1 ARGUING FOR ELIMINATIVISM* In B. Keeley (Ed.), Paul Churchland (Cambridge University Press, 2005) I am sure I am not alone in reporting that the greater exposure I have to experimental work in scientific psychology and neuroscience the less value there seems to be in our commonsense psychological framework of belief, desire, and the other propositional attitudes. Commonsense psychological concepts hardly feature at all in cognitive science and cognitive neuroscience. Theorists in these areas either eschew psychological vocabulary altogether or appeal to shadowy neologisms such as “cognize” or “encode”. It is often difficult to see where the points of contact are between the serious scientific study of the mind and the apparent platitudes that philosophers tend to take as their starting-point. And even when one can see where the points of contact are, scientific psychology and commonsense psychology are often in tension with each other. Many areas of scientific psychology place serious pressure on our image of ourselves as rational, consistent agents with stable character traits. Yet this image of ourselves is at the heart of commonsense psychology. Commonsense psychology tells one story about the “springs of action” – about how and why we behave the way we do – while the story (or rather, stories) told by scientific psychology and cognitive neuroscience seem completely different and in many ways incompatible with a commonsense understanding of human behavior. In the face of all this some philosophers, most prominently of course Paul Churchland, have argued for a wholesale rejection of our commonsense ways of thinking about the mind. Very few philosophers have found Churchland’s eliminative materialism to be a palatable option. Whatever tensions there might be between commonsense psychological thinking and the scientific investigation of cognition and behavior, it is natural to ask whether we can really abandon the * I am grateful to Brian Keeley for very helpful comments on an earlier draft of this paper, and to audiences at the University of Cincinatti and at the 2004 meeting of the Austrian Ludwig Wittgenstein Society at Kirchberg-am-Wechsel. A shortened version of this paper was published in the Schriftenreihe der Wittgenstein Gesellschaft: Experience and Analysis. 2 conceptual framework of commonsense psychology. What would happen if we tried to do without our concepts of belief and desire and the ways of thinking about how and why people behave that those concepts bring with them? This paper considers how best an eliminativist might argue for the radical falsity of commonsense psychology. I will be arguing that Paul Churchland’s “official” arguments for eliminative materialism (in, e.g., Churchland 1981) are unsatisfactory, although much of the paper will be developing themes that are clearly present in Churchland’s writings. My aims are, first, to refocus the debate on what I take to be more interesting and fruitful arguments for eliminativism and, second, to explore how far those arguments might take us. This introductory section explores some general issues about the form that a plausible argument for eliminative materialism should take. In clarifying the issues in this area it will be helpful to begin with an influential attack on eliminative materialism. Paul Boghossian’s ‘The status of content’ argues forcefully that eliminative materialism cannot be coherently formulated. As he brings out, eliminative materialism is best viewed as a species of irrealism. Irrealism with respect to a particular area of discourse is the thesis that there are no real objects or properties corresponding to the kind terms in that discourse. Just as irrealism about evaluative discourse about art or morality holds that there is nothing in the world corresponding to the evaluative predicates used in aesthetics and ethics, eliminative materialism holds that there is nothing in the world corresponding to the psychological kind terms used in commonsense psychology. As Boghossian notes, irrealist theories in general, and hence eliminative materialism in particular, come in two flavors. Irrealism can be developed either as a form of error theory, or as a form of non-factualism. Error-theoretic forms of irrealism hold that, although meaningful and substantive claims are made by the particular discourse in question, those claims are all false. So, for example, an error theory of moral discourse would hold that evaluative statements are no less truth-apt than any other form of statements. It is just that none of them are true. In contrast, nonfactualist versions of irrealism hold that surface grammatical form is misleading and deny that the 3 discourse in question is truth-apt. The non-factualist denies that the apparently assertoric form of the discourse in question should be taken at face value. An example of a non-factualist approach to moral discourse would be the thesis that moral discourse should be understood as expressing particular attitudes to people and/or situations, rather than as making claims about the (moral) properties of those people and/or situations. Plainly, eliminative materialism about the propositional attitudes is an error-theoretic variety of irrealism. Commonsense psychology makes certain claims about the world and, according to the eliminative materialist, those claims are uniformly false. These claims occur primarily in the context of explaining and predicting behavior. They invoke both particular psychological states and generalizations defined over classes of such states. Yet, if eliminative materialism is correct, there are no such states and no true generalizations defined over them. Arguing convincingly for eliminative materialism requires identifying and diagnosing the errors of which commonsense psychology is being convicted. We need to know (a) what the relevant claims are, and (b) why they are supposed to be false. In the remainder of this section and the next section we will focus predominantly on the first question. The rest of the paper will emphasize the second question. As will soon become apparent, however, it is very difficult to keep these two questions distinct. Boghossian explicitly argues that eliminativism has to be construed as an error theory about all forms of content. He objects to all attempts to limit eliminativism to a thesis about the psychological – and in particular he objects to the idea that we can continue to think about language and linguistic behavior in semantic terms while rejecting the conceptual framework of commonsense psychology. In fact, his anti-eliminativist argument depends upon this very broad reading of the scope of eliminative materialism, for what he effectively offers is an argument to the effect that global eliminativism about content is incoherent. Since his argument against global 4 eliminativism is fairly convincing, our first task must be to determine whether eliminative materialism can be formulated without a commitment to global eliminativism. Prima facie, there is no direct entailment from eliminativism about the psychological to eliminativism about the linguistic. It is true that there are models of linguistic meaning and the requirements of communication that effectively bring linguistic behavior and linguistic understanding within the scope of propositional attitude psychology. If one thinks that the meaning of a sentence is given (at least in part) by the communicative intentions of its utterer, and that communication is achieved when the listener manages to work backwards from the heard sentence to the communicative intention that it embodies, then it will seem obvious that the domain of the linguistic stands or falls with the domain of the psychological. But this model of linguistic meaning is far from compulsory. Accounts of meaning in terms of use, for example, offer a natural retreat for the eliminative materialist. Boghossian’s argument in fact hinges on what he takes to be the most powerful arguments in support of eliminative materialism. These are all, he thinks, arguments directed at the notion of content in the abstract, rather than at the content of propositional attitudes. He writes: The real difficulty with the suggestion that one may sustain differential attitudes towards mental and linguistic content stems from the fact that the best arguments for claim that nothing mental possesses content would count as equally good arguments for the claim that nothing linguistic does. For these arguments have nothing much to do with the items being mental and everything to do with their being contentful: they are considerations, of a wholly general character, against the existence of items individuated by content. If successful, then, they should tend to undermine the idea of linguistic content just as much as they threaten its mental counterpart. (Boghossian 1990, 171) The arguments Boghossian discusses are arguments from (i) the indeterminacy of content, (ii) the holistic nature of content, (iii) the irreducibility of content, and (iv) the “queerness” of content. It is quite plausible, I think, that these four arguments are each just as applicable to linguistic content as they are to propositional attitude content. But there is no need to conclude with Boghossian that eliminative materialism entails global eliminativism. We can instead take it as a 5 recommendation to look elsewhere for a plausible argument for eliminative materialism. The eliminative materialist needs an argument that can plausibly be confined to the psychological domain and that will not leave her open to Boghossian’s charge of incoherence. The eliminative materialist’s first task must be to identify a class of putative errors committed by commonsense psychology that can plausibly be confined within the realm of the psychological. The putative error must be more localized than simply a commitment to content-bearing states in general. It must be something deriving either from the particular type of content-bearing states invoked in commonsense psychology or from the use to which they are put (or, of course, both). This paper will explore both possibilities. An eliminativist can argue that there is a fundamental error in the scope of commonsense psychology. The claim here is that we actually use the machinery of commonsense psychology far less frequently than we think we do. Whereas opponents of eliminativism hold that commonsense psychology is an indispensable tool for navigating the social world, the eliminativist can point to many occasions on which we make manage to understand what is going on and engage in coordinated social behavior without bringing into play the conceptual framework of commonsense psychology. Of course, this line of argument can only be part of the overall eliminativist strategy. Showing that commonsense psychology has a far more peripheral role to play in social understanding and social coordination than it is standardly held to have does not give us eliminativism. The