ANGELA ARKWAY
THE SIMULATION THEORY, THE THEORY THEORY AND FOLK PSYCHOLOGICAL EXPLANATION∗ (Received in revised form 22 October 1997)
I. INTRODUCTION
Much interest has been raised recently in cognitive science and in the philosophy of mind by a debate that focuses on the nature of the cognitive mechanism that underlies our folk psychological practices. One side in this debate is represented by proponents of the reigning paradigm, the theory theory. Theory theorists say that our ability to give explanations, predictions and interpretations of intentional behavior is subserved by tacit knowledge of an internally-represented theory of commonsense psychology (Fodor, 1987). The simulation theory challenges this view on the grounds that there is no evidence to support the suggestion that we have such knowledge and some evidence to suppose that we do not (Gordon, 1986; Goldman, 1989). It is more likely, say simulationists, that these abilities are simply underpinned by the innate capacity to simulate others. Simulationists argue that a compelling case can be made against the theory theory. I do not discuss that case here. Instead I examine an issue which, although crucial to the outcome of the debate, has not yet been addressed. This issue concerns the nature of the commonsense psychological explanations produced by the mechanism whose functioning both theories claim to describe. Two assumptions bring the issue of explanation to the fore. One of these concerns the range and the other the type of event that the theories are adduced to account for. The merits of the theory theory and the simulation theory are usually discussed in relation to the practice of the prediction of intentional behavior. The reasonable assumption is that the mechanism that is deployed in prediction will Philosophical Studies 98: 115–137, 2000. © 2000 Kluwer Academic Publishers. Printed in the Netherlands.
116
ANGELA ARKWAY
be the same one that is deployed in the explanation, the description and the interpretation of our own and others’ behavior. The second assumption is that the two theories offer competing accounts of the same sort of event. The theory theory and the simulation theory, in offering alternative views of the mechanism underlying our folk psychological practices of prediction, explanation, description and interpretation of intentional behavior, agree on what these practices consist in. With regard to explanation, the grounds for the latter assumption are not clear. Theory theorists hold that the psychological explanations produced by the folk are robust causal explanations of behavior. They substantiate their claim by spelling out the conditions that a bona fide folk psychological explanation of this type satisfies. Simulation theorists have not specified the type of explanation that the folk produce in deploying the mechanism of simulation. If, as the assumptions above indicate, the simulationist challenge to the theory theory includes an implicit claim to account for the same type of explanation-giving as theory theorists claim to account for, there is work to be done in the simulation camp. A claim that the simulation mechanism is responsible for robust-explanation-giving needs to be substantiated by an account of the conditions that a simulation-produced explanation of this type satisfies. Attempting to provide the simulation theory with such an account is the task I undertake in this paper. It is important to point out what hangs on the outcome of this discussion. Failure to find an adequate simulation-based account of robust-explanation-giving will require the assumptions underlying the debate to be reconsidered. Discarding the assumption that the simulation theory and the theory theory offer competing accounts of the mechanism that underlies all of our folk psychological practices has the unwelcome implication that a different mechanism from the one that underlies prediction, or maybe even several, underlie(s) the folk psychological practices of explanation, description and interpretation. If the assumption that the two theories offer accounts of the full range of our folk psychological practices is retained, the lack of an account of simulation-based causal explanation will entail the rejection of the second assumption; the simulation theory and the theory theory, at least insofar as folk psychological explana-
THE SIMULATION THEORY
117
tion is concerned, will be theories about different kinds of event. In this case, before the debate about the cognitive mechanism responsible for our folk psychological practices can continue, the fundamental issue of the kind of explanation that the folk produce when they account successfully for someone’s behavior will need to be addressed. These difficulties will be avoided however if it turns out that there is an adequate simulation-based account of robust causal explanation. The groundwork for the discussion of this issue is laid out in the next two sections. In Section II I describe the theory theory focusing on what it says about the type of psychological explanation produced by the folk. In Section III I describe the simulation theory of the mechanism deployed in our folk psychological practices. Here the emphasis is on what it says about the practice of explanation. In Sections IV and V I attempt to substantiate the simulationist challenge to the theory theory by giving an account of correct simulation-produced explanation that is of the same type that theory theorists claim to account for.
