Keywords: attribution; empathy; eliminativism; simulation; theory-theory
Contents:
The simulation theory (ST) is an account of our everyday ability to make sense of the behavior of others. One crucial element of this ability is the identification and attribution of inner mental states that generate action, especially propositional attitudes such as beliefs or desires. The successful 'mindreading' of mental states allows us to predict and to explain what others do, and makes possible the rich social dynamic that pervades human life.
Conceived most broadly, ST maintains that one represents the mental activities and processes of others by mental simulation, i.e., by generating similar activities and processes in oneself. For example, one anticipates the product of another's theoretical or practical inferences from given premises by making inferences from the same premises oneself. In more complex simulations, one imaginatively adopts the circumstances of the target and then uses one's own mental apparatus to generate mental states and decisions. Computationally, this exercise of imagination is usually conceived as requiring the ability to feed pretend inputs into one's own decision-making processes as well as the ability to take these processes 'off-line,' so that they do not issue forth in real behaviors.
Some proponents of ST go further to claim that many of the mental state concepts we deploy in understanding other human beings are fundamentally linked to our possession of those same mental states. Some prominent accounts attempt to shed light on the conceptual transformation involved in retooling our first person concepts in such a way that they can be deployed in the third person.
While ST is related to empathetic or verstehen approaches to explanation in the social sciences that have been prominent in the 20th century, most researchers who currently work on ST do not operate directly within that theoretical framework. In its contemporary forms, ST has often been developed with its principal rival, the theory-theory (TT), as an explicit foil. The TT maintains that the mental terms and concepts used in understanding human behavior get their predictive and explanatory credentials by being embedded in a folk theory of the mind. A more proximal historical source for ST came from debates in the philosophy of mind over the status of this putative theory. Eliminativists argue (i) that mental states like beliefs are desires are the posits of a folk theory of the mind, and (ii) that this theory is radically false. The conclusion drawn by the eliminativist is that the faulty folk theory ought to be rejected in favor of some more scientifically respectable theory such as one derived from neuroscience.
Before research on ST, most of the critics of eliminativism focused on the defensibility of (ii). Important articles by Jane Heal (1986), Robert Gordon (1986), and Alvin Goldman (1989) challenged (i) by setting out the ST alternative to the theory view. If it could be shown that mental terms did not derive their intelligibility from their role in a folk theory, then the eliminativist's conclusion would become suspect. Since this important impetus, ST has provoked interest in its own right and has become somewhat independent of concerns over eliminativism.
Some philosophers believe that ST sheds light on traditional topics such as the problem of other minds, referential opacity, broad and narrow content, and the peculiarities of self-knowledge. ST has had a substantial impact on research into "theory of mind" in developmental psychology, as well as on branches of philosophy outside the philosophy of mind, especially aesthetics and philosophy of the social sciences.
In keeping with this widely endorsed strategy in cognitive science, theory-theorists claim that we possess a body of tacit knowledge that governs our judgments about the mental states of others. The theoretical posits of this theory will be mental states like beliefs and desires, and the transitions between the mental states will be described by the theory as mental processes such as inference.
There is a diversity of opinion in the theory-theory camp on the nature and origin of the putative commonsense theory. Some argue that the theory is learned, while others claim that it is innate. Theory-theorists also face an orthogonal issue regarding the modularity of the commonsense "theory of mind." Some claim that the theory is a distinct cognitive module, while others maintain that it is continuous with the system of representations that constitute theories of other, non-mental domains. What unifies theory-theorists is the view that attributing inner states and making sense of the behavior of others is carried out by a capacity that deploys knowledge encoded in a theory.
The most straightforward sense in which ST is an alternative to the theory-theory is that simulation theorists deny that our mental state attribution capacity is subsumed by a body of knowledge about the minds of others. Rather, our own mental processes are treated as a manipulable model of other minds. Such simulation would typically require indexical adjustments, such as shifts in spatial, temporal, and personal point of view, to place oneself in the other's physical and epistemic situation insofar as it differs from one's own. One may also compensate for the other's reasoning capacity and level of expertise, if possible, or modify one's character and outlook as an actor might, to fit the other's background and behavioral history. With these adjustments, the attributer might enter mental states that differ from those she would have were she herself actually in the target's situation. Even when insufficient for making decisions in the role of the other, simulation might allow one to discriminate between action options likely to be attractive to the agent and those likely to be unattractive. Accordingly one would be prepared for the former actions and surprised by the latter.
Moreover, most simulationists are happy to grant that, in some cases, we will develop general rules of thumb or heuristics for attributing mental states. These may be called on to generalize the results of a simulation to cover the same target at future times or a class of targets, such as those who share the conventions of a particular culture, that "as a rule" do x in circumstances c. Still, simulationists deny that general, 'theoretical' considerations play a fundamental part in attributing mental states. ST is often characterized as 'process driven' because it is a cognitive process that is generating the output of the simulation, with little or no influence by general information about minds.
