INTRODUCTION

There are few things that hold a greater fascination for us human beings than the project of explaining ourselves to ourselves. On the one hand, the human mind is the part of us that makes us who we are as individuals. It is also our minds that set us as a species apart from brute matter and from other members of the animal kingdom. The centrality of self-knowledge in Western philosophy goes back at least to Socrates' adherence to the motto inscribed over the temple of Apollo at Delphi: "Know thyself." On the other hand, the mind has proved one of the most intractable mysteries for modern science. Indeed, modern science, conceived as a discipline concerned with the lawful causal interactions of material bodies, has been hard pressed to accommodate the world of thoughts and concepts and images that seem essential to any treatment of the mind. One might even go so far as to say that the central problem of modern philosophy has been one of somehow closing the gap between two apparently incommensurable discourses: a discourse about our minds that speaks of ideas and images, and a discourse about the world of nature that speaks of causal relations between bodies in motion.

Since Alan Turing's introduction of the notion of a computing machine in the late 1930s, there has been a growing interest in a new paradigm for understanding the mind: a paradigm that treats the mind as a digital computer. The arrival of machine computation upon our intellectual landscape has had a profound and widespread impact upon research in the many disciplines that are concerned with the study of the mind. In fields such as cognitive psychology, ethology, linguistics, the philosophy

of mind, and cognitive neuroscience, the computational view of the mind has become a mainstream view—perhaps even the dominant view in recent years. Even though there is no monolithic consensus about how the computer paradigm is to be applied to the mind, and even though there are many researchers in all of the disciplines that study the mind who are working out of other traditions, it is by now generally agreed that the computational approach has emerged as a force to be reckoned with. And thus even writers who view the computer metaphor as essentially bankrupt have nonetheless felt moved to devote considerable ink to refuting it or establishing the merits of their own views against it.

I believe that there are two very different approaches that a philosopher may take to this very rich body of "computationalist" work in the study of cognition or "cognitive science." The first is that of the historian and philosopher of science. As a philosopher of science, one may look at the computationalist paradigm in psychology with an eye towards issues that are internal to the various sciences of cognition: What are the methodological assumptions of computational psychology? How do they differ from those of, say, behaviorism or associationism or neuroscience? What are psychological theorists really committed to in their use of theoretical terms such as 'representation' or 'syntax'? What are the issues that really stand between rival research programmes such as "good old-fashioned AI," which emphasizes rules and representations, and neural network approaches? What are the implicit assumptions of different theorists about the "good-making" qualities of scientific theories in a domain such as psychology?

On the other hand, the philosopher of mind may also look to the computational paradigm for answers to long-standing philosophical problems, such as the mind-body problem, issues about the metaphysical nature of the mind and the relationship between thought and matter, the relationship between psychology and the natural sciences, and the nature of intentionality. While there have been some welcome contributions of late to the history and philosophy of psychology that take a careful look at actual research in the sciences of cognition,^[1] by far the greater portion of philosophical interest in the computer paradigm has been concentrated on the more distinctively philosophical enterprises of explaining intentionality and "naturalizing" psychology by rendering its commitment to mental states and processes compatible with materialism and the generality of physics.

This book is intended as a contribution towards such an understanding of the nature of the computer paradigm and its importance to the

empirical study of cognition and to the philosophy of mind. It combines an extended examination of a "mainstream" approach to the importance of computation—the "Computational Theory of Mind" (CTM) championed by Jerry Fodor and Zenon Pylyshyn—with a preliminary articulation of an alternative approach to examining the importance of computational psychology. The thesis, in a sentence, is that CTM does not provide a solution to the philosophical problems that it is heralded as solving—indeed, it involves some deep confusions about computers, symbols, and meaning—but that this does not undercut the possibility that the computer paradigm may provide an important resource (for all we know, perhaps the key resource) for the development of a mature science of cognition. In short, we will be disappointed if we look to CTM for solutions to long-standing philosophical problems about the mind. But computational psychology is nonetheless a robust research programme that is deserving of philosophical study, and the final section of this book suggests an alternative approach to viewing computational psychology from the standpoint of the philosophy of science rather than that of metaphysics.