eliminativist has to go beyond this to tackle directly the core tenet of commonsense psychology. This core tenet is that propositional attitudes are the “springs of action”. We act the way we do in virtue of our beliefs, desires, hopes, fears, and so on. These states are distinguished by having a particular type of content, namely, propositional content of the type that might be captured by a “that–“ clause. It would be powerful support for the eliminativist thesis if it could be shown that invoking psychological states with propositional content is simply not the right way to think about 6 the springs of action. The eliminativist can try to show that, although we need to appeal to representational states in explaining and predicting behavior, these representational states do not have propositional content and are fundamentally different from states that do have propositional content. The project here would be to establish a fundamental mismatch between two models of representation – the model of representation implicated in just about all ways of thinking about commonsense psychology, on the one hand, and the family of models of representation that seem to provide the best general picture of how the brain can be representational, on the other. This paper is programmatic. Its aim is to explore how eliminativism might best be formulated and defended. Section 1 will clarify the nature of commonsense psychology, and hence clarify the eliminativist thesis. This will allow us to see why Paul Churchland’s standard arguments for eliminativism are unlikely to provide the desired conclusion. It seems to me that the most promising strategy for the eliminative materialist has two components. The first part of the strategy is to put pressure on what I have elsewhere called the broad construal of the scope of commonsense psychology (Bermúdez 2003, 2005). In most general terms “commonsense psychology” simply picks out the complex of skills and capacities that collectively make possible social understanding and social coordination. It is a substantial thesis that these skills and abilities all invoke the conceptual framework of commonsense psychology. The first step for the eliminative materialist is to block this move by arguing that commonsense psychology has a far more peripheral role to play in social understanding and social coordination than it is standardly held to have. This line of argument is discussed in section 2. But narrowing the scope of commonsense psychology does not give us eliminativism. The second part of the eliminativist strategy tries to establish a fundamental mismatch between two models of representation – the model of representation implicated in just about all ways of thinking about commonsense psychology, on the one hand, and the family of models of representation that seem to provide the best general picture of how the brain can be 7 representational, on the other. Sections 3 and 4 discuss this second component of the eliminativist strategy. 1 Commonsense psychology and eliminative materialism In its most general sense the term “commonsense psychology” picks out the complex of social abilities and skills possessed by all normal, encultured, non-autistic and non-brain-damaged human beings. These are the skills and abilities that allow us to navigate the social world. Taken in this very general sense, commonsense psychology is an explanandum rather than an explanans. We would expect it to be the sort of thing of which a theoretical account is given, rather than something that can itself do theoretical and explanatory work. The expression “commonsense psychology” is used more determinately to characterize what is in effect a particular conceptual framework deemed to govern our social understanding and social skills, where this conceptual framework can be thought of as an account of what underlies the general abilities and skills just identified. Here is a useful characterization of this second way of thinking about commonsense psychology from the introduction to a recent collection of essays: It has become a standard assumption in philosophy and psychology that normal adult human beings have a rich conceptual repertoire which they deploy to explain, predict and describe the actions of one another and, perhaps, members of closely related species also. As is usual, we shall speak of this rich, conceptual repertoire as ‘folk psychology’ and of its deployment as ‘folk psychological practice’. The conceptual repertoire constituting folk psychology includes, predominantly, the concepts of belief and desire and their kin – intention, hope, fear, and the rest – the so-called propositional attitudes. (Davies and Stone 1995, p.2) In very general terms, then, our skills in social understanding and social coordination are underpinned by the conceptual framework of propositional attitude psychology. We can make 8 sense of other people and coordinate our behavior with theirs in virtue of our ability to apply the concepts of belief, desire, and so forth. This general characterization leaves unanswered questions about how the conceptual framework of propositional attitude psychology is applied in practice. This brings us to a third way of thinking about commonsense psychology. This is where we find the much-discussed distinction between theory theorists and simulation theorists. A number of influential theorists accept the view that social understanding and social coordination rest upon an implicitly known, and essentially theory-like, body of generalizations connecting propositional attitude states with overt behavior and with each other. Paul Churchland is of course one of these, as are David Lewis, Frank Jackson, and Jerry Fodor. On this view (the so-called theory theory), social understanding involves subsuming observed behavior and what is known of a person’s mental states under these generalizations in order to understand why they are behaving in a certain way and how they will behave in the future. In recent years the theory theory has been challenged within both philosophy and psychology by theorists promoting the simulationist approach to commonsense psychology.1 Simulationists hold that we explain and predict the behavior of other agents by projecting ourselves into the situation of the person whose behavior is to be explained/predicted and then using our own mind as a model of theirs. Suppose that we have a reasonable sense of the beliefs and desires that it would be appropriate to attribute to someone else in a particular situation, so that we understand both how they view the situation and what they want to achieve in it. And suppose that we want to find out how they will behave. Instead of using generalizations about how mental states typically feed into behavior to predict how that person will behave, the simulationist thinks that we use our own decision-making processes to run a simulation of what would happen if we ourselves had 1 Gordon 1986 and Heal 1986 are key statements of the simulationist position. The principal readings in the debate between theory theorists and simulation theorists are collected in Davies and Stone 1995a and 1995b. Further essays will be found in Carruthers and Smith 1996. This collection includes interesting material from developmental psychologists and students of primate cognition. Currie and Ravenscroft 2002 develops a theory of imagination in the context of a simulationist approach to social understanding. 9 those beliefs and desires. We do this by running our decision-making processes off-line, so that instead of generating an action directly they generate a description of an action or an intention to act in a certain way. We then use this description to predict the behavior of the person in question. We can, therefore, distinguish three different ways of construing commonsense psychology, as in the following table. 1 2 3 The complex of skills and abilities that underlie our capacities for social understanding and social coordination A particular conceptual framework for explaining social understanding and social coordination in which the propositional attitudes are central. A particular account of how the conceptual framework in (2) is applied in the service of explanation/prediction. It is plain that these different ways of construing commonsense psychology yield different ways of construing eliminative materialism. There is no prospect of an eliminative materialism defined in terms of the first way of construing commonsense psychology, since this would involve denying the existence of successful social interaction and coordination. However, eliminative materialism looks very different depending on whether it is understood against the background of the second or third construals of commonsense psychology. The most wide-ranging form of eliminativism is directed at the simple idea that we can only make sense of other people’s behavior, and coordinate our own behavior with theirs, through the interpretive framework of propositional attitude psychology. This form of eliminativism takes issue with the very idea that our canonical way of getting purchase on the explanation, prediction, and coordination of behavior is through the machinery of the propositional attitudes. It is completely neutral on the issues that divide simulationists and theory theorists. Any arguments that can show that social coordination and social understanding do not rest upon the machinery of propositional attitude psychology will be effective against both theory theorists and simulationists. 