An analogy may be of help here. If we wished to predict a future state of the solar system, we might appeal to a theory that expresses law-like generalizations about the motion of the planets. So, we might appeal to the theories articulated in a contemporary astrophysics textbook or, more in keeping with our interest here in folk theories, we might appeal to a set of platitudes about the motion of celestial objects of the kind articulated by the ancients. This would be a perfectly sensible approach, but it does not exhaust the ways we could successfully carry out the prediction. If we could build a reasonably accurate physical model (such as an orrery), we could advance this model the correct number of cycles and read off it a future state of the solar system. Depending on how versatile our model was, we might even be able to experiment with counterfactual starting states; or, with a still more sophisticated model, or a digital simulation of one, we might even adjust the number and the orbits of planets to model a wide range of planetary systems quite unlike our own.
This gives us some insight into the difference between a theory driven account of a domain and a process driven account of that domain. ST can be viewed as the proposal that we use our own mental states and processes the way that the orrery is being used, above. Of course, there is at least this important difference: In the case of ST, it is our own perceptual, cognitive, motivational and emotional systems that are used as the model.
It is obviously a desideratum that any relevant disparities between simulator and target be removed or offset in some way. If the behavior to be predicted or explained is crucially dependent on beliefs, desires, or emotions not shared by the simulator, then it is crucial that the simulator either adopt the appropriate pretend-beliefs, -desires, or -emotions, or compensate by a heuristic rule. However, this desideratum says nothing about the nature of belief, desire, or the various emotions. It is consistent with but certainly not committed to, for instance, a functionalist account of belief. In a well-known box diagram, Stich and Nichols (1992) portray ST as the empirical hypothesis that the same belief-desire system that generates one's own decisions and actions also generates our predictions of the decisions and actions of others, adding for this purpose a pretend-belief generator and a pretend-desire generator. Some simulationists find this portrayal too restrictive, committing ST to a questionable conception of mental states and possibly also a mistaken understanding of the dependence of actions on these states. These proponents of ST conceive our everyday ascriptions of belief and other mental states as part of an explanatory enterprise quite unlike the attempt to fill in a "boxological" theory like those of cognitive science. Our ascriptions specify the "internal states" of a system only in the sense of attempting an essentially first person glimpse into a subject looking out onto the world.
There is also diversity among simulationists on another front. Some proponents of ST hold that to ascribe mental states to others by simulation, one must already be able to ascribe mental states to oneself by introspection, and that to do this one must already possess the relevant mental state concepts. On this view, simulation is understood as essentially an application of the argument from analogy. A rather different approach attempts to build on the subject-looking-out-on-the-world idea of mental state ascription. Those who accept it hold that in such ascriptions, whether concerning oneself or another, one is saying something about the world, albeit in a way that is relativized to a particular "point of view." Rather than resting on an analogy between what lies "inside" one individual and what lies within another, this account rests on the assumption that, unless there is evidence to the contrary, all subjects look out on one and the same world.
A number of arguments have been put forward in favor of the simulation theory, including the following:
1. Parsimony
The most important distinguishing feature of folk psychology, according to many writers, is the central and essential role it gives to the semantic content of the states it posits, particularly the propositional or sentential "objects" of propositional attitudes such as beliefs, desires, and intentions. Most theory theorists try to accommodate this feature with the hypothesis that folk psychology comprises laws or principles that quantify over this content, connecting, for example, what x believes and what x desires to what x chooses to do. Moreover, the connections are said generally to mirror the semantic relations that hold among these contents, particularly relations that can be represented abstractly by rules of logic and rational argument such as modus ponens and the practical syllogism. Thus the theory theory posits an internal store of causal laws corresponding to these rules.
However, insofar as the store of causal generalizations mirrors the set of rules our own thinking typically conforms to, the simulation theory appears to render it otiose. For whatever rules our own thinking typically conforms to, our thinking continues to conform to them within the context of simulation, unless, of course, adjustments are made to accommodate evident differences. In short, we can use our own reasoning as a model of the reasoning of beings that reason the way we do. In the light of this alternative, it is argued, the hypothesis that people must be endowed with a special stock of laws corresponding to rules of logic and reason appears unmotivated and unparsimonious.
2. Simulation has important uses other than prediction and explanation
The procedure that the ST posits as essential to the commonsense methodology for predicting and explaining behavior also appears to be - with modifications - essential to emotional empathy and important if not essential to ethical evaluation. Even if one didn't think simulation essential to commonsense explanation and prediction, one who wishes to account for empathy and ethical evaluation would probably have to posit such a procedure anyway.