CTM claims that the mind literally is a computer. And what it is to be a computer, according to CTM, is to be a device that stores symbols and performs transformations upon those symbols in accordance with formal (or, more precisely, syntactic) rules. There are two distinct and important strands to this theory. The first strand is representational and consists in the claim that individual mental states, such as particular beliefs and desires, are relationships between an organism and mental representations . These mental representations are physically instantiated symbol tokens having both semantic and syntactic properties. This view, taken alone, Fodor sometimes calls the "Representational Theory of Mind." The Representational Theory of Mind (RTM) is a theory about the nature of individual mental states. The second thread of CTM is the claim that mental processes, such as forming and testing a hypothesis or reasoning to a conclusion, are computational processes that the mind performs upon these representations. That is, when the mind moves from one thought to another, it is generating new mental representations, and it does so by applying syntactically based rules to the representations already present in it, just as a digital computer generates new symbolic representations by applying syntactically based rules to existing representations.

CTM has generated a great deal of interest among philosophers because it goes beyond claims of (mere!) empirical utility for the computer

paradigm and makes substantive philosophical claims as well. Two such claims are of particular importance. First, it is claimed that CTM—or, more specifically, the representational component of CTM, RTM—provides an account of how mental states have such properties as meaning, reference, and intentionality. According to Fodor, mental states "inherit" their semantic properties from those of the representations they involve. The second claim is at least as bold: namely, that CTM provides a "vindication" of "intentional psychology" (that is, of psychology that is committed to a realistic construal of explanations in the intentional idiom) by showing that intentional explanations can be tied to nomologically based causal explanations that are in no way incompatible with materialism or with the generality of physics.

These two claims are bold and ambitious, to say the least. A theory that could accomplish either of these goals in isolation would be of considerable importance. A theory that accomplished both, while also being closely linked to a burgeoning methodological approach to actual research in cognition, could hardly draw more attention than it deserved. Indeed, if CTM succeeds as its advocates claim, the emergence of the notion of computation will have provided the basis for a revolution in the study of mind as fundamental and important as the Copernican revolution in astronomy.

I shall argue, however, that while the computer may ultimately provide the basis for the extension into psychology of the Galilean project of the mathematization of science, CTM's attempts to explain intentionality and to vindicate intentional psychology are based upon subtle but fundamental confusions. At the heart of CTM is the claim that mental states are relations to mental representations—to meaningful symbols—and that this accounts for their semantic properties and their intentionality. The crucial questions one must ask of CTM, therefore, are these: (1) Just what does it mean to say that mental states are "relations to meaningful symbols"? And (2) just how is the postulation of meaningful symbols supposed to explain the semantic properties and the intentionality of mental states? The first question calls for an examination of just what we are saying of a thing when we call it a "meaningful symbol." The second calls for an application of the results of such an examination to the formulations of CTM offered by Fodor and Pylyshyn.

It is both curious and unfortunate that these questions have received so little attention from philosophers of mind: curious because the ques-

tions seem so crucial to the assessment of a theory that is generally acknowledged to be of great interest and importance; unfortunate because an examination of these questions uncovers significant ambiguities in such notions as "representation," "symbol," "meaning," and "intentionality." Until we have acknowledged these ambiguities, it is impossible either to assess CTM or even to determine exactly what it is that it is claiming.

This problem has, I think, been touched upon by some writers—notably by Kenneth Sayre and John Searle, both of whom urge upon us the conclusion that there is something about symbols, and particularly about symbols in computers, that renders them unsuitable for an explanation of the meaningfulness of mental states. My criticisms of CTM run in the same vein. Where I part ways with Searle and Sayre is that they look for the problem specifically in the use of symbols in computers. In my view, however, the fundamental issue turns out in the end to have curiously little to do with computers. The issue, rather, is whether the notion of symbolic representation provides the bedrock upon which a theory of the intentionality of mental states may be built. My answer to that is no; and insofar as the project of vindicating intentional psychology (at least as envisioned by advocates of CTM) can be shown to depend upon its ability provide a theory of intentionality, that vindication fails as well.

Why can't one establish a theory of intentionality for mental states upon a foundation of symbolic representations in the mind? I am afraid that I do not know how to give an answer to that question that satisfies my own standards of rigor in less space than the several chapters it occupies in this book, but I shall try to give a short answer here that may prove helpful and not too inaccurate. When one takes a close look at what one is saying when one calls something a "meaningful symbol" or a "symbolic representation," it turns out that one is tacitly saying things about the conventions and intentions of symbol users. This is just part of what we are saying of a thing in calling it a symbol, and of what we are saying of a symbol when we say that it has semantic properties. But conventions and intentions of symbol users are ultimately facts about people's mental states. And so any explanation of the intentionality of mental states that rests upon the meaningfulness of symbolic representations ends up explaining the intentionality of mental states in a way that refers to other meaningful mental states. Thus one important problem with CTM's account of intentionality is that it turns out to be circular and regressive: circular because it explains the meaningfulness of

mental states by appealing to the meanings of symbols, while one must also explain the meaningfulness of symbols by appealing to the meanings of mental states; regressive because the explanation of any particular mental state will ultimately refer back to other mental states.