II. THE THEORY THEORY AND FOLK PSYCHOLOGICAL EXPLANATION
The basic premise of the theory theory of folk psychology is that the mechanism underlying the performance of our folk psychological practices of the prediction, the explanation and the interpretation of intentional behavior exploits an internally-represented body of information, or a theory, about psychological processes and the ways in which they give rise to behavior. Proponents of the standard version of the theory theory hold that the internally-represented body of information consists of knowledge of a network of causal laws which link mental states to behavior, to other mental states, and to sensory input (Fodor, 1990, p. 145). This type of theory theorist acknowledges that psychological laws will not be strict laws of universal form, laws that state that whenever you have a G event you have an F event. A suggestion that is popular is that the laws include clauses that state that they hold in cases where all else is equal. On this view intentional laws are unstrict laws, laws that admit of exceptions, that is, “ceteris
118
ANGELA ARKWAY
paribus” or “hedged” laws. Laws of this type are usually expressed as universally quantified conditional statements. The antecedent of an intentional ceteris paribus law will consist of a conjunction of the relevant explanatory factors and the consequent will be the behavioral event to be explained. Such a law might take the following form: if x wants that p, and x believes that not-p unless q, and x believes that x can bring it about that q, then, ceteris paribus, x tries to bring it about that q. Theory theorists who hold that the psychological knowledge used in our folk psychological practices is internally represented in the form of laws also hold that correct commonsense psychological explanations are covering law explanations of the deductivenomological type: such explanations are correct when the event being explained is subsumed by an internally-represented, (perhaps only tacitly-known), commonsense psychological law and when the explainer alludes (in some sense) to this law in giving his explanation. That these laws are ceteris paribus does not disqualify them from playing the role in explanation usually reserved for strict laws since, in the words of theory theorist Jerry Fodor (1990, p. 154), “[s]trict laws and hedged laws with satisfied ceteris paribus conditions operate alike in respect of their roles in covering law relations and in respect of their roles in covering law explanations.” So this version of the theory theory holds not only that lawdeployment is the mechanism that underlies the performance of our folk psychological practices, but also and importantly, that a correct commonsense psychological explanation produced by this mechanism is a deductive argument in which a covering law is implicitly deployed (Fodor, 1987, p. 7). The psychological explanations produced by the folk are of the same type as explanations in basic science and as explanations in the special sciences; causal explanations of the covering law model. They cite what it is that is causally relevant to the event being explained in just the same sense that correct scientific explanations do. The fact that its account of commonsense psychological explanation fits the standard model of explanation makes this version of the theory theory particularly attractive.
THE SIMULATION THEORY
119
III. SIMULATION AND FOLK PSYCHOLOGICAL EXPLANATION
A common version of the simulation theory is that when we predict the behavior of another person, or of ourselves at a time remote from the present, the mechanism which governs the daily interaction of our beliefs, desires and other intentional and qualitative states, the practical-reasoning mechanism, is disengaged from its actual inputs, viz. from external stimuli and from our own salient beliefs and desires. At the same time, this mechanism is disconnected from the action controllers, the mental mechanisms responsible for a decision-to-behave being translated into actual behavior. Operating “off-line” in this manner, the decision-making mechanism is fed pretend-input in the form of those beliefs and desires we imagine we would instantiate ourselves if we were in the situation of the person whose behavior we are about to predict. The practical-reasoning mechanism then processes these pretend inputs and a pretend decision-to-behave is generated. This pretend decision-to-behave is transformed into the prediction of behavior. The mechanism that subserves predictive practice also subserves explanatory practice, claim simulationists. In explaining a behavioral episode we feed into our disengaged decision-making mechanism those beliefs and desires which we imagine are likely to produce a decision to perform the behavior we want to explain. Alvin Goldman (1989) describes how the simulation procedure can be used to make this explanatory or ‘retrodictive’ assignment of mental states: If you make a surprising chess move, I may infer a new strategy on your part, one that might have led me, if I were in your situation, to the observed move. To assure the plausibility of this being your strategy, I would see whether I could simulate both (a) arriving at this strategy from your presumed antecedent states, and (b) choosing the observed move given this strategy. (p. 169)
He points out that although my inference to a new state of yours draws on assumptions about your prior mental states, assumptions which will also be the result of simulation, the regress stops at perceptual cases and at basic likings and cravings. He says that we infer from the perceptual situation of the explainee that he has certain perceptual input, the same ones that we would have if we were in his situation, and that we may also assume that he has the same need for food, love, warmth, etc.