3. Simulation utilizes capacities that clearly are evident in young children
Skill at sophisticated pretend play, including role playing and role exchanging, is unequivocally evident before age three. These capacities seem to many theorists to be indicative of the pervasiveness and importance of simulative abilities.
ST has enjoyed attention from a range of disciplines and methods in the cognitive sciences.
Insight into the empirical details of the developmental trajectory of mindreading has reached a high level of sophistication in psychology. There is thus consensus among both theory-theorists and simulation theorists on the mindreading capacities of young children. The ability to attribute mental states to others emerges between the ages of three and four years, and the most dramatic aspect of this transition is that children seem to adopt a representational view of mind where they begin to understand that their own mental states may differ from other's given differences in perceptual access and perspective. This is demonstrated by the children's success at the much-discussed false belief task. In standard versions of the false belief task, children are introduced to a scenario where they have full knowledge of the course of events in the scenario, but where some other participant (either another child or a doll) is only partially privy to the scenario. The lack of full knowledge of the goings on of the scenario should produce a false belief on the part of the non-subject participant. Before age three, children fail to attribute the false belief to the other participant, even though experimental controls show that they understand the task and have the linguistic ability to succeed at it if they conceptualized false beliefs the way adults do.
The majority of developmental psychologists who carry out this research are theory-theorists, so the usual explanation for the change in performance on the false belief task is that children at age 3 are in the process of acquiring a theory of mind that enables them to successfully attribute beliefs. Simulationists have proposed a new interpretation of the false-belief task results by claiming that it is the ability to simulate that is being refined. A related area of research investigates the possibility that one of the capacities that is impaired in autism is the ability to mindread. Research on autistic children indicates a pattern of failure on the false belief task that is not observed in mentally retarded children without autism. For the most part, false-belief task research is now seen as neutral between the theory-theory and ST. These debates were important, however, in highlighting the potential prospects for investigating ST from empirical perspectives.
Consequently, there has been significant research on ST within artificial intelligence, neuroscience, and cognitive psychology. In order to show how ST has been pursued in these fields, it is useful to develop further the "process driven" conception of ST introduced earlier. One way of refining the claim that simulations are process driven is to take ST as the hypothesis that essentially the same set of mechanisms is called upon to provide two different competences. One of is the intelligent control of behavior. This would include, among other things, the capacity to make inferences from beliefs to new beliefs and the capacity to make decisions on the basis of beliefs. The second competence is the anticipation and comprehension of intelligently controlled behavior, by anticipation of the underlying inferential and decision-making processs.
There are thought to be testable consequences of this 'double-duty' mechanism construal of ST. Researchers in artificial intelligence have claimed that the same models that provide an account of practical reasoning (and of other dynamics between the propositional attitudes) can efficiently be adapted to perform mental state attribution. These computational efficiency arguments have in turn sometimes been used as arguments in favor of ST. Both traditional AI programming methods as well as neural network techniques have been employed in research on ST along these lines.
In AI research, the 'double-duty' hypothesis is a functional claim. The idea is that the same mental program is implicated in both practical reasoning and in mental state attribution, with changes allowed for taking the system 'off-line' or feeding in pretend states. So, even if it turned out that different parts of the physical computational system were responsible for practical reasoning and mental state attribution, this would not jeopardize the AI approach to ST.
If, on the other hand, the double-duty conception of ST is understood not solely functionally and includes a commitment to the commonality of the underlying neural substrate, we arrive at another avenue of cognitive science research on ST. Such research would seek to show that there was a shared neuronal mechanism that is responsible for practical reasoning and mental state attribution. This approach is analogous to a line of research that has been fruitfully pursued on vision and visual imagery. These are two capacities that appear to share substantial portions of the underlying neural architecture.
There is emerging evidence for the existence of mirror neurons in both non-human primates and humans (Fadiga et al (1995); Rizzolatti et al (1996)). Single-cell recordings in macaque monkeys and magnetic stimulation techniques in human beings show that these mirror neurons exhibit increased activity when another primate is observed performing some characteristic action, such as grasping an object. This is relevant for ST because these are the same neurons that show increased activity in the first-person performance of that action. This research is far from showing that very sophisticated cognition is subsumed by double-duty mechanisms, but it is suggestive. Some simulation theorists maintain that this research reveals an evolutionary precursor to the perspective taking capacity that ST requires (Gallese and Goldman, 1998).
Evolutionary considerations have been lurking in the background of much empirical research in this domain. Thus elements of cognitive ethology have been thought to bear on ST, and conversely. Indeed, it was in their research on primate behavior that Premack and Woodruff (1978) introduced the term 'theory of mind'. Though efforts to develop ST within cognitive ethology have been limited, this potentially presents another source of data.