Moreover, it is not only the semantic properties of symbols that are conventional in nature; syntactic properties, and the very symbol types themselves are ultimately dependent upon conventions. (The fact that something is a letter p or an inscription of the English word 'dog' depends upon conventions that establish the existence of those symbol types.) In particular, the kinds of syntactically based rules that are necessary for compositionality are conventional in nature: in order to generate semantic properties for complex representations, it is not enough to have interpretations for the primitives and "syntax" in the weak sense of rules for legal concatenation or equivalence classes of legal transformations . Rather, one needs a stronger kind of syntax that involves rules for how syntactic patterns contribute to meanings of complex representations—for example, a rule to the effect that if 'A' means "X" and 'B' means "Y," then 'A-&-B' will mean "X and Y." The only way we know of getting this kind of compositionality is by way of conventions. It is not clear that there is any other way of getting compositionality; at very least, CTM's advocates would have to show how semantic composition could be achieved without the aid of conventions.

Now of course this argument rests upon a particular construal of what it is to be a symbolic representation or to be a meaningful symbol. But, as far as I am aware, this is the only sense of 'symbolic representation' and 'meaningful symbol' that we have. One is, of course, inclined to wonder whether perhaps writers like Fodor really mean something different when they speak of the mind containing "meaningful symbols." But if they do, it is curious that they never inform the reader that they are using familiar expressions in novel ways. Indeed, in one place Fodor gives a brief glance at this possibility only to dismiss the issue as unlikely to prove important.

It remains an open question whether internal representation, so construed, is sufficiently like natural language representation so that both can be called representation 'in the same sense'. But I find it hard to care much how this question should be answered. There is an analogy between the two kinds of representation. Since public languages are conventional and the language of thought is not, there is unlikely to be more than an analogy. If you are impressed by the analogy, you will want to say that the inner code is a language. If you are unimpressed by the analogy, you will want to say that the inner code is in some sense a representational system that is not a language. But in

neither case will what you say affect what I take to be the question that is seriously at issue: whether the methodological assumptions of computational psychology are coherent. (Fodor 1975: 78-79)

My contention, by contrast, is that it does indeed matter a great deal whether words like 'representation' are used in their usual sense within CTM, because I believe that the conventionality of linguistic symbols is not something that can be divorced from their symbolhood. It is not that we find a nonconventional property called "meaning" in linguistic symbols that additionally happens to be conventional in nature. Rather, the very notion of "meaning" that we apply to symbols is interwoven with conventionality through and through. And thus if we apply these familiar notions of "representation," "meaning," and "syntax" to CTM, we are led to circularity and regress. As Fodor says, this criticism does not undercut the methodological assumptions of computational psychology . But this is only so because computational psychology (that is, empirical science inspired by the computer paradigm) is not committed to the view that its "mental representations" literally are symbols in a language, as I shall argue in later chapters. A merely analogous usage of the word 'representation' is just fine for computational psychology. The plausibility of CTM's philosophical claims, by contrast, would seem to turn precisely upon the assumption that the "symbols" in question are "symbols" in precisely the same sense that we speak of "symbols" in a language. Formalization and computation show us how to tie meaning to causation for (convention-based) linguistic symbols, and not for anything else. If mental representations are something other than linguistic symbols, we need to see how the link from meaning to causation works for some new class of entities. The arguments Fodor and others give for their claims about intentionality and the vindication of intentional psychology simply do not go through as stated if words like 'representation' and 'symbol' are used in a merely analogous or metaphorical manner.

On the other hand, it is clear that one might try to develop CTM in a way that divorces the technical notion of mental representation from convention-based linguistic signs. The fact that CTM's advocates do not try to do this in any explicit detail does not mean that this avenue might not prove more fruitful in the end. One might, for example, say that the "semantic properties" of mental representations are not the same sort of "semantic properties" possessed by garden-variety symbols. That is, one might say that expressions such as 'semantic property' are homonymous, and have different senses when applied to garden-variety symbols

and to mental representations, in which case results of conceptual analysis of semantic terminology as applied to discursive symbols cannot be used to create problems for a theory of mental representations.