10 One striking feature of the standard version of eliminative materialism is that, although the position is clearly targeted against commonsense psychology in its second construal (its intended target is the legitimacy of any use of content-bearing propositional attitudes in the enterprise of explaining, predicting, and coordinating behavior), the arguments standardly put forward make most sense in the context of the third construal. The enterprise of showing that commonsense psychology does not count as a legitimate and productive scientific theory clearly presupposes the truth of some version of the theory theory. Of the three canonical arguments for eliminativism, two clearly fall into this category. Consider the following: The folk psychology of the Greeks is essentially the folk psychology we use today, and we are negligibly better at explaining human behavior in its terms than was Sophocles. This is a very long period of stagnation and infertility for any theory to display, especially when faced with such an enormous backlog of anomalies and mysteries in its own explanatory domain. (Churchland 1981, p.124) The idea that commonsense psychology is a theory is both explicit in the argument and essential to it. The same holds for the argument from the irreducibility of commonsense psychology: If we approach Homo sapiens from the perspective of natural history and the physical sciences, we can tell a coherent story of his constitution, development and behavioral capacities which encompasses particle physics, atomic and molecular theory, organic chemistry, evolutionary theory, biology, physiology and materialistic neuroscience. . . But folk psychology is no part of this growing synthesis. (idem.) Only if commonsense psychology is a theory can it even be a candidate for reduction. It might seem that an easy way to respond to the concerns just raised would be to argue on independent grounds for the inadequacy of the simulationist alternative to the theory theory. This in fact is what Paul Churchland has done. His 1989 paper “Folk psychology and the explanation of human behavior” contains a sustained argument against what he calls the ‘anti-theoretical view of our self understanding’. Surely, if simulationism is not a serious contender, then arguments against the theoretical construal of commonsense psychology will be sufficient to establish the bankruptcy of commonsense psychology tout court. This would be a mistake. There is a very significant difference between a successful local argument against a particular way of applying the conceptual framework of propositional attitude 11 psychology and a successful global argument against propositional attitude psychology as such. What needs to be shown for the global conclusion to emerge is that no possible version of the theory theory could resolve the problem that the three local arguments identify. And nothing like this follows from the three arguments we have briefly considered. If they are telling at all (and of course this has been fiercely contested) they are telling only against our current commonsense psychological theories. To take just one example, if our commonsense psychology were to change then it would no longer be the same commonsense psychology enjoyed by the ancient Greeks, and the first argument would be blunted. But nothing has been said to rule out the possibility of a significantly developed commonsense psychology that continues to operate within the parameters of propositional attitude psychology. Nor is it clear how the sort of considerations that Churchland brings into play could possibly show this. What would be needed would be a principled argument to show that certain very general features of propositional attitudes (features general enough to apply to any possible version of propositional attitude psychology, and hence to any possible descendant of our current commonsense psychology) rule them out of the project of explaining and predicting human behavior. The type of arguments that Churchland brings into play in support of the standard version of eliminative materialism just seem to be the wrong type of arguments. On the other hand, however, the resources for such an argument can be found elsewhere in Churchland’s writings. There is an illuminating passage in “Eliminative materialism and the propositional attitudes”. Churchland is (implicitly) criticizing the idea that we can think of what a person believes in terms of the sentences to which that person would assent, so that we can use those sentences to characterize the content of what they believe. This is, of course, simply a special case of the general principle that propositional attitudes have contents that can be specified by means of “that –“ clauses, where the complement of a “that –“ clause is a declarative sentence. A declarative sentence to which a speaker would give confident assent is merely a one-dimensional projection – through the compound lense of Wernicke’s and Broca’s areas onto the idiosyncratic surface of the speaker’s language – of a four- or five-dimensional “solid” that is an element in his true kinematical state. . . Being projections of that inner reality, such sentences do carry significant information regarding it and are thus fit to function as elements in a communication system. On 12 the other hand, being sub-dimensional projections, they reflect but a narrow part of the reality projected. They are therefore unfit to represent the deeper reality in all its kinematically, dynamically, and even normatively relevant respects. (Churchland 1981, p.129) The motivation for eliminativism suggested here is not that it is an impoverished theory, nor that it cannot be reduced to neuroscience, nor that it is limited in its explanatory scope. Rather, we should be eliminativists because commonsense psychology rests upon an untenable model of representation. Before going on to consider that line of argument in sections 3 and 4, however, we should note that the distinction between different ways of thinking about commonsense psychology provides a way of clarifying the first line of argument discussed earlier. It is effectively a matter of definition that we navigate the social world in virtue of commonsense psychology in the first of the three identified senses. But it by no means follows that our skills and abilities in social understanding and social coordination are underwritten by commonsense psychology in the second sense. The eliminativist can argue that many of our social skills and abilities have nothing to do with propositional attitude psychology. This line of argument will be explored in the next section. 2 Narrowing the scope of commonsense psychology2 No eliminativist denies that we are capable of accommodating ourselves to other people’s behavior and of engaging in coordinated social action. The eliminativist’s question is how these forms of social accommodation and coordination are achieved. To what extent, if any, are the propositional attitudes involved? One obvious way to broach this general question is to ask whether success in social exchanges and social interactions rests upon attributions of propositional attitudes. Do we navigate the social world by attributing beliefs, desires, hopes, and fears to the people whom we encounter in order to explain and predict their behavior? Or do we employ more 2 This section incorporates material from Bermúdez 2003 and forthcoming. 13 basic mechanisms that allow us to interact successfully with others without using the machinery of propositional attitude psychology? It will be helpful to break this question down into two further questions. (1) Does successful social behavior always require explaining and/or predicting the behavior of other participants? (2) In those cases where social behavior does depend upon explaining and predicting the behavior of others, do such explanations and predictions have to involve propositional attitude psychology? In this section I will briefly survey some reasons for answering these two questions in the negative – by giving examples of (a) social interactions that seem to proceed without any form of explanation or prediction, and (b) processes of explanation and prediction that seem to proceed without involving attributions of propositional attitudes. Let me begin with the first question. Are there any social interactions that can be modeled without assuming that the parties involved are engaged in explaining or predicting each other’s behavior? Surprisingly, game theory has some interesting implications in this area. Game theorists have long been interested in a particular class of strategic interaction where the dominant strategy for each player leads inevitably to an outcome in which each player is worse off than he could otherwise have been. A dominant strategy is one that is more advantageous than the other possible strategies, irrespective of what the other players do (it dominates those other strategies). The classic example of this kind of strategic interaction is the so-called prisoners’ dilemma, which has the following pay-off table. Player B BETRAY KEEP SILENT BETRAY 5, 5 0, 10 KEEP SILENT 10, 0 2, 2 Player A 14 Each entry represents the outcome of a different combination of strategies on the part of prisoner A and B. The bottom left-hand entry represents the outcome if prisoner A keeps silent at the same time as being implicated by prisoner B. The outcomes are given in terms of the number of years in prison that will ensue for prisoners A and B respectively. So, the outcome in the bottom left-hand box is 10 years for prisoner A and none for prisoner B. Imagine looking at the pay-off table from Prisoner A’s point of view. You might reason as follows. Prisoner B can do one of two things – betray me or keep quiet. Suppose he betrays me. Then I have a choice between five years in prison if I also betray him – or ten years if I keep silent. So, my best strategy if he betrays me is to betray him. But what if he keeps silent? Then I’ve got a choice between two years if I keep quiet as well – or going free if I betray him. So, my best strategy if he keeps quiet is to betray him. Whatever he does, therefore, I’m better off betraying him. Unfortunately, prisoner B is no less rational than prisoner A and things look exactly the same from her point of view. In each case the dominant strategy is to defect. So, the two prisoners will end up betraying each other and spending five years each in prison, even though they both would have been better off keeping silent and spending two years each in prison. The psychology of the prisoners’ dilemma is not very complicated. There is no question of either player having to predict what the other player is doing – if betrayal is the dominant strategy then one should follow it whatever one thinks that the other person is going to do, which just means that there is no need to think about what they are going to do. One can, in effect, read off what to do from the pay-off table. Things get more complicated when we come to social interactions that have the same logic as the prisoner’s dilemma but are iterated. When it is not known how many plays there will be and/or the rationality of the other participant is not known scope opens up for cooperative play. This is where we rejoin the question of the domain of commonsense psychology. Suppose that we find ourselves, as we frequently do, in social situations that have the structure of an indefinitely repeated prisoner's dilemma. How do we negotiate the situation? One answer might be that I make a complex set of predictions about what the other player (or players) will do, based on my 15 assessment of their preference orderings and their beliefs about the probability of each of us betraying as opposed to cooperating, and then factor in my own beliefs about how what will happen in future depends upon whether or not I cooperate - and so on. This, of course, would be an application of the general explanatory framework of commonsense psychology (on the simplification that utilities and probability assignments are regimentations of desires and beliefs).3 But even if we can make sense of the idea that strategic interaction involves these kinds of complicated multi-layered predictions involving expectations about the expectations that other people are expected to have, one might wonder whether there is a simpler way of determining how to behave