120
ANGELA ARKWAY
Simulationist Robert Gordon (1986, p. 164) suggests that in the case where several sets of pretend-beliefs and pretend-desires, when fed into the decision-making mechanism, render the desired outcome, that is, a decision to perform the behavior we want to explain, we simply apply the principle of least pretending: we choose the set that requires the shifts in input that are the least different from those upon which we operate most of the time. It is clear from the above that the simulation theory, just like the theory theory, purports to offer an account of the mechanism underlying the folk practices of the prediction and of the explanation of intentional behavior. The theory theory tells us that the mechanism responsible for the production of explanations of behavior by the folk is law-deployment. It also says that when the folk give a correct explanation of this type it is because the behavior to be explained is subsumed by a (maybe only tacitly known) causal law and this law is alluded to by the explainer in giving this explanation. The explanations produced by the mechanism of law-deployment are robust causal covering law explanations. The simulation theory, although it claims to describe the mechanism deployed in the production of commonsense belief-desire explanations, does not say what kind of explanations simulation-produced explanations are. In particular, the simulation theory does not say whether correct simulation-produced explanations explain in the same robust sense that scientific explanations do. Since there is no argument to the contrary in the literature on the debate, the working assumption in the discussion that follows is that simulation-produced explanations are of the same robust type as those produced by law-deployment, that they explain behavior in terms of its causes. IV. SIMULATION, EXPLANATION, AND CAUSAL LAWS
Let us suppose that the conditions that are satisfied by the correct simulation-produced commonsense psychological explanation “Oprah went to London because she believed p and desired q” are the following: (SE) The statement that Oprah went to London because she believed p and desired q is correct explanation of her going to London iff
THE SIMULATION THEORY
121
(1) she went to London (2) she believed p and desired q, and (3) a decision to go to London is the result of a simulation of Oprah run on the explainer’s decision-making mechanism and that simulation satisfies two conditions: (a) it essentially involves the belief p and the desire q (b) there are no facts about Oprah such that were the explainer to know them a different decision would result from his simulation of her.1 A prima facie problem with this formulation is that if, as assumed in the simulation condition at (3), the behaver, Oprah, is endowed with a decision-making mechanism that essentially involves her psychological states, it is not clear what the goings-on in the explainer’s head will have to do with what makes an explanation of Oprah’s behavior correct. If Oprah is thus equipped, then it seems true that part of what will make the statement a correct explanation is that the decision to go to London was the result of a decision-making process in Oprah that essentially involved the belief p and the desire q, and the fact that she did go to London. However the suggestion that whatever else is needed in order to make that statement a correct (robust) explanation will be something to do with the explainer’s decision-making mechanism is just implausible. And if it is the case that what the explainer does when giving the explanation is irrelevant to what it is that actually explains Oprah’s decision to go to London, then it follows that satisfaction of the simulation condition at (SE) (3) is neither necessary nor sufficient for the correctness of the explanation. A second objection, one that supports the suggestion that a simulation performed by one person is irrelevant to whatever it is that explains the behavior of another, is that there are counterexamples that seem to indicate that satisfaction of (SE) is not necessary for the correctness of an explanation. Let us suppose that “Oprah went to London because she believed that Michael Jackson was there and she wanted to interview him” is the correct explanation of the fact that Oprah went to London. It seems that an explainer could give this correct explanation of Oprah’s going to London and at the same time know that if he were in Oprah’s situation with her relevant beliefs and desires he would not decide to go to London because
122
ANGELA ARKWAY
he believed that Michael Jackson was there and he, the explainer, wanted to interview Jackson. That is, even though his explanation of Oprah’s behavior does not satisfy (SE) because a simulation of her run on his mechanism does not produce a decision to go to London, it is still correct. If it is true that an explainer retains his ability to produce a correct explanation of behavior in spite of the fact that his processing of the pretend beliefs and desires produces a different decision from the one that the behaver’s processing produces, then satisfaction of the simulation condition will not be necessary for a correct explanation. Moreover, it seems that were an explainer to rely on simulation, the correct explanation would probably elude him. A likely simulationist response to this type of counterexample would be to point out that although the above correct explanation does not appear to satisfy the simulation condition at (SE) (3), that condition is in fact satisfied. Such a theorist might argue that since our capacity for practical simulation operates at a sub-verbal level most of the time, and since, when simulating, our decisionmaking system “gets partially disengaged from its ‘natural’ inputs and fed instead with suppositions or images (or their ‘subdoxastic’ or ‘subpersonal’ counterparts)” (Gordon, 1986, p. 170), the correct explanation does satisfy (SE) (3) in spite of the fact that the explainer knows on a personal level that he would not have made the same decision as Oprah had he been in her position. This response to the second objection, however, elicits a third. Martin Davies (1994) has pointed out that under some construals simulation runs the risk of collapse into the theory theory. When the inputs to the simulating system are taken to be representational states whose contents themselves concern mental states, contents of the form “I believe that p” and “I desire that q”, what will be generated by their processing and then ascribed to the person being simulated is either a conclusion about a further mental state, for example “I believe that r”, or a conclusion about an action, “I V”, or about an intention to act, “I intend to V” (p. 114). Davies holds that this kind of simulation will count as process-driven, as opposed to theory-driven, if the processing in the simulator is isomorphic to the process being simulated. However, besides being isomorphic
THE SIMULATION THEORY
123
the processing in the simulator could also follow the contours of the derivational structure of a proof of a conclusion about, say, the agent’s beliefs and desires – a proof cast in a psychological theory. This matching of structure between simulation process and deductive derivation could be quite general. wherever, for example, two proofs draw upon a common axiom in the theory, the two pieces of simulation might draw upon a correspondingly common cognitive mechanism. (p. 115)
Davies points out that the threat of collapse ensues because it is the idea of the matching of structure between a causal process and a derivational structure that is used in some versions of the theory theory: Roughly speaking, a component processing mechanism embodies tacit knowledge of a particular rule or axiom if it plays a role in mediating causally between representational states that is structurally analogous to the role that the rule or axiom itself plays in mediating derivationally between premises and conclusions. (p. 115)