Unfortunately, to the best of my knowledge, no advocates of CTM explicitly pursue this course. But since it does seem to be the only way of saving the theory from the results of my conceptual analysis, I develop two ways of pursuing this line of thought. The first is to take causal theories of content like that supplied in Fodor (1987) as supplying a causal definition of semantic terminology as applied to mental representations. The second is to treat semantic terminology as applied to mental representations as being theoretical and open-ended in character: that is, to treat terms such as 'meaningful' and 'referential' as applied to mental representations as terms whose meaning we do not presently know but might discover as the result of further investigation. I shall argue in Part III that neither of these strategies seems likely to be able to provide an account of intentionality or to vindicate intentional psychology.

These problems for CTM as a philosophical thesis, however, do not entail that the computer paradigm is of no use for the philosopher or the empirical researcher. For I think that there is a much better way to understand the nature and importance of the computer paradigm for the study of cognition. If one adopts this alternative view, the importance of providing an account of intentionality wanes significantly, while the need to justify intentional psychology disappears altogether. To arrive at this standpoint, however, we must cease looking to CTM as a source of solutions to old philosophical puzzles and begin to look at computational psychology as a research programme in psychology from the perspective of historians and philosophers of science.

The basis of the alternative approach is the premise that two of the traditional distinguishing marks of a mature science have been the mathematization of its explanations and the clarification of connections between the domain and the laws of that science and those of other areas of knowledge. So, to take a paradigm example, chemistry progressed towards mathematical maturity through the development of the periodic table, the notion of valences, and the discovery of rules governing reactions between different classes of molecules. It progressed towards connective maturity as the explanations given in chemical terms were able to provide explanations for phenomena described at a higher level (e.g., as described in the vocabularies of metallurgy or genetics) and as categories such as valence were in turn explained at a lower level in terms of such ideas as elementary particles and orbitals.

The interpretation of the importance of the computer paradigm that I wish to urge is the following: what the notion of computation may be able to provide for the empirical scientist is the right kind of technical machinery for the mathematization of the study of cognition—particularly, cognitive psychology. (I emphasize the word 'may' because I seek only to illuminate what computational psychology would provide if successful, and not to make any predictions about its eventual successes or failures.) That is, what computer science gives us is an abstract vocabulary that might turn out to provide the resources for psychology to progress towards mathematical maturity. I think that it should be clear that this is of enormous interest, even without the philosophical benefits claimed for CTM. Surely a large part of what psychology is about is providing an inventory of cognitive processes, "mapping" the relations between these, "unlocking the black boxes" underlying high-level processes by specifying lower-level processes that would account for them, and showing how mental processes are connected to behavior. That is, part of what psychology is about is specifying the form of the mind by tracing out the functional relations mental states bear to one another and to behavior. Many researchers interested in cognition have staked their careers upon their belief that computational notions allow them to carry out this project in ways that were previously unavailable. Indeed, the strength of this belief is evidenced by the emergence of "cognitive science" as an approach to the mind that is organized around the premise that cognitive processes can be described in computational terms. I think this research project is of great interest regardless of whether the notion of computation can contribute to the solution of any philosophical problems as well.

Moreover, viewed in this way, cognitive science as an empirical research programme is not imperiled by my criticisms of CTM. What I argue against CTM is that if you take it as central to the very notion of computation that computation consists in the manipulation of meaningful symbols, then there are serious problems involved in saying that cognition is computation. If, on the other hand, what is essential to the notion of computation is functional specifiability —in, say, the form of a machine table—these problems do not arise. If cognitive science is oriented towards the thesis that cognitive processes are functionally specifiable, then it can attempt to apply the technical resources of computer science to the domain of psychology without worrying about problems with the notions of symbol or representation . Indeed, one might even propose theories that depend upon the premise that there are men-

tal states or brain states that play a role in thought, a role that is formally analogous to the roles played by symbols in the execution of particular computer programs, without threat of incoherence from misuse of such words as 'symbol' and 'meaningful'. (One might, with some risk, even use the word 'symbol' in describing such states, so long as one was careful that the illegitimate importation of the ordinary meaning of the word 'symbol' did not do any illicit work in one's explanations.)

