The current debate between theory and simulation accounts of folk psychology is in part about the nature of our ability to predict and explain people's behavior. Theory-theory accounts of folk psychology claim that the candidate explanation or prediction is found by the application of psychological generalizations. Simulation accounts of folk psychology claim that the candidate explanation or prediction is found by simulating or pretending to be the other person. Despite these differences, I will argue that both theory and simulation face a version of the frame problem. Both accounts do not explain how, from the many different possible characterizations of a person's behavior or situation, salient background conditions can be pared down to find a characterization which enables explanation or prediction. More specifically: in order to generate a prediction or explanation, a description of both the context and the behavior of the other person must be found so that I, the interpreter, have something to simulate or something to theorize about. This frame problem is prior to the application of any psychological generalizations or running of simulative transformations, and solving it will likely show that psychological competence cannot be strictly separated from a more general competence with getting around in the world.
This issue should be distinguished from the issue of whether the mental state attribution is justified by its correctly capturing a person's beliefs, desires and motivations. This epistemological question is certainly important, but before it can be answered, we must be sure that theory or simulation can actually even arrive at a candidate explanation or prediction that can then be tested for its truth. I am investigating what conditions must be met to generate this candidate prediction or explanation.(1)
Simulation and Theory.
Simulation theories hold that I generate explanations and predictions by simulating the other - by placing myself in the other's position and asking myself what I might believe or desire in a similar situation. When I simulate another, I attribute mental states to that person by pretending (in some sense of that term) that I have the same states as the other person. Several philosophers have endorsed such a theory, although its primary advocates have been Robert Gordon (Gordon 1995a, 1995b & 1995c), Alvin Goldman (1995a & 1995b) and Jane Heal (1996, 1998a & 1998b). Although they differ on many of the details, the basic contours of their ideas are relatively similar. As an example of this theory, I will look at Gordon's version of simulation.
Gordon claims that simulation takes place via an actual imaginative transformation of myself into the other person. When simulating, I do not imagine what I myself would do in the other's situation, I imagine what the other would do in her or his situation. That is, I "recenter my egocentric map" and transform myself into the other, so that even the referent of the first person pronoun 'I' changes to the person being simulated. Gordon claims that this obviates the need to transfer the mental state simulated to the other, since I already am the other in the context of my pretense. Despite the term 'transformation,' one does not have to transform oneself into the other in the sense of taking on the other's beliefs. One simply finds what would motivate you to act as another did in a similar situation. For example, if you are hiking and your friend turns around suddenly and says "go back," you would search the environment for something that would make you act like that (like a bear up ahead on the trail). Only in special cases, when the other has beliefs very different from your own do you need to "patch" simulation by pretending how you would act if you had such differen beliefs. Hence, Gordon's account effectively has two levels of simulation. In the first level, I transform myself into the other without making any changes in my own beliefs; I simply "recenter." Gordon's discussion of the hikers who encounter the bear on the trail is a case of this level. In the second level, I recenter my egocentric map and adjust my beliefs and desires. If were to simulate the behavior of someone who liked something I detested, I would have to engage in this second kind of simulation in order to find a prediction or explanation. (Gordon 1995a, 1995b, 1995c)
In contrast with simulation, theory-theory holds that a tacit theory generates predictions and explanations. In its nomological form (i.e. Fodor 1987), this tacit theory is thought to be made up of psychological generalizations that we all share as a part of our cognitive architecture. These generalizations describe inferential relations between various mental states, such as beliefs and desires, as well as relations between environmental stimuli and responses to these stimuli. One deploys these generalizations in order to find some reasonable explanation or prediction for a person's behavior. An example of such a generalization would be:
(g) If a person desires X, and knows that s/he must do Y to bring X about, then ceteris paribus, the s/he will attempt Y.
Such generalizations all contain ceteris paribus clauses, and with good reason. Suppose we fill in the belief/desire rule as follows: a person desires lots of money, and she knows that she will not have lots of money unless she kills her rich relatives, so she will attempt to kill her rich relatives. Certainly, people's behavior can be described by (g), and there are people who are even in the situation I have described, but there are very few rich relatives being killed. Human behavior is both unpredictable and sensitive to fine differences of situation in a way that rigid laws cannot capture. So we must have some way of "letting people off the hook" when a particular generalization does not hold. And, more importantly, any theory of our ability to attribute mental states must be able to work in a new and unfamiliar situation. The psychological generalizations cannot suddenly be rendered useless by an unfamiliar situation, even if applying them in a new situation requires some work. For these reasons, psychological generalizations must hold all things being equal.