Furthermore, Davies says, since inputs such as “I believe that p” exhibit the general form “x believes that p”, the states of entertaining them are appropriate inputs to a mechanism that embodies tacit knowledge of a psychological theory. What these remarks underline is that the construal of the simulation process which has the simulator entertaining hypotheses about mental states, not only does not exclude the possibility that the simulating process is theory-driven, but, given the form of the content of the input that this construal presupposes, seems to indicate that it is theory-driven. Davies goes on to suggest that by appealing to a different construal of simulation the threat of collapse into the theory theory can be avoided (1994, p. 117). Under this construal, the simulator simulates the mental processing of another by identifying with that other in his imagination. That is, the simulator, instead of imaginatively entertaining hypotheses about mental states, projects himself onto the other and imaginatively adopts what he takes to be the other’s mental states; he imagines believing that p and desiring that q. Thus the states processed by the decision-making mechanism while simulating another are simply pretend-states with the contents “that p” or “that q” and what is generated is either pretend-beliefs, or a pretenddecision to behave in such and such a manner. Processing mechan-
124
ANGELA ARKWAY
isms that mediate transitions amongst states with such contents are not going to be embodiments of tacit knowledge of the principles of a psychological theory, says Davies (p. 117). However even if this construal avoids the threat of collapse there is still a problem. A robust reading of folk psychological explanation requires that the “essential involvement” of psychological states in decision-making be construed in terms of their causal relevance to the decision made. A widespread philosophical assumption is that events related as cause and effect fall under causal laws. So a theorist who supports a robust account of simulation-produced explanation must either say what kind of causal laws there are that subsume psychological states or else find a causal account of folk psychological explanation that does not appeal to psychological laws. The remainder of this section is devoted to examining the first option. In the next section I discuss a suggestion for a robust simulation-based account of explanation that does not appeal to psychological laws. It is important to note that the appeal to psychological causal laws in this construal of simulation does not entail the risk of the collapse of the simulation theory into the theory theory. The theory theory is usually taken as an internally-represented body of information. In the case at hand the posited background causal laws are not internally represented and tacitly known. They are laws which are just assumed to exist. If one finds the explanation of folk physics “the window broke because it was hit by a rock” an acceptable commonsense causal explanation of the fact that the window broke, and if one holds that this is so because there are unknown background laws which subsume these events, then there is no reason to deny that there are unknown underlying laws in the folk psychological explanation “Oprah went to London because she believed p and desired q.” The most likely candidates for these unknown causal laws are intentional ceteris paribus laws. There is, however, a standing argument in the literature against the existence of laws of this type. Stephen Schiffer pointed out recently that if psychological ceteris paribus sentences express true propositions, these propositions will accommodate the fact that psychological states have multiple realizations. Schiffer argues that such propositions are not, nor will they determine, anything worth calling psychological ceteris paribus
THE SIMULATION THEORY
125
laws, anything that could play the role of psychological covering law in a folk psychological explanation of behavior (Schiffer, 1991, p. 7).2 Another likely candidate for the causal laws that subsume psychological states and account for the correctness of our explanations are laws of a type which, were they knowable, could not be expressed in intentional terms. The psychological states referred to in our folk psychological explanations and predictions of behavior in commonsense vocabulary might have another description, maybe a physical description, by virtue of which they are subsumed by unknown causal laws. In the explanation of Oprah’s going to London, for example, the neural event tokens in the causal chain that ended in Oprah’s going to London would include one that is a belief that Michael Jackson is in London and another that is the desire to interview him. This is, there is belief and a desire that issue in some other events and finally in an event which is the piece of behavior that is the going to London. It is not clear that this suggestion will contribute anything to a simulationist account of correct commonsense psychological explanation. Let us suppose that one of the neurological events that occur in the causal chain between the belief and the desire about Jackson and the behavior is a neurological even which happens to be a crazy intention to go to London just for the hell of it. In this case the explanation that Oprah went to London because she believed Jackson was there and she wanted to interview him is false even though the neurological events that underpinned that belief and desire were part of the causal chain whose final link was the event that was her going to London and even though those events are subsumed by a strict causal law under a non-intentional description. The fact is that whereas the belief, the desire and the intention were all causally related to the decision to go to London, only the intention to go just for the hell of it played an explanatory role in relation to that decision.3 So the states referred to in our folk psychological explanations having another description by virtue of which they are subsumed by causal law is not sufficient to account for the correctness of those explanations. What we need to know is what it is about an intentional notion that entitles it to occur in a correct becausestatement. That is, we need a condition on explanation that weeds
126
ANGELA ARKWAY
out what it is that is explanatorily relevant about a state or event in relation to another state or event from what it is that is explanatorily irrelevant. It does not appear likely that a simulation-based condition will be up to performing this task. For even if the explainer were to succeed in simulating all of the specifically-weighted inter-related complex of mental states that is causally relevant to Oprah’s behavior it is no more likely that the intentional states selected by the explainer will be those that are actually explanatorily relevant to Oprah’s behavior than any of the others that are causally related to the decision to behave and that he has successfully simulated. This will be the case even if, and maybe especially if, he were to apply Gordon’s principle of least pretending to the nexus of correctly-simulated causally-relevant states. Let us review the situation. Satisfaction of the conditions at (SE) is unlikely to be necessary or sufficient for correct robust folk psychological explanation of behavior. First, if what (robustly) explains human behavior is something to do with the decisionmaking mechanism then it seems that it will be something to do with what is occurring in the behaver’s decision-making mechanism, not in the explainer’s. Second, if the simulation account is to replace the theory theory, a construal of simulation that does not collapse into theory deployment is needed. A common construal of simulation relies on the entertaining of hypotheses in the imagination about the contents of the mental states of another. The form of the input that this construal presupposes together with the posited isomorphism between the processing in the simulator and the person simulated leaves this construal of simulation open to the objection that the processing mechanism embodies tacit knowledge of the laws of a theory and thus incurs the threat of collapse. The construal of simulation as imaginative identification of the behaver by the simulator evokes, absent an argument to the contrary, the need for psychological causal laws external to the system to account for the causal relation of the psychological states to each other and to behavior. I argued that there are good reasons to believe it unlikely that there are laws of a type that can play this role. In the next section I discuss a proposal for robust simulation-based explanation that does not appeal to psychological laws.
