad to find in Gooding's chapter (p. 45) the statement that "The new naturalism in science recognizes that people learn by active intervention in a world of objects and other people," I was surprised that the point needed to be made. If science is not interaction with the world, what is it? To those who wish to appeal to knowledge structures, what good are knowledge structures unless they involve knowledge of properties of the world? When Gooding later (p. 47) refers to "the fact that scientists manipulate material things as well as representations," the puzzle is that material things are given lower priority than representations, though the former are prerequisites for the latter and are their most important derivatives (the Wright brothers' airplane was a crucial embodiment of their knowledge of aerodynamics). It is my very strong impression that those who approach science from a cognitive point of view typically spend too much time on knowledge structures and too little on the activities that produce them.
4. But thought experiments deserve attention and Gooding's concern with procedure is commendable. I agree with him that much that is important in science occurs at what he calls "the frontiers of observation" (p. 52), describing it as "activity that is unpremeditated as well as preverbal." It is too bad that he does not seem aware of how closely his distinction (p. 53) between "explicit, transparent, declarative knowledge" and "implicit, opaque, and unpremeditated learning and invention" corresponds to Houts & Haddock's (p. 390) distinction between rule-governed and contingency-shaped behavior. Precisely the same contrast between verbal and nonverbal processes is involved in both distinctions, and a substantial empirical literature exists on the properties of rule-governed behavior and contingency-shaped behavior (cf. Catania, Matthews & Shimoff, 1990).
5. Both Gooding and Houts & Haddock are concerned with computational models that draw scientific conclusions from propositions (as in the chapters by Nowak & Thagard and by Freedman). Gooding (p. 53) identifies declarative knowledge with "the stuff of which computational discovery programs are made." As Houts & Haddock (p. 381) put it, such models "are indeed impressive at solving problems that have been well defined by a programmer. What such powerful demonstrations fail to show is how scientists find their problems in the first place." In other words, such models are crucially deficient without an account of how propositions are derived from observations.
6. Freedman's simulation of the latent learning controversy illustrates the difficulty; it pits the hypotheses of Tolman against those of Hull, and so completely misses other ways to handle the controversy that are orthogonal to Tolman's and Hull's treatments. For example, the various input statements fail to handle a critical problem in all latent learning experiments: even though the rat can negotiate the maze in the absence of food reinforcers, it is impossible to eliminate all other conceivable reinforcers (e.g., return to the home cage upon removal from the maze, or even just entry into an open alley instead of a cul-de-sac; for more detail, see pp. 85-86 in Catania, 1992). The simulation lacks appropriate alternative definitions of reinforcement. Terms such as learning, reinforcement, drive state, and habit strength were derived in some way from behavior, and computational simulations do not take that derivation into account.
7. I see the problem as one of how verbal behavior is determined by events in the world, as in learning the names of things and events. The issues are psychological, and involve the functions of language (psycholinguistics and linguistics have had little to offer, because they have stressed language structure to the neglect of language function). I have treated this issue in more detail elsewhere (Catania, 1993); a related point is made by Liberman and Aarsleff in their discussion of the importance to chemistry of Lavoisier's naming of chemicals (pp. 14-15 in Harnad, Steklis & Lancaster, 1976). Before words can be involved in the framing of definitions, the speaker must already have made the relevant distinctions in some nonverbal way (i.e., Gooding's unpremeditated and preverbal activities and Houts & Haddock's contingency-shaped behavior).
8. Houts & Haddock (pp. 388-390) deal with another language issue in their outline of B.F. Skinner's argument about the inevitable ambiguity of the vocabulary of private events (e.g., words such as "thinking," "feeling" and "imagining"). Skinner (1945) discussed the difficulties created by positing a language of private events not based on the public practices of the verbal community (his paper has often been taken as a defense of operationism but was in fact a renunciation of it). The issue was not the existence of private events such as thinking or feeling or imagining; rather, it was how we learn to talk about them. The words we use are taught to us by the same verbal community that provides us with all of the other words of our language, and this teaching can only be accomplished on the basis of what is publicly shared by the speakers and listeners of the verbal community.
9. The argument, often misunderstood, does not deny the private; instead it points out the limitations imposed on the vocabulary of private events by the fact that common vocabularies can be based only on what is mutually accessible to and therefore shared by speakers and listeners. If a private feeling has no public correlate, how can anyone ever tell when anyone else has it? If one cannot tell, how can one ever teach the other the word for it? (The argument is consistent with Wittgenstein's rejection of private languages.)