Theory-theory and the Frame Problem - an Argument for Simulation?
Both Goldman (1995a) and Gordon (1995b) objected to the theory-theory because they view these psychological generalizations as incapable of doing their work. Goldman objects that the ceteris paribus clauses in folk psychological generalizations render them too vague to provide explanations and predictions. How can people using these laws tell when the C.P. clauses are actually satisfied? Just taking our belief/desire rule stated above, we can see that too many differing circumstances could intervene, so that the law is far too vague to do any work. If someone wants curry, and believes that going to an Indian restaurant will allow her to obtain it, then she will go to an Indian restaurant, unless she can't afford it, or already agreed to meet someone at the Mexican restaurant, or is drugged and won't want curry in five minutes, or . . . the list goes on.
Gordon (1995b) claims that we may indeed use generalizations like (g) but only in the context of a simulation. Knowing whether a generalization applies in a specific case requires us to decide which features of the other's environment, history, and background beliefs are salient at the moment that we are simulating. Even if a generalization I apply to another is learned from experience with that very person, as in:
(b) Bill always drops his coat on the chair by the door,
this person-specific generalization only holds ceteris paribus, and this requires the theorizer to decide when it may not hold. For example, I know not to apply the generalization if a punch bowl sits on the chair, or if Bill breaks into the house instead of comes in through the front door. But my knowledge of when to apply this generalization about Bill comes from a quite general competence with human behavior and thought, a competence that is difficult to capture through the mere addition of further and further generalizations. For any further generalization, such as:
(b-1) One does not follow one's ordinary pattern of behavior when one breaks into the house,
would also only hold in some cases, and further circumstances could intervene to make it invalid in the particular situation. We can imagine a situation in which Bill breaks into the house because he forgot his key, and when he comes in through his front window he puts his coat right on his ordinary chair as he runs to disarm the burglar alarm.
When formulated this way, Gordon's criticism of theory-theory accounts has close affinities with the frame problem. (See Pylyshyn 1987, Heal 1996) While some disagree about exactly what the frame problem is (or even if there is a single problem that goes by this name), it is certainly possible to give a general characterization of the problem as it originated in artificial intelligence. A. I. researchers attempting to build and program computers that could mimic intelligent human behavior soon faced two problems which collectively go under the name of the frame problem: 1) getting a system to have information stored in such a way that the system could access the right and relevant information at the appropriate time, and 2) getting the system to notice the salient features of the environment given a task it was required to perform.
These two problems are particularly evident in Gordon's discussion of generalizations. A theory account faces the problem of explaining how salient features of the environment or situation can aid in determining which generalizations to use in prediction and explanation. In actual practice, humans have little difficulty with predicting that Bill will not put his coat on the chair if a full punchbowl sits on the chair and also deciding that he will put his coat in the chair if just a magazine sits on the chair. We notice, without advance cueing, what rests on the chair and we know that putting a coat on the chair with the punchbowl implies the possibility of staining the coat, and that people don't want to stain their clothing. Yet imagine trying to spell out, in the detail necessary for AI, the knowledge base that makes possible this predictive competence. First, we have to get the system to learn that Bill leaves his coat on the chair, (which involves competence with things like coats, chairs, and the like). However, we also need to instruct the system to determine under what circumstances Bill would leave his coat on the chair. This would mean teaching our system how to detect when Bill is in a hurry, as well as the difference between a break-in and a regular entrance, and so on. Finally, we need to get the system to detect relevant features of the environment that would prevent Bill from leaving his coat on the chair: the absence of a chair, the presence of specific objects on the chair (but not all objects), even the quality of the chair and whether Bill's coat had mud splashed on it. In short, we need to spell out the ceteris paribus clauses that go with these generalizations and teach the system how to deal competently deciding when they are met and when they are not.
The general point that Goldman and Gordon make against theory-theory is that it cannot possibly hope to spell out our competence with predicting and explaining behavior in terms of lawlike generalizations. Gordon further claims that the theory-theory needs to explain how we determine relevant features of a particular situation, and this turns out to be very difficult for it to do if only theory-like reasoning is available. While only Jane Heal (1996) puts such criticisms of theory-theory in terms of the frame problem, it seems that all three of the leading simulation theorists have noticed some general version of this general problem.