THE SIMULATION THEORY
127
V. SIMULATION AND MAKING SENSE OF BEHAVIOR
Recently there has appeared in the literature on the debate a suggestion for a type of theory which, its proponents claim, could be construed as a variation on the simulation theory theme. Martin Davies and Tony Stone (1996, p. 136) point out that the type of theory they have in mind stresses both the importance of first order thought about the world and the idea of a distinctive type of explanation, that of “making sense of another person.” They describe how, on this view, we come to a decision to act in our own case: When I am considering how to act in a given situation, I bring to bear my knowledge about the world, and arrive at a judgement about what is the thing to do. Both knowledge and imagination are certainly drawn upon, but there need be no intrusion into my own decision taking of any body of empirical theory about psychology – about what people in certain situations and with certain propositional attitudes generally tend to do. (p. 136)
They point out that the same kind of normative judgment is relevant when we explain or understand the behavior of another person. They agree with John McDowell (1985) that [T]he concepts of the propositional attitudes have their proper home in explanations of a special sort: explanations in which things are made intelligible by being revealed to be, or to be approximate to being, as they rationally ought to be. (p. 389)
Davies and Stone hold that this type of theory clearly differs from the theory theory: the only empirical assumption involved here is that people will generally do the sensible thing and this does not amount to the body of psychological knowledge that theory theorists envisage. They point out that this type of theory also differs from the version of the simulation theory in which simulation is construed as imaginative identification. There is no special role here for that notion in the transition from the first to the third person, the normative judgment not being essentially a first person judgment. However, they remark, it may still count as a variation on the simulation theme according to Stich and Nichols’ way of drawing the “battle lines.” In Stich and Nichols’ view (1992, p. 47) any theory that involves processing in the decision-making mechanism when it is disconnected from its natural inputs and from the action controllers
128
ANGELA ARKWAY
(“off-line” processing), even if that processing follows the contours of a theory, counts as a version of the simulation theory. The strategy described by Davies and Stone does involve some “offline” processing. The explainer imagines the circumstances in which the decision to behave was made and then feeds these pretend “facts” into his practical reasoning mechanism along with his (relevant) knowledge about the world and the beliefs that constitute his knowledge about the sensible thing to do in different sets of circumstances. The behaver’s processing of these “facts” is then simulated in the explainer’s decision-making mechanism and a pretend decision-to-behave is generated. Since the mechanism is disengaged from the action controllers, no behavior ensues. Some theorists will disagree that this type of theory belongs in the simulationist camp. Simulation is mental simulation, they will point out. Its basic premise is that in our folk psychological practices we exploit the innate ability to simulate the mental goings-on in another or in ourselves at a time remote from the present. Simulationist Jane Heal however, would simply respond that this objection misdescribes the direction of gaze. A simulator looks not at the subject to be understood but at the world around that subject: “It is what the world makes the replicator (simulator) think which is the basis for the beliefs he attributes to the subject” (1986, p. 48). Heal, Davies, and Stone agree that psychological understanding of other people, including giving some sort of answer to a whyquestion, needs only one simple empirical assumption: that “they are like me in being thinkers, that they posses the same fundamental cognitive capacities and propensities that I do” (Heal, 1986, p. 47; Davies and Stone, 1996, p. 136–137). In Heal’s view, another way to express the truism that “in giving a psychological explanation we render the thought or behavior of the other intelligible, . . . ” is to say that we see them as “exercises of cognitive competence or rationality,” terms which she considers interchangeable and which should be understood very broadly to mean what is exercised in the formation of intention, desire and belief (1986, p. 52). Heal points out that although the notion of rationality or cognitive competence is central on the replication account, no substantive definition of it can be given.