10. It follows from this analysis of the vocabulary of private events that the issue of verbal origins extends to all psychological terms. One problem with operationism was its acceptance of the common vocabulary in terms of which operations, especially including those of observation, were defined. For any such term, an essential question concerns how it could have been learned. The question always brings the argument back to the practices of the verbal community.
11. No doubt some readers will object to the argument, raising questions about the status of private events from an epistemological perspective. But epistemological arguments too are framed in the vocabularies of drawing inferences, making assumptions, being objective, making sense of things, and so on. To blunt the force of the argument, objections to it must explicate such vocabularies. From a behavior analytic point of view, they have no special priority. Just as we can ask about the origins of the language of private events, we can ask about the origins of these specialized terminologies, once part of folk psychology and now, perhaps functioning very differently, part of the technical discourse of epistemologists. After all, talking about knowing is behavior too.
12. It follows that any discussion of the internal processes of scientists is subject to the ambiguities of the language of the private. Unfortunately, the problems seem not to be recognized by many of the chapter contributors. For example, how can Nersessian (p. 17) judge whether Maxwell's visualizations or quantitative judgments came first? Nersessian (p. 20) wants "to show how we might conceive of what scientists like Faraday and Maxwell were doing with their images and analogies" and goes on to argue (p. 24) that "the function of the image is... to make certain structural relations visualizable." But what does it mean to say that the image makes something visualizable, and how does one distinguish between imaging and visualizing or decide which comes first? The problems seem analogous to those that doomed the enterprise of introspection in the early decades of psychology.
13. Carey's chapter provides another example. Categories such as concept, belief, and reference are subject to the same types of problems. Carey is concerned with teaching the concept of density to children. The relevant discriminations must involve higher order contingencies (e.g., you must be able to handle both volume and weight before you can talk about density; it is conceivable that you can make one of the others derivative, but you can do so only by invoking unusual and artificial contingencies).
14. I find Carey's educational application (p. 125) disturbing: "Students work with the models, and then work on mapping the models to the world." Scientists work the other way around, so Carey has it backwards. What the students should be learning is that we start with the world and then make models of it. I find it hard to imagine that the students would have had the same problems if the teachers had started with actual physical systems and had introduced the analogs later. We know enough about pedagogy to be able to serve students more effectively than this (both Piaget and Skinner have provided good direction). In considering the commensurability of child and adult concepts, Carey at least approaches a distinction that I wish more developmental researchers would take into account. It has been carefully enunciated by Wellman (1990): the child's theory of mind and the developmentalist's theory of the child's theory need not resemble each other.
15. In their discussion of verbal behavior, Houts & Haddock may have put some readers off by suggesting that the main source of verbal behavior is the reinforcing consequences provided by the verbal community. This makes it seem (incorrectly) as if they would place the verbal behavior of an individual whose psychiatrist has shaped up stories of interactions with flying saucer crews on par with that of a scientist who has conducted experiments and provided a full account of methods and data and conclusions drawn from them. Verbal communities are crucial, but what has been left out are the special practices of scientific verbal communities. Those practices demand certain kinds of correspondences between what has been done and what has been found and what is said about them. Operationism was a first step in formalizing such verbal practices, but it was based upon an inadequate characterization of the functional properties of verbal behavior.
16. Scientific practices concerned with the integrity of scientific reports have been shaped up in scientific verbal communities because our effective interaction with events in the world is incompatible with the social contingencies that exert much of the control over our verbal behavior in other contexts. Anyone who is in doubt about this need only try to talk about the health hazards of cigarettes with a representative of the tobacco industry or to present the facts of evolution to an audience of religious fundamentalists. Implications of the relevant facts, and therefore the facts themselves, are not what these audiences want to hear.
17. Many contributors treat scientific theories as representational systems. For example, Nersessian (p. 9) distinguishes among propositions, mental models, and images as types of representational systems. Once again the emphasis is on knowledge structure rather than on acts of knowing. Saying, simulating, and imagining are things we do, and are not out of reach of a behavioral account. Some cognitivists might object that these activities do not involve muscle movements or glandular secretions. Yet the obsolete idea that behavior is restricted to muscle twitches and glandular squirts, though part of John B. Watson's methodological behaviorism early in the century, is not essential to contemporary behavior analysis.
18. Most of Watson's interpretation of thought and other complex processes was in terms of the development of abbreviated (and therefore covert) forms of overt behavior. Watson rejected an alternative view that might have included various central processes as behavior and thereby conceded to others the activities that would later be included under the rubric of cognitive processing: "The alternative sometimes advanced to this theory is that so-called central processes may take place in the brain so faintly that no neural impulse passes out over the motor nerve to the muscle, hence no response takes place in the muscles and glands" (Watson 1930, p. 239).