in that sort of situation. And in fact game theorists have directed considerable attention to the idea that social interactions taking the form of indefinitely repeated prisoner's dilemmas might best be modeled through simple heuristic strategies in which, to put it crudely, one bases one's plays not on how one expects others to behave but rather on how they have behaved in the past. The best known of these heuristic strategies is TIT-FOR-TAT, which is composed of the following two rules: A. Always cooperate in the first round B. In any subsequent round do what your opponent did in the previous round The TIT-FOR-TAT strategy is very simple to apply, and does not involve any complicated folk psychological attributions or explanations/predictions. All it requires is an understanding of the two basic options available to each player, and an ability to recognize which one of those strategies has been applied by other players in a given case. The very simplicity of the strategy explains why theorists have found it such a potentially powerful explanatory tool in explaining such phenomena as the evolutionary emergence of altruistic behavior (see Axelrod 1984 for an accessible introduction and Maynard Smith 1982 and Skryms 1996 for more detailed discussion).4 3 For more on this simplifying assumption see Lewis 1994 and Pettit 1991. TIT-FOR-TAT has only a limited applicability to practical decision-making. In a situation in which two players are each playing TIT-FOR-TAT, a single defection will rule out the possibility of any further cooperation. This is clearly undesirable, particularly given the possibility in any moderately complicated social interaction that what appears to be a defection is not really a defection (suppose, for example, that my colleague misses the examination meeting because her car broke down). So any plausible version of the 4 16 Strategies such as TIT-FOR-TAT plainly do not involve any exploitation of the categories of folk psychology. In fact, they do not involve any processes of explanation or prediction at all. In order to apply TIT-FOR-TAT, or some descendant thereof, I need only work out whether the behavior of another player should best be characterized as a cooperation or a defection - and indeed to work out which previous behaviors are relevant to the ongoing situation. This will often be achievable without going into the details of why that player behaved as they did. Of course, sometimes it will be necessary to explore issues of motivation before an action can be characterized as a defection or a cooperation - and sometimes it will be very important to do this, given that identifying an action as a defection is no light matter. But much of the time one might well get by perfectly well without going at all deeply into why another agent behaved as they did. The TIT FOR TAT heuristic is an interesting example of how one might navigate the social world without explaining or predicting the behavior of others. But suppose that we are dealing with a social situation where some form of explanation and/or prediction of the behavior of other participants is required. Is this a situation that we can only navigate by using the conceptual framework of commonsense psychology? Not necessarily. Consider everyday social interactions, such as buying food in a shop or ordering in a restaurant. These are coordination problems that can only be successfully negotiated because one has certain beliefs about why people are doing what they are doing and about how they will continue to behave. But there is no need for these beliefs to be second-order beliefs about the psychological states of other participants in the interaction. In such routine situations all that is required is to identify the person approaching the table as a waiter, or the person standing behind the counter as a butcher. Simply identifying social roles provides enough leverage on the situation to allow one to predict the behavior of other participants and to understand why they are behaving as they are. There is no need to think about what the TIT-FOR-TAT strategy will have to build in some mechanisms for following apparent defections with cooperation, in order both to identify where external factors have influenced the situation and to allow players the possibility of building bridges back towards cooperation even after genuine defection. One possibility would be TIT-FOR-TWO-TATS, which effectively instructs one to cooperate except in the face of two consecutive defections. 17 waiter might desire or the butcher believe – any more than they need to think about what I believe or desire. The social interaction takes care of itself once the social roles have been identified (and I've decided what I want to eat). One lesson to be drawn from highly stereotypical social interactions such as these is that explanation and prediction need not require the attribution of propositional attitudes. Identifying someone as a waiter is a matter of understanding him as a person who typically behaves in certain ways within a network of social practices that typically unfold in certain ways. This is a case where our understanding of individuals and their behavior is parasitic on our understanding of the social practices in which their behavior takes place. Nor, of course, is this understanding of social practices a matter of mastery of a primitive theory. We learn through experience that certain social cues are correlated with certain behavior patterns on the part of others and certain expectations from those same individuals as to how we ourselves should behave. Sometimes we have these correlations pointed out to us explicitly -- more often we pick them up by monitoring the reactions of others when we fail to conform properly to the "script" for the situation. These programmatic remarks provide some support for eliminative materialism. Commonsense psychology, understood as a conceptual framework for making sense of behavior in terms of propositional attitude psychology, is based on the overarching assumption that people act in virtue of their beliefs and desires – and hence that we can only make sense of other people’s behavior by interpreting it within the web of propositional attitude psychology. We have been considering examples of social interactions that (if my interpretation is correct) it would be a fundamental mistake to view in these terms. To the extent, then, that these social interactions are typical and widespread, rather than isolated outliers, we have a clear sense in which commonsense psychology could turn out to be false in a significant number of cases. 18 3 Neural representation and eliminative materialism Let us turn now to the argument for eliminativism from neural representation. This is an argument that can be found in Churchland’s writings, although it is not foregrounded as the “standard” arguments briefly considered earlier. Cognition is, Churchland thinks, a form of informationprocessing and the representations over which that processing takes place are distributed in something like the way that representations are distributed over a large number of units and weights in an artificial neural network. Here is a very general presentation of the approach: The basic idea is that the brain represents the world by means of very highdimensional activation vectors, that is, by a pattern of activation levels across a very large population of neurons. And the brain performs computations on those representations by effecting various complex vector-to-vector transformations from one neural population to another. This happens when an activation vector from one neural population is projected through a large matrix of synaptic connections to produce a new activation vector across a second population of nonlinear neurons. Mathematically, the process is a process of multiplying a vector by a matrix and pushing the result through a nonlinear filter. (Churchland 1992, reprinted in Churchland & Churchland 1998, p. 41) Churchland is betting, in effect, that a complete account of neural computation will be defined over complex patterns of activation across large populations of neurons. This means that neural representations will have a huge number of degrees of freedom. There will be as many dimensions of variation in a neural representation as there are neurons whose activation values can vary independently. In mathematical terms we need to consider neural representations as n-place vectors where n is the number of neurons. Philosophical discussion of the ramifications of the distributed nature of neural representations have tended to focus on the issue of whether mental representations are structured (see, for example, the papers collected in Macdonald and Macdonald 1995). The parameters of the debate should be familiar. Proponents of the language of thought hypothesis argue that propositional attitudes cannot be causally efficacious unless the vehicles of those propositional attitudes are isomorphic to the structure of their contents (so that the vehicle of the belief that Paris is in France has distinguishable components corresponding to the separable and recombinable components of 19 the content of the belief). It is noted that there seems little prospect of capturing such structure if representations are distributed in anything like the way they are in artificial neural networks. Then the conclusion is drawn either that artificial neural networks cannot be a good guide to the structure of mental representation or that propositional attitude psychology is seriously cast in doubt. There is a sense, of course, in which this can be taken as an argument for eliminativism from the (putatively) distributed nature of neural representations. But the argument is, to put it mildly, far from compelling. I would imagine that opponents of eliminativism have far more confidence in the validity of propositional attitude psychology than they have in the principle that causally efficacious propositional attitudes require isomorphically structured vehicles. It is hard to imagine this line of argument having much suasive force. Yet there is a much more direct line of argument from the distributed nature of neural representation to eliminativism. If neural representations are distributed in a way that yields as many degrees of freedom as there are neurons whose activation values can vary independently of each other, then we need to think of each individual neural representation as a location in a multidimensional state space where each dimension is given by the range of possible activation values for each unit. The state space of a neural representation defined over with n independently varying neurons contains n dimensions, and assigning to each unit a particular activation value uniquely determines a single point within that n-dimensional space (a point that could equally be represented by a vector comprising an ordered sequence of those activation values). Paul Churchland is betting