THE SIMULATION THEORY
129
Heal (1986), Davies and Stone (1996, p. 136) are engaged in describing the strategy people use in giving explanations. They do not address the issue of whether the psychological explanations produced by the folk are explanations of the occurrence of the thought or of the behavior in question in the same sense that scientific explanations explain the occurrence of the events they subsume. I argued earlier that the simulationist challenge to the theory theory includes an implicit claim to account for the same sort of explanation-giving that the theory theory claims to account for. If I am right about this, then, since on the theory theory account folk psychological explanations are robust explanations, explanations that “make sense of another person” will constitute a challenge to the theory theory only if they are robust explanations also. Let us look at a suggestion for the conditions that are satisfied by a correct simulation-produced folk psychological explanation of this type: (SE)∗ The statement that Philip smoked a cigarette because he believed that p and desired that q is a correct explanation of his smoking a cigarette iff (a) he smoked a cigarette, (b) he believed that p and desired that q, and (c) a decision to smoke a cigarette is the result of a simulation of Philip run on the decision-making mechanism of the explainer and that simulation satisfies the following conditions: (i) the belief that p and the desire that q emerge from the simulation as the belief and the desire whose instantiation in circumstances similar to those that obtained when Philip decided to smoke a cigarette would make that decision intelligible to the explainer (ii) no other belief and desire emerge from the simulation as the belief and the desire whose instantiation in those circumstances would make the decision to smoke a cigarette more intelligible to him. Heal would be the first to point out that there are correct folk psychological explanations of behavior that do not satisfy these or any other simulation-based conditions. She says that “. . . the only cases that a simulationist should confidently claim are those where (a)
130
ANGELA ARKWAY
the starting point is an item or collection of items with content, (b) the outcome is a further item with content, (c) the latter content is rationally or intelligibly linked to that of the earlier item(s)” (1996a, p. 56). She does not say that it is impossible that there should be psychological simulation involving non-content or non-intelligible linkages, but just that there should be no priori assumption in favor of the existence of such cases. Limiting the domain of simulation to these cases however raises doubts about the role of simulation in accounting for commonsense psychological explanation. Although it seems true that some behavior has a rational explanation and it seems possible that these cases are taken care of by simulation, it also seems true that there are many commonsense psychological explanations that account for behavior for which there is no rational explanation, behavior that cannot be made sense of. So even if simulation were to account for the rational ones there would remain an important body of commonsense psychological explanations to be accounted for by a non-simulation-based theory. There is no reason to believe that an account that subsumes psychological explanations of non-rational behavior will not subsume rational explanations also. For the sake of the argument let us suppose that rational explanations are produced by simulation and consider whether satisfaction of the simulation-based condition at (SE)∗ is likely to be necessary and/or sufficient for a correct one. Let us take the case of Philip who decides to take up smoking and smokes his first cigarette while working on his dissertation. The correct commonsense psychological explanation of Philip’s smoking is “Philip smoked because he wants to finish writing his dissertation and he believes that he writes more easily while smoking.” Philip’s case clearly qualifies as a candidate for Heal’s type of simulation in that his belief and his desire are contentful states that are intelligibly linked to each other and to his decision to smoke. The next step is to examine what would be the case if this correct explanation were to satisfy the simulation condition (c) at (SE)∗ . Heal points out that recent thought in moral psychology and in the philosophy of action more generally has developed a view about the conception of desires, emotions and intentions which affects how we should conceive of simulating these states (1996a,
THE SIMULATION THEORY
131
p. 59). This conception emphasizes the links between the whole motivational and affective side of our nature and the concept of the valuable: the desire that p should be understood as being closely linked to conceiving that it would be valuable in some way, enjoyable, health promoting, just, if p. In a similar way emotion connects closely with the recognition that something conceived as valuable exists, or has been destroyed, or is threatened or is to be hoped for, and intention connects with the judgment that the intended action is a way of promoting or defending some value. Heal points out that since phenomena like akrasia, depression and overreaction show that the strength of motivation or feeling can get out of line with what is rationally licensed by the associated judgment, desires, emotions and intentions cannot be identified with these value judgments. However, she adds, this does not indicate that there is more to simulating a desire, an emotion or an intention than entertaining the content of the associated value judgment. She points our that “[A]s far as rationalizing and making intelligible are concerned it is the value judgments that do the work” (p. 60). On this view then, the explanation of Philip’s smoking will satisfy (c) at (SE)∗ iff the value judgment attached to his desire to finish his dissertation, i.e. that finishing his dissertation would be a very good thing indeed, is correctly simulated such that this desire emerges from a simulation of Philip, together with the belief that he writes more easily while smoking, as the belief and the desire that make best sense of Philip’s decision to smoke in the circumstances in which he actually made that decision. Heal (1996b) suggests that what is responsible for our success in simulating is the remarkable cognitive machinery with which we are endowed. This machinery empowers us to pick out from the volumes and volumes of our own world view the factors relevant to any given problem, and this not only in the sense that we can instantly provide answers to extremely diverse questions but in the more important sense that if there are configurations of information in the total view that make something other than a stereotypical answer proper to the question in hand then that information tends to become prominent to us and to influence our judgment appropriately. (p. 80)
What is more, she says, we can apply this remarkable machinery to the world view of another and “extract from it the thoughts relev-
132
ANGELA ARKWAY