Of course, what one loses in this alternative is the hope CTM excited of finding a level of explanation—a domain of meaningful mental representations over which mental computations are defined—at which there is a clear meeting of the ways between mentalistic description cast in the intentional idiom and one of the natural sciences. On this view, cognitive science does not "close the gap" between mind and nature. Here, however, there is a parting of the ways between the interests of the philosopher of mind and those of the empirical scientist. For the computer paradigm might help psychology progress to one or both types of scientific maturity without providing a philosophical account of intentionality in the process. First, it might provide the tools for the mathematization of psychology without providing for connective maturity as well. But it does seem likely that a good mathematization of cognitive explanation is just the sort of thing that would be helpful in correlating states specified in the intentional idiom with states specified in neurological terms: that is, it is arguable that the only way of finding out how cognitive states are instantiated is to find out what in the brain has the right (functional) "shape" to realize them. So if connection between cognitive explanation and other kinds of explanation is to take place, it may partially be through the mathematization of both levels of explanation. And for this it is a plausible hypothesis that computer science provides the appropriate resources.

But it is important to see that one might get the kind of connectivity that the scientist desires without thereby solving any philosophical problems. The researcher committed to intentional explanation and natural explanation wants to find out what neural processes are specially associated with what intentionally specified processes. And she is interested in this association just to the extent that intentional and naturalistic predictions will track one another. The metaphysical nature of this "special association" really does not matter as far as empirical science is concerned. Empirical science is largely blind to the differences between relationships stronger than empirical adequacy, and hence a good integrated psychological theory could be equally compatible with material-

ism and supervenience or with thoroughgoing parallelist dualism. And this, I think, should be viewed as a virtue rather than a vice: in my book, consistency with a wide range of ontological options counts as a significant virtue for an empirical research programme.

Now there is a sense in which such an integrated psychology would provide an "account" of intentionality and a sense in which it would not. If by "an account of intentionality" one means (a) a model of the relations between intentional states, stimuli, and behavior, and (b) a specification of the natural systems through which intentional states and processes are—to use an intentionally neutral term—realized, then an integrated psychology might well involve an "account of intentionality." But if "an account of intentionality" means something stronger —say, if it involves providing natural conditions upon which intentional properties would have to supervene, then an integrated psychology might well not provide this kind of "account of intentionality." I believe, however, that it is fundamentally misguided to seek such a naturalistic account of intentionality, for reasons that I shall develop in chapters 9 and 11. If I am right, then inability to provide an account of intentionality in this strong sense is not a fault.

I shall argue for a similar attitude towards the other goal CTM has sought to achieve: that of vindicating intentional psychology. To put it very briefly, I do not believe that intentional psychology is presently in need of vindication. The perceived need for a vindication turned upon some concerns about methodology and ontology that came to prominence in the writings of behaviorists and reductionists. One might do well to ask whether these concerns ought to have survived the theories that brought them to prominence. But even if one finds these concerns to be serious ones, they must at very least be put off for the present. By just about everyone's reckoning, any full-scale meeting of the ways that might take place between intentional explanation and neuroscience (much less physics) is a long ways away and depends upon a great deal of research, much of which almost has to be pursued through top-down strategies in cognitive psychology. So, in a sense, any real assessment of cognitivism's compatibility with the generality of physical explanation could only take place once we had a reasonably successful predictive cognitive psychology.

Of course we all look for occasions when our top-down strategies get us to a level where we can find some plausible candidate for a known neurological mechanism that has the right functional features to support the kind of cognitive process we have postulated. Such moments are land-

marks that provide some of the best kinds of reasons to believe one's research is on the right track. That being said, it is nonetheless the case that (1) intentional explanation seems an indispensable starting point for cognitive psychology, regardless of whether such research would ultimately allow us to "throw away the ladder"; (2) it could, in principle, turn out that research in cognitive science could produce a good predictive psychology without ever hooking up with neuroscience in a comprehensive fashion (e.g., we would not throw out psychophysics if we could not produce neutral models to account for the data); (3) if this were to happen, it is not at all clear that we should, as a result, regard such psychological theories as flawed, much less metaphysically perverse; and (4) in the meantime, it is absolutely pointless to expect empirical researchers to care about whether their work meets such ideological tests as conformity with one's favorite ontological theory.

In short, I do not think that intentional psychology is in need of vindication at the present time. The pressing question for the philosophy of psychology is whether intentional explanation can be systematized and mapped out using something like the techniques afforded us by the notion of computation or by some alternative notions, and whether in the course of this project our ordinary mentalistic notions like "desire," "belief," and "judgment" will be retained, built into a larger framework, transformed, or abandoned altogether. Certain outcomes of this project might call for the reassessment of intentional psychology. (And of course there are already those who believe that it is a mistake to view it as an explanatory science in the first place.) There is a separate, and largely empirical question about how cognitive states are realized through specific physiological structures. The connections between the success or failure of this project and the status of intentional psychology are far more tenuous, but really need not be fretted over at this stage of the game.