that this does indeed turn out to be the case. Here is the possibility he envisages. Suppose that research into the structure and activity of the brain, both fine-grained and global, finally does yield a new kinematics and correlative dynamics for what is now thought of as cognitive activity. The theory is uniform for all terrestrial brains, not just human brains, and it makes suitable conceptual contact with both evolutionary biology and non-equilibrium thermodynamics. It ascribes to us, at any given time, a set or configuration of complex states, which are specified within the theory as figurative ‘solids’ within a four- or five-dimensional phase-space. The laws of the theory govern the interaction, motion and transformation of these ‘solid’ states within that space, and also their relations to whatever sensory and motor transducers the system possesses. (Churchland 1981, p. 129) 20 How will this conceptual framework of solids within multi-dimensional phase space relate to the familiar framework of the propositional attitudes? The argument for eliminativism from the distributed nature of neural representation is simply that there is a fundamental mismatch between the vocabulary of the propositional attitudes and the complex multi-dimensional representational states that our explanations of behavior are trying to capture. We act the way we do because of how our brains represent the world. Yet the complexity of those neural representations far outstrips the linguistic resources with which we are trying to capture them. As Churchland notes in the passage quoted in the first section, there seems to be a fundamental difference between one-dimensional linguistic characterizations of mental states with what he describes as the underlying multidimensional neural reality It is built into our thinking about propositional attitudes that the content of a propositional attitude can be specified in a “that –“ clause where what follows the “that” is a declarative sentence. The contents of propositional attitudes are exhausted by the meanings of the sentences that express them. If I believe that Paris is in France then there can be no more to the content of what I believe than can be expressed through the sentence “Paris is in France”. We can see this general notion of propositional attitude content at play in the familiar debate about the relation between the content of belief and the content of perception. Theorists who argue, surely correctly, that there is a fundamental distinction between the content of belief and the content of perception tend to highlight features of the content of perception that outstrip the concepts and words that we can deploy in specifying how a given perception represents the world. So, for example, the acuity of our color perception allows us to make discriminations that far outstrip our color vocabulary. The same holds for shape perception. We can perceptually represent the world as containing shapes for which we have no concepts or words. This is frequently taken to indicate that the content of perception is fundamentally different from the content of belief – if the content of perception were content of the same type as in beliefs then it would be fully expressible without remainder in a “that –“ clause. 21 We do not have anything like a plausible model of neural representation, but if Churchland’s bet is well-founded and neural representations are distributed in a way that gives them a very high number of degrees of freedom then it seems clear that there will be an even greater lack of fit between our propositional attitude vocabulary and the multi-dimensional representations in virtue of which we behave the way we do than there is between our propositional attitude vocabulary and our perceptual representations. Yet why should this provide an argument for eliminativism about the propositional attitudes? How do we get from the premise that propositional attitude vocabulary does not get things fully right to the conclusion that it gets things fundamentally wrong? What is at issue here, of course, is the precise nature of the lack of fit between our propositional attitude talk and the underlying neural representations. Let us look again at the lack of fit between perceptions and epistemic reports of those perceptions (reports that specify what is sometimes called the concept content of perceptions). If I look out of the window, see my car in the drive and then say that I see that my car is in the drive, it is certainly true that I have failed fully to characterize my perceptual state. There are indefinitely many ways in which things could visually appear to me that it would be correct to describe as my seeing that my car is in the drive. But I have not, of course, said anything false. Not telling the whole story is not the same as telling a story that is wholly wrong – or even, for that matter, partly wrong. The eliminativist is arguing for something much stronger than the claim that commonsense psychology stands to the underlying neural reality in the relation that perceptual reports employing “that –“ clauses stand to the underlying perceptual reality. Clearly, then, if we are to derive an argument for eliminativism from the multi-dimensional nature of neural representation there must be more going on than in the perceptual case. 4 Eliminative materialism and sub-symbolic representation 22 How might such an argument be motivated? Suppose we identify a locus of representational content at a very low level in neural network models. That is to say, suppose that we identify particular units, or small groups of units, as carrying out particular representational functions and tasks. This would be to make something like the “natural assumption” that Churchland sketches out in the following passage. If we are to assign specific semantic or representational contents to collective units of this kind, a natural first assumption is that any unit must in some way inherit its overall content from the individual and more basic representational significance of each of its many constituting elements, namely, the activation level of each of its many neurons. After all, it is these individual neurons that are the recipients of information from the environment: either directly, through their interaction with ambient light, heat and various mechanical and chemical impingements; or indirectly, through their many synaptic connections with neurons earlier in the processing hierarchy. (Churchland 1998, in Churchland and Churchland 1998 p. 83) We can think of these units or groups of units as the representational primitives of the network – the place where we need to start if we are to build up to an account of the representational character of the network as a whole.5 It is natural to think that these representational primitives will be representing what are often called microfeatures of the environment. That is to say, they code features that are much more finely grained than those encoded within the vocabulary that we employ to specify the content of propositional attitudes. These micro-features are, to use the familiar jargon, sub-symbolic. As far as eliminativism is concerned the crucial question is the relation between this subsymbolic level of representation in terms of micro-features and the symbolic level of representation in terms of objects and properties. The eliminativist needs to argue that there is a fundamental mismatch between the two different types of representation, so that what is going on at the sub-symbolic level comes apart from what is going on at the symbolic level. Strikingly, discussions of the relation between sub-symbolic and symbolic levels have tended to try to assimilate the two levels, rather than to drive a wedge between them. Theorists such as 5 In fact, for reasons brought out in Churchland and Sejnowski 1993, the representational primitives are far more likely to be distributed across groups of units than to individual units. See their discussion of local coding vs vector coding on pp. 163 ff. 23 Smolensky, for example, have tried to show how approximations to our symbolic level concepts can be constructed from complexes of micro-features (see, for example, Smolensky 1988). Smolensky’s well-known example of the distributed representation of coffee cup has some heuristic value, but is not very helpful because the micro-features are all located squarely at the symbolic level. Smolensky’s micro-features include hot liquid, burnt odor, finger-sized handle, and so on. Clearly there will no difficulties integrating these micro-features into a symbolic level account of what is going on in the network. For present purposes the following characterization from Fodor and Pylyshyn’s well-known critique of neural network models is more helpful. Fodor and Pylyshyn note that the relation between sub-symbolic and symbolic representations is standardly taken to be analogous to the relation between a defined expression and its feature analysis, and go on to say: Since micro-features are frequently assumed to be derived automatically (i.e. via learning procedures) from the statistical properties of samples of stimuli we can think of them as expressing the sorts of properties that are revealed by multivariate analysis of sets of stimuli (e.g. by multi-dimensional scaling of similarity judgments). In particular, they need not correspond to English words; they can be finer-grained than, or otherwise atypical of, the terms for which a non-specialist needs to have a word. (Fodor and Pylyshyn 1998 in Macdonald and Macdonald 1995). Even though clusters of sub-symbolic micro-features approximate to the relevant symbolic representations, rather than mapping precisely onto them, it is clear that there is a basic continuity between the sub-symbolic and the symbolic level stories. And this is not at all surprising, given that (as Fodor and Pylyshyn note) the sub-symbolic characterization of what is going on in a network is often arrived at by working backwards from the symbolic characterization. This might be by the feature analysis of concepts (as in Smolensky’s coffee example) or by more sophisticated methods of analyzing the relevant activation space, such as the similarity measures proposed by Laakso and Cottrell (1998) or the dendogram analysis of NETtalk offered by Rosenberg and Sejnowski (1987). If there is such a mapping, even if only an approximate mapping, from the sub-symbolic to the symbolic level then there seems no prospect of arguing for an interesting version of eliminativism 24 from the mismatch between the multi-dimensionality of neural representation and the unidimensionality of commonsense psychology. It is striking, therefore, that Churchland himself places considerable emphasis on the existence of such a mapping. In his 1998 paper “Conceptual similarity across sensory and neural diversity: The Fodor-Lepore challenge answered” he uses the Laakso-Cottrell similarity measure to argue that the semantic properties of a point in a network’s activation space should be understood in terms of “stable and objective macrofeatures of the environment” (Churchland 1998, p. 85). If this is how we are to understand the semantic properties of neural networks then there seems little prospect of arguing that the vocabulary and conceptual framework of propositional attitude psychology is fundamentally and in principle unsuitable for characterizing how brains represent the world – which, of course, is what the eliminativist is trying to establish. 