The logic of such experiments is as follows. In order for a simulation to be successful, the operation of the attributer's mental processes must be substantially similar to the operation of the target's mental processes. The attributer does not need to know about the vagaries of human psychology because, by hypothesis, her own mental mechanism is subject to those same vagaries. One test of this version of ST, then, would be to select some surprising feature of our mental life to see whether subjects can successfully attribute mental states to others where the mental processes in question exploit the surprising feature. If the attribution fails, this suggests that either the attributer's mind does not share the surprising feature or that it is the specific lack of information about the surprising feature that is generating the failure to attribute a correct mental state. Neither of these results would be friendly to ST. The first possibility is against the very spirit of ST, while the second possibility implicates a theory of mental states.
In one attempt to pursue this line of criticism, researchers explored the counterintuitive effect (reported by Langer (1975)) where subjects demand more money in selling a lottery ticket that they had chosen over a ticket that had been given to them with no choice involved. Although there has been some worry about the replicability of Langer's initial results, the effect reported has been thought to secure something unexpected about human psychology, namely, that subjects are more attached to items that they choose than they are to identical items that they do not choose. Nichols et al (1995) asked naïve subjects to predict the outcome of the Langer experiment without actually putting them under the experimental conditions themselves. It was assumed that if in making the prediction the subjects simulated making a decision themselves under each of the two experimental conditions, they would be prone to the same surprising effect, valuing the ticket they had chosen more than the one they had not picked themselves. Thus they would predict correctly. Subjects do not predict correctly, and Nichols et al concluded that their subjects were not simulating. However, several problems with the Nichols et al experiment have been noted. A more discriminating set of experiments has been reported, with mixed results for ST (Perner et al, 1999).
These experiments are specific instances of a general empirical test of ST. In cases where our own minds operate in unexpected ways, there should be an equally unexpected attribution of mental states to others. Consider the case of attributers with psychopathologies. All things being equal, ST should predict that psychopathological patients will attribute to others similar pathological mental states and processes. It is not clear whether this prediction is borne out, and developments of these criticisms await further research.
Fadiga, L, Fogassi, L, Pavesi, G, & Rizzolatti, G (1995) Motor facilitation during action observation: a magnetic stimulation study. Journal of Neurophysiology, 73: 2608-2611.
Gallese, V and Goldman, A (1998) Mirror neurons and the simulation theory of mind-reading. Trends in Cognitive Sciences, 2: 493-501.
Goldman, A (1989) Interpretation Psychologized. Mind and Language, 4: 104-119.
Gordon, R (1986) Folk Psychology as Simulation. Mind and Language, 1: 158-171.
Heal, J (1986) Replication and Functionalism. In Butterfield J (ed) Language, Mind and Logic, pp. 135-150. Cambridge: Cambridge University Press.
Langer, E (1975) The illusion of control. Journal of Personality and Social Psychology, 32: 311-328.
Nichols, S, Stich, S, Leslie, A and Klein, D (1996) Varieties of off-line simulation. In: Carruthers D and Smith E (eds) Theories of Theories of Mind, pp. 39-74. Cambridge: Cambridge University Press.
Perner, J, Gschaider, A, Kühberger, A and Schrofner, S (1999) Predicting others through simulation or by theory? A method to decide. Mind & Language, 14: 57-79.
Premack, D and Woodruff, G (1978) Does the chimpanzee have a theory of mind? Behavioral and Brain Sciences, 4: 515-526.
Rizzolatti, G, Fadiga, L, Matelli, M, Bettinardi, V, Paulesu, E, Perani, D, & Fazio, G (1996) Localization of grasp representations in humans by PET: 1. Observation vs. execution. Experimental Brain Research, 111: 246-252.
Stich, S, and Nichols, S (1992) Folk Psychology: Simulation or Tacit Theory? Mind and Language, 7: 35-71.
Carruthers, P and Smith, P (eds) (1996) Theories of theories of mind. Cambridge: Cambridge University Press.
Davies, M and Stone, T (eds) (1995a), Folk psychology: the theory of mind debate. Cambridge, MA: Blackwell.
Davies, Martin & Stone, Tony (eds) (1995b) Mental simulation: evaluations and applications. Cambridge, MA: Blackwell.
Goldman, Alvin I. (1995), "Simulation and interpersonal utility", Ethics, 105, 709-726.
Gordon, R and Barker, J (1994), Autism and the 'theory of mind' debate. In: Graham, G and Stephens, G (eds) Philosophical Psychopathology, pp. 163-181. Cambridge, MA: Bradford.
Heal, Jane (1998b), "Co-cognition and off-line simulation: two ways of understanding the simulation approach", Mind & Language, 13: 477-498.