To repeat, on my reading of the significance of the computer paradigm, what it offers is a project that might hasten the progress of psychology towards scientific maturity by providing the right technical resources for mathematizing the functional relationships that mental states bear to one another and to behaviors. Interpreted in this fashion, the computational approach to cognition is one that is distinguished principally by the conceptual tools it borrows from computer science. However, the computational approach is only one research programme among several that seek to provide the right formal tools for studying cognition. It is a research programme that has rivals that supply different tools for the mathematization of psychology. Notable among these are the information-

theoretic approach favored by Sayre and the network-based mathematical models of various sorts offered over the past thirty years by Pitts and McCulloch, Grossberg, Anderson, and others. What formal techniques end up providing the best descriptions is a question to be answered by the fertility of these research projects.

What this book calls for, then, is a separation of two kinds of issues. The first set of issues involves questions about how to compare competing theories about the mind that emerge out of empirical science. For example, apart from their abilities to give some description of the phenomena in their own canonical vocabularies, just where do two approaches to cognition, such as CTM and connectionism, really differ? What are the "good-making" qualities that are relevant to the assessment of empirical theories in psychology, and which are possessed in greater abundance by whose theories? The second set of issues is made up of more purely philosophical questions about the mind-body problem, the exact metaphysical relationship between mental states and the physical states through which they are realized, and attempts to give a logically necessary and sufficient account of notions such as meaning and intentionality. It is the thesis of this book that, contrary to popular rumor, CTM does nothing to solve the latter problems. Nonetheless, it is quite possible to "bowdlerize" CTM in a fashion that avoids the problems of interpretive regress and to construe it as a special version of machine functionalism; and interpreted in this fashion, computational psychology can be seen as an interesting contender with respect to the first set of issues. With CTM's claims to solving philosophical problems out of the way, however, there is now a level playing field, and computational psychology may be compared with its competitors in terms of their purely scientific merits. And in a roundabout way, I think this counts as progress.

A Brief Guide to This Book

This book is divided into four sections. Part I, comprising chapters 1 through 3, gives an exposition of CTM and of its claims to solve important philosophical problems. It also provides an initial statement of some potential problems for CTM arising from criticisms raised by Searle and Sayre. I make a case that their criticisms are not definitive, and call for a more careful analysis of the notions of "symbol," "syntax," and "symbolic meaning." This analysis is provided in Part II. Chapter 4 presents a conventionalist analysis of symbols, syntax, and symbolic mean-

ing, which is then applied to symbols in computers in chapter 5 and defended against some likely objections in chapter 6. (The reader who comes out of chapter 4 with burning objections will lose nothing by reading chapter 6 before chapter 5.)

The results of this analysis are then applied in Part III (chapters 7 through 9) in a critique of the philosophical claims of CTM. Chapter 7 argues that, if you interpret CTM's talk of "symbols," "syntax," and "semantics" in the ordinary convention-laden way, you are left with an account that is circular and regressive. Chapter 8 argues that CTM fares no better if the semiotic vocabulary is reconstructed in a nonconventionalist way. In short, CTM maintains an illusion of explaining intentionality only by slipping back and forth between semiotic notions based on conventional symbols and talk of an alternative "pure semantics." Chapter 9 briefly makes the case that CTM is unlikely to be supplemented by an independent naturalization of content: some features of the mind do not seem susceptible to naturalization at all, while others seem likely to be naturalized (if at all) only in a fashion incompatible with the constraints laid down by CTM. The dialectical situation at the end of Part III is that CTM's claims to producing distinctively philosophical fruit have been undermined.

Part IV then presents an alternative view of the importance of the computer paradigm. Chapter 10 outlines how computation might provide psychology with important good-making qualities without naturalizing intentionality or vindicating intentional psychology. The book concludes, in chapter 11, with a philosophical examination of the assumptions that intentionality needs naturalizing and mental states need vindicating. I argue that, in the absence of strong a prioristic arguments for naturalism, we are better off letting the special sciences flourish as best they may and shaping our metatheoretic views about intertheoretic connections on the basis of the shape that real psychology takes rather than upon any preconceived notions of what it should look like.