5 Evidence for micro-featural representation The eliminativist needs to show that we can explain behavior in terms of a level of neural representation that is incommensurable with the conceptual framework of propositional attitude psychology. This requires showing that the representations serving as the “springs of action” are representations of features of the environment that cannot be assimilated to the features of the environment that are represented in propositional attitude psychology. How might such an argument be constructed? Let me begin with some more general comments. The explanatory power of folk psychology depends upon beliefs, desires, and other propositional attitudes being the “springs of action”. This way of thinking about the springs of action brings with it a particular interpretation of the architecture of cognition – specifically, a sharp distinction between “central” cognitive processes that involve propositional attitudes and “modular” cognitive processes that are not defined over propositional attitudes but instead provide inputs to the propositional attitude system. These 25 modular processes have certain characteristics (such as informational encapsulation, domainspecificity, speed, and so on) that make it natural to classify them as subpersonal, in opposition to the personal-level propositional attitude system, which has none of these characteristics. We have seen a number of ways of putting pressure on this way of thinking about the architecture of cognition. We looked briefly at ways of making sense of the behavior of others that do not involve the attribution of propositional attitudes and hence that do not involve the explanatory framework of commonsense psychology. Much of our understanding of other people rests upon a range of relatively simple mechanisms and heuristics that allow us to identify patterns in other people’s behavior and to respond appropriately to the patterns detected. The simplest such patterns are a function of mood and emotional state, while the more complex ones involve social roles and routine social interactions. Of course, our ways of explaining behavior are not invariably a good guide to how that behavior came about, and so we need to look further. In the remainder of this section I will point towards some experimental findings and research programs that might be deployed in support of the eliminativist project. I will focus on four different areas, two low-level and two high-level. Two visual pathways Psychologists and neuroscientists are agreed that there are (at least) two different ways of thinking about how our perceptions of the environment feed into action. There is considerable controversy about precisely how we are to understand both the function and the neuroanatomy of these two different pathways, but a considerable consensus that some sort of distinction needs to be made between “vision for action’ and “vision for identification’.6 Neuropsychological dissociations are an important source of evidence. Researchers have reported a double dissociation between the 6 See the essays in section I of Prinz and Hommel 2002 for up-to-date surveys of current thinking in this area. The tutorial by Rossetti and Pisella is particularly helpful. 26 capacity to act upon objects and the capacity to name them. Patients with optic ataxia are able to identify objects but are severely impaired in tasks that involve reaching objects or working out their orientation, while patients with various types of agnosia have the reverse impairment – they can act on objects but are often completely unable to identify them. Neuroanatomical evidence points towards a distinction between two different visual pathways leading from the visual cortex – the dorsal pathway projecting to the posterior parietal cortex and the ventral pathway leading to the inferotemporal cortex. The functional distinction between the dorsal and ventral pathways was originally construed in terms of the distinction between “where” and “what”, with the dorsal stream primarily involved in computing distance and location and the ventral stream specialized for the type of color and form processing that feeds into object identification (Mishkin and Ungerleider 1982). More recent investigation has suggested that the dorsal pathway is also involved in computing the “how” of action (Milner and Goodale 1995). The two visual systems hypothesis offers an interesting example of how behavior can be explained in terms of the representation of micro-features. One of the striking experimental data that has emerged from investigation of the differences between vision for action and vision for identification is that the two different systems can come into conflict. We see this, for example, in work that has been done on visual illusions, where the illusions have a much greater effect on perceptual reports than on action performance. The Ebbinghaus size contrast illusion is a case in point. 27 As the diagram indicates, a circle surrounded by other circles will appear smaller if the surrounding circles are enlarged. When (normal) subjects are presented with two circles of the same size, one of which is surrounded by small circles and the other surrounded by large circles, they will reliably judge the one surrounded by small circles to be larger than the one surrounded by large circles. Yet the illusion does not carry over to action. When subjects are asked to reach out as if they were going to pick up the circles their grip aperture is constant for the two circles (Aglioti, DeSouza and Goodale 1995). Similar effects have been observed with Muller-Lyer and Ponzo illusions. The dissociations between behavior and report in these visual illusions suggest that we respond to properties such as graspability that are a function of the size of the object and yet that are clearly distinct from the object’s perceived size. These are micro-features that it is difficult to assimilate within the conceptual framework of commonsense psychology. It is a key tenet of commonsense psychology, for example, that we act upon objects in virtue of how they appear to us, so that it is because an object looks a certain size to us that we make the appropriate hand movements for grasping it. Yet the experimental evidence suggests that things cannot be as simple as this. If subjects acted upon how objects appear to them (more strictly: on how they report objects appearing to them) then they would act differently in the two cases. Instead it looks 28 as if subjects are sensitive to properties of objects that are correlated with actual size but are independent of perceived size in a way that operates outside the realm of conscious awareness. The dimensional-action system We have seen that the eliminativist can place pressure on the hegemony of propositional attitude psychology by identifying links between perception and action that are based on the representation of micro-features and that shortcut anything that might plausibly be described as “central processing”. Intriguing recent results in perceptual psychology provide further examples of such microfeature-based perception-action links. Traditional information processing accounts make a sharp distinction between perceptual processing and post-perceptual processing and see all motor processing and response selection as falling clearly on the post-perceptual side of the divide. Whereas perceptual processing is widely held to involve the separate processing of micro-features (with shape, form, color and so on all being processed in neurally distinct areas), post-perceptual processing is thought to take place downstream of the “binding” of those micro-features to form representations of objects. Yet some intriguing recent experimental evidence has led theorists to postulate highly specialized perception-action links that are explicitly tied to the perception of micro-features (Cohen and Fientuch 2002). As with the dissociations discussed in the previous section, the experimental evidence seems to show that we can act upon isolated features of objects in complete independence of other features of those objects. It has been know for a long time that there are regions of the mammalian visual system perceptually sensitive simply to color or to shape, but it has always been thought that we can only act upon the world by somehow combining these separately processed features into representations of objects – into representations that operate at the symbolic level of commonsense psychology, rather than at the sub-symbolic level of microfeatures. Yet this assumption appears to be called into question by research into the dimensional action system. 29 One representative set of experiments was carried out with the so-called flanker task interference paradigm. Subjects are instructed to make differential responses to types of object presented at the center of a display while ignoring peripheral distractors flanking the target object. In the experiments reported in Cohen and Shoup 1997 the targets and responses were as follow. The first response was to be made to the appearance either of a red vertical line or of a blue right diagonal line, while the second was to be made to the appearance either of a green vertical line or of a blue left diagonal line. So, if the target object has a vertical orientation the appropriate response can only be made on the basis of color, while if it is blue the appropriate response can only be made on the basis of orientation. The distractors were lines of varying colors and orientations. Strikingly, interference effects were only observed when the relevant responses for target and distractor were on the same dimension. INTERFERENCE Trial User! 8/20/04 3:26 PM Formatted NO INTERFERENCE 30 So, for example, there would be an interference effect if a red vertical line were flanked by differently colored vertical lines, but not if it were flanked by red diagonal lines. Similarly, there would be interference if a blue left diagonal were flanked by other diagonal lines, but not if were flanked by differently colored diagonal lines. The conclusion drawn is that there are distinct processing channels linking the detection of individual micro-features (a particular orientation, or a particular color) with particular responses. These processing channels operate without any “binding” of the relevant micro-features. In this case it is not the micro-features themselves that is surprising, or the fact that we are able to act upon perceived color and perceived shape. What is surprising, and difficult to assimilate within the conceptual framework of propositional attitude psychology, is that there seem to be perception-action links triggered by representations of micro-features that are independent of representations of objects, or indeed of representations of other micro-features. The ecological approach to perception and action J. J. Gibson’s ecological approach to perception and action is best known for the claim that the line between perception and cognition is far less sharply defined than it is standardly taken to be. There are ways of perceiving the world that have direct implications for action. Frequently what we perceive are the possibilities that the environment “affords” for action, so that we can act on how we perceive the world to be, without having to form or exploit beliefs and other propositional attitudes. An affordance is a resource or support that the environment offers a particular creature (the possibility of shelter, for example, or the availability of food). These affordances are objective features of the environment that are supposed to be directly perceived in the patterns of light in the optic flow. Gibsonian psychologists describe organisms as “resonating to” or “directly sensitive to” the affordances in the distal environment. 