19. Many processes called cognitive (e.g., imagining and visualizing) are kinds of behavior that do not manifest themselves as muscle movements. They are typically observed indirectly. Consider the behavior called paying attention. It is not enough to ask whether someone is attending. The problem of introspection is the same as in Watson's time: it is unreliable. As already discussed, reports of covert behavior (private events) are constrained by the limited ways in which verbal communities can establish reliable vocabularies based on such events.
20. For visual stimuli, attending may superficially resemble looking toward or even pointing at something. But treating attending as equivalent to eye movement implies inappropriately that looking without seeing is the same as looking and seeing. Separate acts of looking at a given place differ if different things are looked for each time. For example, scanning a page to find a name is not the same behavior as scanning it to find a definition (some researchers try to capture this difference by speaking of search images). Sperling and Reeves (1980) provide an illustration of how to measure shifts of visual attention that do not involve eye movements. Contingencies can be arranged for such classes of behavior, even though they cannot be identified as particular muscle movements. Methods for measuring attending, imagining, and other covert processes are necessarily indirect but make such varieties of behavior accessible to analysis (see Chapter 15 in Catania, 1992, for a more detailed account; note also that such measurement makes the private public).
21. The cognitive processing called encoding is also a type of behavior. (Like other behavior, it is affected by contingencies, in the sense that ineffective methods of encoding are likely to drop out in favor of effective ones.) When letters are presented visually, errors of remembering are typically based on acoustic rather than visual similarities (Conrad, 1964). This tells us that in remembering the subject's prior behavior with respect to the stimulus is more important than the stimulus itself. At recall, the subject can reconstruct the stimulus by reproducing that behavior.
22. This point is relevant to the concept of representations. Giere (p. xxvi) states that a behavior analytic approach rejects representations, but it rejects only copy theories. Representations need not be copies. For example, the spoken letter A represents the written letter A but is not a copy of it. Such representations are more like responses than like stimuli. Behavior is the way in which you can present (or re-present) something to yourself, but it does not imply that the organism stores copies. In this regard, behavior analysis shares its views with advocates of neural nets and connectionist systems (cf. Donahoe & Palmer, 1989). Too often, however, the ways in which the environment may enter into the selection of active units in networks are neglected (Churchland's chapter provides an example). To the extent that cognitive processes can be treated as varieties of behavior, behavior analytic and cognitive approaches should eventually converge on common treatments of knowing.
23. Information processing is a metaphor that appears at several points in the book. Grandy (p. 187) is almost certainly wrong when he maintains that our information processing "is mainly a filtering process that eliminates vast quantities of information to reduce the flow of information to something manageable by our central processing unit." We have known long enough that the relevant units are chunks and not Shannon-Weaver bits. But even without regard to such quantitative questions, we must face a simple fact about behavior: not all informative stimuli are attended to equally.
24. Of the various behavioral analyses carried out in recent years, one of the most significant in its practical implications for human behavior is Dinsmoor's (1983) body of work on observing behavior. Its main conclusion is that organisms do not work to produce informative stimuli per se; instead, they work to produce informative stimuli correlated with reinforcers rather than those correlated with aversive events. Because the work has been discussed extensively in a Behavioral and Brain Sciences treatment, I will not recapitulate the details here. The point is that the finding undercuts the reliance on information processing as a primary cognitive process. Perhaps more important, it also implies that the effectiveness of a message depends more on whether its content is reinforcing or aversive than on whether it is correct or complete. For example, it is consistent with this finding that people often hesitate to have medical symptoms diagnosed. The phenomenon has long been recognized in folklore, as in accounts of the unhappy treatment of messengers who bring bad news or of Cassandra's fate in Greek mythology.
25. What if the acceptance of a scientific theory depends less on whether it is effective or true or coherent than on whether its competitors are more likely to tell a story that contemporary audiences want to hear? Darwin stands as perhaps the most relevant case in point. We admire him at least in part because he persisted even in the face of the unwelcome implications of his conclusions. His reputation for keeping careful notes on evidence that contradicted his theory is especially important. At some level (perhaps "unpremeditated and preverbal"), he must have recognized that such evidence might be the kind most likely to be forgotten. His notebooks were one device for countering the effects of unwelcome information. Galileo and Freud provide other clear examples. The obvious scientific inferences are not the problem. What needs to be explained is coming to a conclusion when the answer is not what you wanted to hear.
26. Behavior analysis itself might qualify as another example. Though it shares its opposition to dualism with contemporary cognitivism, it is much more likely to engender adverse reactions. That organisms work for information correlated with reinforcers rather than information per se is just one more fact about behavior that has been derived from behavior analysis, but to the extent that it may be resisted as a fact, it is a fact that illustrates itself.