ant to answering a particular question . . . ” (p. 85). She suggests that our success in this is due to the fact that in attempting to answer that question the other’s “. . . remarkable machinery is doing exactly the same (as ours), namely sieving through varied contents to find and use the relevant ones. Each of us is relying on his or her understanding of the question and of the content of the world view to drive forward the thought process which delivers the answer” (1996b, p. 85). In dealing with others, she says, in particular in predicting their thoughts “we take account of the fact that they have the ability to cope sensibly when other things are not equal or circumstances are not normal” (p. 81). Heal’s claim about the role of simulation in our folk psychological practices relies on two assumptions: first, that our cognitive machinery is such that we have the capacity to simulate successfully the rationally-linked contentful states that another instantiates on a specific occasion and, second, that this cognitive machinery/capacity is relevantly similar or isomorphic in human beings. It is the posited relevant similarity of this competence that I take issue with below. It seems that even if it is true that human beings are cognitively competent in a way that allows some sort of simulation, it does not follow that the cognitive competence in question will be isomorphic in human beings, or even similar enough to account for the role that Heal attributes to simulation in our folk psychological practices. Let us see what sort of similarity would have to obtain if one of us were to succeed, via simulation, in picking out from another’s world view the thoughts relevant to answering the particular question “Why did I (i.e., the other) behave in that way on that occasion?” One possibility is that the similarity consists in the fact that the cognitive mechanism incorporates some sort of norm of what it is for one thought to be rationally connected to another and what it is for a thought to be rationally linked to behavior. However it does not look as though our being similar in this respect will do the job. If we imagine the ideal case in which a potential explainer, thus equipped, succeeds in simulating all of the relevant contentful states that a similarly-equipped behaver instantiated on the occasion of his decision to behave, satisfaction of (c) by a simulation will still not be sufficient for a correct explanation to emerge. What processing
THE SIMULATION THEORY
133
according to a norm might extract from the complex of correctlysimulated mental states are those which, among all the ones the behaver actually instantiated on the occasion of his decision to behave, are the most likely to be instantiated by a rational behaver in relation to the kind of behavior in question in circumstances of the type in question. There is no reason to believe that these are what actually account for the behaviorial episode at issue. The posited isomorphism between the cognitive mechanisms is simply too weak to support such a suggestion. Let us see what would be the case if we construe the similarity in cognitive competence as strong to the degree that it allows the rationally-linked contentful states relevant to a specific behavior on a particular occasion to emerge from a simulation more often than not. The issue about cognitive similarity then becomes a question of the degree of similarity necessary to account for the number of our folk psychological explanations that are (apparently) successful. The problem is that this degree is likely to be implausibly high. The states responsible for someone’s behavior (or decision to behave) will not emerge from a simulation performed by a would-be explainer whose cognitive mechanism is only similar enough to the behaver’s to correctly simulate the content of his rationally-linked states. The degree of similarity sufficient for content-simulation will not be sufficient for selecting those states which, among the correctly-simulated rationally-linked ones, are those without whose instantiation the behaver would not have decided to perform the behavior in question. A minimum additional requirement for these states to emerge from a simulation is that the explainer simulate correctly the ranking in the value judgments; attached to the behaver’s desires, emotions and intentions and the hierarchy in the strength of the (relevant) beliefs that obtained at the moment the behaver made his decision to behave. To return to the case of Philip, if the desire to finish writing his dissertation and the belief that he writes more easily while smoking are to emerge from a simulation of Philip as those that are explanatorily relevant to his decision to smoke, then the simulator will succeed in at least the following tasks. He will simulate both the content of the value judgment attached to Philip’s desire to finish his dissertation, viz., that finishing it would be a very good thing
134
ANGELA ARKWAY
indeed, and its ranking in relation to the value judgments attached to any other state in the inter-related complex of relevant states that Philip instantiated at the time he made the decision to smoke, viz., that finishing his dissertation ranked higher for Philip in the circumstances in which he made the decision than, for example, the value judgments attached to his firm intention not to expose his two-year old to the noxious effects of second-hand smoke, to his deep-seated fear of becoming addicted to nicotine and to his standing desire to increase his chances to live a long and pain-free life. The simulator will also simulate successfully the degree of strength carried by Philip’s belief about smoking’s effect on his writing in relation to his other beliefs about writing and about smoking. It is clear that the kind of simulation involved here simply falls out of the domain of the type of simulation endorsed by Heal. In describing the view she refers to as epistemological holism Heal points out that “no thought, whatever its subject matter, can be ruled out a priori as certainly irrelevant to a given question” (1996b, p. 79) and that our reasonings and memories are open “to influences from an immense variety of sources such as emotion, mood, subliminal suggestion, etc. etc.” (1996a, p. 52). If this is true, it follows that the degree of value ascribed to the judgment linked to a particular desire, emotion or intention on a specific occasion in relation to the degree of value ascribed to judgments linked to other desires etc., as well as the degree of strength attributed to a belief on that occasion in relation to other beliefs is likely to be a function of the interaction of a large number of mental states. It is hard to deny that the degree of similarity between cognitive mechanisms that would allow successful simulation in these circumstances is just unlikely to obtain.4 What the preceding discussion shows is that it is no more probable that correct robust folk psychological explanations that make sense of behavior will satisfy a simulation condition than those that claim to explain behavior by entertaining hypotheses about the content of another’s mental states or by identifying with another in the imagination. Attempting to account for robust simulation-based explanation without appealing to psychological laws seems to lead to the postulation of a degree of similarity between the explainer and the bahaver that is implausibly high. This is not to deny that
THE SIMULATION THEORY
135
our remarkable cognitive machinery might be such that it helps us pick out the correct explanation often enough to account for our successful interaction. However simulation being a useful heuristic device does not entail it being either necessary or sufficient for a correct rationalizing explanation.