31 This basic claim about affordances is not particularly radical, and in itself provides little or no support for the eliminativist project. If we describe affordances in more neutral terms as the instrumental properties of objects in the environment then it is hard to see why there should be any incommensurability between perceptual sensitivity to those instrumental properties and the conceptual framework of commonsense psychology. Scope for incommensurability does appear, however, when we consider the general account of the workings of perception that underlies Gibsonian claims about the direct perception of affordances. One of Gibson’s major contributions to the study of vision is the proposal to reconstrue the visual field as a constantly moving and constantly reconfiguring set of illuminated surfaces and concomitant solid visual angles, rather than in terms of empty space containing bounded objects (figures on a ground). We do not, he thinks, ever see empty space surrounding discrete objects. What we see is a complex and gapless structure of surfaces. Some of these surfaces are surfaces of objects, while others are not (the various surfaces in the sky, for example). To each surface there corresponds a solid visual angle with its base at the face of the visible surface and its apex at the point of observation. We can, for simplicity’s sake, think of these solid angles as cones, although of course their shape will vary with the visible outline of the surface in question. As the observer moves through the environment the solid angles change, as one surface moves in front of another (relative to the perceiver) or as the observer approaches or moves away from the surface. This is what Gibson terms optic flow. The ecological analysis of visual perception gives us an example of how representation in terms of micro-features might work in practice. There is a fundamental mismatch between a characterization of the distal environment in terms of objects and properties (of the sort that might feature in specifications of propositional attitudes) and a characterization of the distal environment in terms of optic flow. Gibson’s perspective on perception rests upon the perception of microfeatures that resist assimilation to macro-features. These micro-features include, for example: • texture gradients (the decrease with distance of discriminable fineness of detail) 32 • the focus of expansion (the aiming point of locomotion that is also the vanishing point of optic flow) • visual solid angles Gibsonian accounts of perception attempt to show how behavior is controlled by perceptual sensitivity towards these micro-features, which are properties of the global optic array, rather than of individual objects. Within its sphere of applicability, Gibsonian psychology certainly seems to provide an account of the “springs of action” that supports eliminativism by offering explanations of behavior that resist assimilation to the concepts and categories of commonsense psychology. Dynamic touch and rotational inertia The eliminativist needs to identify behaviors that can be shown to involve responding to microfeatures of the environment that resist assimilation within the conceptual framework of commonsense psychology. A good example of this has come from perceptual psychologists working within a broadly Gibsonian tradition and exploring the phenomenon of dynamic touch. It is well known that people can make accurate assessments of the spatial properties of objects by manipulating those objects. So, for example, people are remarkably accurate at detecting the length of objects by grasping those objects at a single point and moving the object around (without running their fingers over the whole object). One can get a feel for the phenomenon by picking up a pen, closing one’s eyes and rotating the pen with one’s fingers – or taking a slightly larger object, such as a metal rod, and rotating it around one’s wrist. Dynamic touch is a puzzling phenomenon, because the haptic system does not have access to any direct perceptual information about the length of the pen or the ruler. The physiological underpinnings of dynamic touch are mechanoreceptors yielding information about the stretching, contraction and twisting of muscles and tendons. Very little work is done by receptors on the surface of the skin, even at the point where contact is made with the object. Clearly there is some mechanical property (or properties) of objects that is reliably correlated with changes in the mechanoreceptors. The obvious candidates are weight and rotational force (torque), but neither of 33 these can do the job. Perceived length is independent of both weight and torque, as can easily be appreciated by manipulating a pen and a pencil of the same length but different weights, and by manipulating both of them with very different twisting forces. The key mechanical property must remain invariant through changes in torque and weight It turns out (Turvey 1996, Carello and Turvey 2004) that the relevant physical invariant is what is known as the inertia ellipsoid. The inertia ellipsoid is, roughly speaking, a way of characterizing an object that measures the object’s resistance to being rotated. It is derived from the object’s principal moments of inertia, where a principal moment of inertia quantifies an object’s resistance to rotation around one of its axes of symmetry. An object’s moment of inertia will vary according to the distribution of its mass, with higher concentrations of mass away from the object’s center of gravity yielding a higher moment of inertia. So, for example, suppose we hang a weight from a metal rod. The further away the attachment point is from the rod’s center of gravity the greater the moment of inertia – and the more force will be required to rotate it. Once we know an object’s axes of symmetry and its principal moments of inertia we can characterize its overall resistance to being rotated in terms of an ellipse whose center is the intersection of the three axes of symmetry and whose surface is obtained from the reciprocal of the square roots of the principal moments of inertia. INSERT FIGURE 1 FROM P. 26 OF CARELLO AND TURVEY 2004 It is a robust finding that an object’s rotational inertia, as given by the inertia ellipsoid, is the invariant underlying perceived length. Nor is length the only quantity that can be detected by perceived touch. People can make reliable estimates of an object’s weight from wielding the object. Amazeen and Turvey (1996) have established that perceived heaviness is also a function of 34 rotational inertia, both when perceived heaviness accurately tracks and object’s weight and when (as in the size-weight illusion) it leads to misleading estimates.7 This sensitivity to rotational inertia is a further example of the type of microfeatural sensitivity suggested by the Gibsonian approach to perception and action. We act upon the world in virtue of our perceptual attunement to properties of objects and of the optic array that are fundamentally alien to the conceptual framework of commonsense psychology. The inertial ellipsoid is a mathematical object that stands to our commonsense thinking about objects and their dynamic and kinematic properties in something like the relation that the rules of transformational grammar stand to our everyday use of English. Subtle experimental work is required to identify rotational inertia as the relevant parameter in our haptic sensitivity to the spatial properties of objects. The influence of situation in social psychology The “springs of action” have been investigated by social psychologists as well as cognitive psychologists and neuroscientists. The research has been two-pronged, investigating both why people behave the way they do and how we interpret that behavior. Two features of this research are particularly salient in the present context. The first has to do with the genesis of behavior. There is an overwhelming body of evidence highlighting the importance of the situation in determining behavior. Situational changes that might seem at first sight to be insignificant have been shown to have a serious effect on behavior. The second feature is that subjects systematically underestimate the significance of the situation, making what has come to be known as the “fundamental attribution error” of over-estimating the significance of character traits and personality in explaining and predicting behavior. 7 The size-weight illusion is the illusion that larger objects of the same weight are perceived as heavier. 