27. If many cognitivists do not know about the status of information as a reinforcer, it is probably because they have ignored the knowledge on which contemporary behavior analysis is built (perhaps for reasons related to Dinsmoor's findings). In criticizing behavior analysis, they have worried so much about epistemological issues that they have neglected the subject matter (which is something like trying to understand the Copernican revolution while remaining ignorant of elementary physics). Too often judgments have been based on superficial readings or on secondary sources such as book reviews in linguistics journals.
28. For example, the concept of reinforcement is often debated among those outside behavior analysis without considering the elementary facts about the phenomenon. Reinforcement is no more a theory in behavior analysis than osmosis is a theory in cell biology. It is, first and foremost, a name for something that can be seen in behavior. As a phenomenon, it is ubiquitous but not universal, and one of the first skills a behavior analyst must acquire is to be able to identify it when it happens. The only increases in responding that qualify are those that occur because the responding has had consequences.
29. One common misunderstanding, pointed out by Houts & Haddock (p. 386), is that reinforcement assumes associationism and stimulus-response bonds. But the three-term contingency, in which a discriminative stimulus sets the occasion upon which responding has consequences, is not reducible to pairwise stimulus-stimulus or response-stimulus relationships (recently even conditioning theorists such as Rescorla have come to recognize this, though it is unfortunate that instead of acknowledging the longstanding and well-elaborated terminology of behavior analysis they have couched their insight in a separate vocabulary of occasion-setters).
30. In place of associationism, selectionism has become the dominant focus for behavior analysis. The shaping of behavior by its consequences can be demonstrated both in the laboratory (artificial selection) and in the world outside the laboratory (natural selection). The selection or survival of patterns of behavior in an organism's lifetime parallels the selection or survival of individuals in evolutionary time. Both involve variations that provide the source materials upon which selection acts (parallels among different varieties of selection have been explored in considerable detail: see, for example, Catania, 1992; Donahoe & Palmer, 1994; Skinner, 1981; Smith, 1986).
31. Selection provides a good basis for a philosophy of science. I wish a thorough treatment of it had been included in Giere's book. It is relevant because effective scientists must do at least two things well: they must produce many variants, and they must select among them. As their notebooks show, eminent scientists such as Darwin and Faraday were successful because they explored many different possibilities and tested them against the evidence. Keep a notebook, be generous in your entries into it, and review and edit it often. What better advice could we give to the aspiring scientist?
Catania, A.C. (1992) Learning (3rd ed.). Englewood Cliffs, NJ: Prentice-Hall.
Catania, A.C. (1993) The unconventional philosophy of science of behavior analysis. Journal of the Experimental Analysis of Behavior 60: 449-452.
Catania, A.C., Matthews, B.A. and Shimoff, E.H. (1990) Properties of rule-governed behaviour and their implications. In D. E. Blackman & H. Lejeune (eds.), Behaviour analysis in theory and practice. Hillsdale, NJ: Erlbaum.
Conrad, R. (1964) Acoustic confusions in immediate memory. Journal of Experimental Psychology 92: 149-154.
Dinsmoor, J.A. (1983) Observing and conditioned reinforcement. Behavioral and Brain Sciences 6: 693-728.
Donahoe, J.W. and Palmer, D.C. (1989) The interpretation of complex human behavior: Some reactions to Parallel Distributed Processing. Journal of the Experimental Analysis of Behavior 51: 399-416.
Donahoe, J.W. and Palmer, D.C. (1994) Learning and complex behavior. Boston: Allyn & Bacon.
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.
Harnad, S.R., Steklis, H.D. and Lancaster, J. (eds.) (1976) Origins and evolution of language and speech. Annals of the New York Academy of Sciences, Vol. 280.
Skinner, B.F. (1945) The operational analysis of psychological terms. Psychological Review 52: 270-277; (1984) Reprinted with commentary in Behavioral and Brain Sciences 7: 547-581.
Skinner, B.F. (1981) Selection by consequences. Science, 213, 501-504; Reprinted with commentary in Behavioral and Brain Sciences 7: 447-510.
Smith, T.L. (1986) Biology as allegory: A review of Elliott Sober's The nature of selection. Journal of the Experimental Analysis of Behavior, 46, 105-112.
Sperling, G. and Reeves, A. (1980) Measuring the reaction time of a shift of visual attention. In R. S. Nickerson (ed.), Attention and Performance VIII. Hillsdale, NJ: Erlbaum.
Watson, J.B. (1930) Behaviorism (revised ed.). New York: Norton.
Wellman, H.M. (1990) The child's theory of mind. Cambridge, MA: MIT Press.