VI. CONCLUSION
In conclusion it remains to say how this discussion has contributed to the debate between the theory theory and the simulation theory. I have argued that if the simulationist challenge to the theory theory is a claim to be a better account of the mechanism that underpins the full range of our folk psychological practices, then simulationists need a story about folk psychological explanation. The mechanism posited by the theory theory is supposed to underpin the giving of robust psychological explanations of behavior by the folk. In this paper I have examined the most likely candidates for robust simulation-based folk psychological explanation and I have shown that none of them will do. I suggest that besides continuing the search for other candidates we should consider the possibility that what is at the root of the simulationist challenge to the theory theory is more than a disagreement about the nature of the cognitive mechanism that underpins our practices. What may be driving simulationists is the intuition that folk psychological explanations of behavior are not robust explanations at all but are explanations in the sense that they provide some sort of understanding to the explainer. If this is so then, just as it was for R.G. Collingwood, W. Dilthey, and the Verstehen theorists who argued this case against the logical positivists, it is simulationists who have the burden of proof.
NOTES ∗
I am very grateful to Christopher Gauker, Robert Gordon, Donald Gustafson, William Morris, Jenefer Robinson, Stephen Schiffer, and Stephen Stich for their helpful comments on earlier drafts of this paper. 1 I am indebted to Hartry Field for his suggestions with regard to this formulation. 2 It is worth noting that Schiffer’s target in this paper is theorists who appeal to commonsense psychological ceteris paribus laws to account for the truth of folk
136
ANGELA ARKWAY
psychological explanations, i.e. the type of theory theorist referred to earlier as holding that folk psychological explanations are subsumed by the covering law model. So if Schiffer is right and no such laws exist, then this version of the theory theory will lose its account of explanation and, at the same time, an important advantage over the simulation theory. 3 The importance of this point was impressed upon me by Stephen Schiffer. 4 If I am right about the high degree of similarity needed for correct explanation, the question arises of whether or not this is likely to pose a problem for a simulationist account of prediction. One might argue that if successful simulation were indeed required for the correctness of our predictions of behavior and if successful simulation did call for the high degree of similarity described here, there would be few, or no, correct predictions of behavior. But, one might continue, since there are impressively many successful commonsense predictions of behavior, such predictions cannot be the result of simulation. I hesitate to endorse this conclusion without reservation. The reason is that there are several different types of mechanism that the simulation theory could be construed as describing. For the purpose of attempting to substantiate its challenge to the theory theory I assume, in this paper, that the simulation theory describes the same sort of mechanism that the theory theory describes, a causal mechanism. If the simulation theory is construed as describing a different sort of mechanism, perhaps an heuristic device, there may well be some sort of role for simulation to play in our folk psychological practice of prediction as well as, (as I point out in the last paragraph of this section), in relation to the practice of explanation.
REFERENCES Davies, M. (1994): “The Mental Simulation Debate”, in C. Peacocke (ed.) Objectivity, Simulation and the Unity of Consciousness: Current Issues in the Philosophy of Mind: Proceedings of the British Academy, 83, pp. 99–127. Oxford: Oxford University Press. Davies, M. and Stone, T. (1996): “The Mental Simulation Debate: A Progress Report”, in P. Carruthers and P. K. Smith (eds.) Theories of Theories of mind, pp. 119–137. Cambridge: University Press, Cambridge. Fodor, J. (1987): Psychosemantics. Cambridge: MIT Press. Fodor, J. (1990): “Making Mind Matter More”, A Theory of Content and Other Essays, pp. 137–159. Cambridge: MIT Press. Goldman, A. (1989): “Interpretation Psychologized”, Mind and Language 4, 165–182. Gordon, R. (1986): “Folk Psychology as Simulation”, Mind and Language 1, 158–171. Heal, J. (1986): “Replication and Functionalism”, reprinted in M. Davies and T. Stone (eds.) Folk Psychology: The Theory of Mind Debate, pp. 45–59. Oxford: Blackwell, 1995.
THE SIMULATION THEORY
137
Heal, J. (1996a): “Simulation and Cognitive Penetrability”, Mind and Language 11, 44–67. Heal, J. (1996b): “Simulation, Theory, and Content”, in P. Carruthers and P. K. Smith (eds.) Theories of Theories of Mind, pp. 75–89. Cambridge: Cambridge University Press. McDowell, J. (1985): “Functionalism and Anomalous Monism”, in E. Lepore and B. P. McLaughlin (eds.) Actions and Events: Perspectives on the Philosophy of Donald Davidson, pp. 387–398. Oxford: Blackwell. Schiffer, S. (1991): “Ceteris Paribus Laws”, Mind, 100: 1–17. Stich, S. and Nichols, S. (1992): “Folk Psychology: Simulation or Tacit Theory?”, Mind and Language 7, 35–71.
Department of Philosophy New York University New York, NY 10003 USA