35 In one famous set of experiments (Darley and Batson 1973) groups of students at a Princeton theological seminary were sent from one building to another as part of an experiment putatively on religious education. Their task in the second building was to give a talk, with one group giving a talk on the Good Samaritan and another on jobs in seminaries. On the way over they passed an experimenter slumped in a doorway and moaning. Overall 40% of the subjects stopped to offer some sort of assistance. What is striking, though, is the drastic difference between subjects who were told that they were running late (only 10% offered assistance) and subjects who were told that they had time to spare (where the figure was 63%). This discrepancy did not seem to be correlated with any other differences between the participants. Other experiments have found what seem prima facie to be even more trivial situational factors having a large impact on behavior. Mathews and Canon 1975 explored the influence of ambient noise, showing that subjects are 5 times less likely to help an apparently injured man who has dropped some books when there is a power mower running nearby than when ambient noise is at normal levels. Isen and Levin 1972 found an even more striking effect, discovering that people who had just found a dime were 22 times more likely to help a woman who has dropped some papers than people who had not found a dime. Experiments such as these have been carried out many times and the powerful influence of situational factors has proved a very robust finding. This experimental tradition poses numerous interesting philosophical problems, particularly with respect to the role that character plays in ethical theory (Doris 2002). For present purposes what is interesting is the perspective that situationist social psychology casts on the genesis of behavior. It looks very much as features of situations that do not in any sense count as commonsense psychological reasons for action can play a large role in determining how people behave. At least in the experimental paradigms it seems fundamentally inappropriate to seek explanations in terms of propositional attitudes that act as reasons for action. An important element in the springs of action seems to be relatively low-level features of the situation – what might, in fact, be termed situational micro-features. 36 6 Prospects for eliminativism? The constraints upon a successful argument for eliminativism should by now be clear. Eliminative materialism is an error theory and hence the most plausible way of arguing for it is to identify precisely the error that commonsense psychology is supposed to be committing. The error must come in the very idea that the springs of action can be explained through the concepts and categories of propositional attitude psychology, rather than in a particular way of thinking about how those concepts and categories are to be applied or in a very general feature such as a commitment to the role of representational content in psychological explanation. The eliminativist’s aims are not best served by focusing, as Paul Churchland sometimes does, on the idea that commonsense psychology is a proto-scientific theory to be judged by the standards appropriate to scientific theories. Nor is the eliminativist well served by mounting a global attack on the very idea of content-bearing states. The eliminativist needs an argument that will engage those who take a fundamentally different of what commonsense psychology is and how it is to be applied. And it is important that the eliminativist thesis be formulated in a way that avoids the charges of incoherence that might plausibly be leveled against global attacks on the status of content. The most plausible way of meeting these various constraints, I have argued, is for the eliminativist to identify two very different types of error in commonsense psychology and how we apply it. The first putative error is an error about the scope of commonsense psychology. The eliminativist can argue that we rely far less on commonsense psychology than is generally assumed by pointing to far more primitive mechanisms that control social coordination and governing our understanding of ourselves and others. The case here seems plausible. It seems very likely that the significance of commonsense psychology is significantly over-stated by philosophers of mind. 37 But the real motivation for eliminativism has to come from a more direct attack on the role that propositional attitudes are supposed to play in the genesis of behavior. The eliminativist needs to argue that the representations that feed into action are fundamentally different from those invoked by propositional attitude psychology. The “springs of action” are representations of features that are much more finely grained than those encoded within the vocabulary that we employ to specify the content of propositional attitudes. The eliminativist’s most promising strategy is to argue that, whatever we might think about why we behave the way we do, careful experimental work will show that we are in fact acting in virtue of representations of properties and microfeatures that fall completely outside the ambit of propositional attitude psychology. We have considered a number of examples of how this eliminativist strategy might be developed, ranging from the implications of the two visual systems hypothesis to research in social psychology into the role that situational factors play in controlling action. Of course, even if the examples I have proposed on the eliminativist’s behalf are accepted, the eliminativist case is a long way from being made. With the exception of the experiments on situational factors in social psychology, the examples given are all of relatively low-level motor tasks. The real question is how well the examples scale up. Can the eliminativist give comparable accounts of what is going on in the far more complex cases that tend to be cited when the thesis of the ineliminability of commonsense psychology is proposed? It is very difficult to say. Simpler tasks have been much more comprehensively studied than complex ones. No doubt many will think that beliefs, desires and other propositional attitudes have an irreducible role to play in explaining complex actions and social coordination. But, on the other hand, the case for ineliminability needs to be made. One clear benefit of exploring the dialectic of eliminativism in the way we have been doing is that it makes clear the level at which the debate about the status of commonsense psychology will have to be made. It is simply not good enough to proclaim the ineliminability of the conceptual framework of propositional attitude psychology, or to make intuitive appeals to the possibility of formulating generalizations that capture patterns in behavior 38 that cannot be captured without bringing in commonsense psychological concepts. The debate about the status and future of commonsense psychology must take place on the basis of a detailed study of the genesis and execution of action. One thing that we learn from looking at simply motor behavior is that our intuitions about what is going on when we make simple discriminations or wield objects are systematically misleading. Could the same be going on in more complex behaviors? If, so then it may well turn out, as the eliminativist claims, that the scope of commonsense psychology will shrink to an extensionless point, as the representations that drive behavior are discovered to have little in common with the belief-desire pairs that are the mainstay of propositional attitude psychology. That would be an outcome very much in the spirit of Paul Churchland’s eliminative materialism, even if the means by which it is reached are rather different from Churchland’s “official” arguments for eliminativism. References Aglioti, S., DeSouza, J.F.X., & Goodale, M.A. (1995). Size-contrast illusions deceive the eye but not the hand. Current Biology, 5, 679-685. Amazeen, E.L., & Turvey, M.T. 1996. Weight perception and the haptic size-weight illusion are functions of the inertia tensor. Journal of Experimental Psychology: Human Perception and Performance, 22,213-232. Axelrod, R. (1984) The Evolution of Cooperation, Harmondsworth: Penguin. Bermúdez, J. L. 2003. The domain of folk psychology. In A. O’Hear (Ed.), Mind and Persons. Cambridge. Cambridge University Press. Bermúdez, J. L. 2005. Philosophy of Psychology: A Contemporary Introduction. London. Routledge. 2002. Boghossian, P. 1990. The status of content. The Philosophical Review 99, 157–184. 39 Carello, C., & Turvey, MT (2004). Physics and psychology of the muscle sense. Current Directions in Psychological Science, 13, 25-28. Churchland, P. M. 1981. Eliminative materialism and the propositional attitudes, Journal of Philosophy, 78, 67-90. Churchland, P. M. 1989. Folk psychology and the explanation of human behavior’, Philosophical Perspectives 3, 225 – 241. Churchland, P.M. 1992. Activation vectors vs. propositional attitudes: How the brain represents reality, Philosophy and Phenomenological Research 52, 419-524. Churchland, P.M. (1998). Conceptual similarity across sensory and neural diversity: The Fodor/Lepore challenge answered, Journal of Philosophy 65, 5-32 Churchland, P. M. and Churchland P. S. 1998. On the Contrary : Critical Essays, 1987-1997. Cambridge, Mass.: MIT Press. Churchland, P.S. and Sejnowski, T. J. 1992. The Computational Brain, Cambridge: MIT Press/Bradford Books.\ Cohen, A. and Feintuck, U. 2002. The dimensional-action system: A distinct visual system. In W. prinz and B. Hommel (Eds.), Common Mechanisms in Perception and Action: Attention and Performance XIX. Oxford. Oxford University Press. Cohen, A, & Shoup, R. (1997). Perceptual dimensional constraints on response selection processes. Cognitive Psychology, 32, 128-181. Darley, J. M. and Batson, D. 1973. From Jerusalem to Jericho: A study of situational and dispositional variables in helping behavior. 1973, Journal of Personality and Social Psychology, 27, 100-108. Davies, M. K., Stone, T. (eds.) 1995. Folk Psychology: The Theory of Mind Debate, Oxford: Blackwell. Doris, J. 2002. Lack of Character. New York. Cambridge University Press. Fodor, J. and Pylyshyn, Z. 1988. Connectionism and cognitive architecture: A critical analysis’, Cognition, 28: 3-71. 40 Isen, AM, & Levin, PA 1972. Effect of feeling good on helping: Cookies and kindness. Journal of Personality and Social Psychology, 21, 384-388. Laakso, Aarre and Cottrell, Garrison W. 1998. How can I know what You think?: Assessing representational similarity in neural systems. Proceedings of the Twentieth Annual Cognitive Science Conference, Madison, WI, Mahwah: Lawrence Erlbaum. Lewis, D. 1994. Reduction of mind. In S. Guttenplan (Ed.), Companion to the Philosophy of Mind. Oxford. Blackwell. Macdonald, C. and Macdonald, G. 1995. Connectionism: Debates in Psychological Explanation. Oxford. Basil Blackwell. Matthews, KE, & Canon, LK (1975). Environmental noise level as a determinant of helping behavior. Journal of Personality and Social Psychology, 32, 571-577. Maynard Smith, J. 1982. Evolution and the Theory of Games, Cambridge: Cambridge University Press. Milner, A. D. and Goodale, MA. 1995. The Visual Brain in Action. Oxford. Oxford University Press. Pettit, P. 1991. Decision theory as folk psychology. In M. Bacharach & S. Hurley (eds.), Foundations of Decision Theory, Oxford. Blackwell, 1991. Sejnowski, T.J. & Rosenberg, C. (1987) Parallel networks that learn to pronounce English text. Complex Systems 1:145-68. Skyrms, B. 1996. The Evolution of the Social Contract, Cambridge: Cambridge University Press. Smolensky, P. 1988. On the proper treatment of connectionism, Behavioral and Brain Sciences, 11-1: 1-23. Turvey M T, 1996. Dynamic touch. American Psychologist 51, 1134-1151. Ungerleider, L.G. & Mishkin, M. (1982). Two cortical visual systems. In D.J. Ingle, M.A. Goodale & R.J.W. Mansfield (Eds). Analysis of Visual Behavior. Cambridge, MA: MIT Press. 41