These considerations about salience and the frame problem are further advanced against theory as part of an argument for simulation, by Gordon and Heal in particular. The problems Gordon notes in deploying his generalization about Bill can be met, he claims, if the generalization is used in the context of a simulation of Bill.
One way in which you are a better predictor than your generalization [b] is that you can, at least for a start, use your own continually updated knowledge of the actual environment (including the punch bowl), together with your own reactions to throwing your own coat on the chair in such circumstances. (1995a, p. 106)
Thus Gordon thinks that, because one uses one's own competence in deciding when the generalization might hold, one does not have to face the frame problem that one would face if one had to rely on only psychological generalizations. Heal (1996) makes a similar point: only by conceding that we use our own thought processes to predict and explain the other, can the theory-theorist solve the frame problem.
We can see this point in Gordon's example of hiking with a friend. When your friend starts back down the trail, you require no generalization to decide what is happening. Such generalizations would not work anyway: how do you decide if you should apply the generalization, "people avoid frightening objects" versus "people give up what they are doing when they are physically exhausted" unless you had already found the salient features of the situation? As Gordon tells it, it is not even necessary to ascribe pretend beliefs and desires to the other to find an explanation, so there is no problem of deciding which pretend beliefs and desires would be salient. Instead, having noted your friend's behavior, you just ask what would move you to act in the same fashion, and you find this in the environment when you notice the bear up ahead of you.
Simulation and the general frame problem.
Contra Gordon's and Heal's claims, however, simulation is not immune from the frame problem. Indeed, a version of the frame problem undercuts both simulation and theory accounts of folk psychology. To see this more general problem, consider the following example: you and a friend are riding in her car, and you notice that her car is not running properly. You are wondering what your friend will do. (Will she take it in to have it fixed? Will she ignore it? Will she act upset or nonchalant?) It seems that the frame problem does not come up, because, as with the example of hiking with your friend, you need only ask yourself what you would do in the relevant situation. (Some adjustments may be necessary if, for example, you know that your friend hates spending money on her car and you do not.) But why do I assume that the car's running poorly is a salient event in my friends life? It seems that I have already "decided" that the poorly running car counts as a significant event in that person's everyday life, in the way that blinking or simply driving does not count as significant. I have made a "cut" within the stream of human affairs, and I have seized upon a specific characterization of my friend's situation, a poorly running car, as a source of reflection about future actions, rather than some other feature of my friend's situation, such as her wearing a blue shirt or having her shoes tied. We can, of course, imagine situations in which either of these would be relevant to predicting or explaining behavior, and that is precisely the problem: why the car's condition and not the state of my friend's clothes?
It might be that I find the car's poor performance to be an important situation because I have already attributed background beliefs and desires such as: having a non-functioning car is a significant problem, one that must be faced and that also one wishes to avoid, one that can cost money, and so forth. It could even be claimed that I made this background attribution via a simulation which put those background beliefs and desires in place, and that this background attribution explains why I take this as a salient event in the other's life.
However, this response will not work. For it must then be explained how I singled out beliefs and desires about functioning cars as important, prior to having already noticed that the troubled car is a salient feature of her current situation. And it is absurd to suggest that I attributed beliefs and desires about functioning cars in the absence of any particular car problem, as if I could foretell the future. It is equally absurd to suggest that I have somehow attributed a near infinite set of beliefs and desires to my friend, at each instant, in order to notice that her poorly running car will be important.
Finally, it cannot be said that I know that the car's poor performance will be salient because it would be salient for me, since what we are seeking is precisely an explanation for how this feature became salient for me in the first place, and then further why it might be salient for my friend. Consider Gordon's example of the two hikers. He claims that the frame problem does not come up because I merely have to ask why I would turn and walk down the path. But, in fact, walking back down the path has already been singled out as a possible subject for explanation. In that situation, I have already solved the frame problem by the time I notice my friend's walking down the path.
To put this point another way: posing an example in the context of a philosophical discussion of simulation already gives us actions and situations under a description and hence within a contextual frame. This frame enables discussants to imagine that they can easily develop a prediction or explanation. So, all of Gordon's examples function by offering us a description of a person's behavior and a situation that then frames the simulation to take place. Furthermore, this explains in part why the frame problem does not seem to affect simulation. It is not just because I can easily imagine what I would do in a given situation, as when my friend turns around on the hiking trail or when her car breaks down; it is because the situation and the activity - the contextual frame - has already been provided. Having already characterized a situation, the advocate of simulation can simply appeal to our intuition that we would know what to do given that situation. The simulation advocate is of course right that we would know what to do given a situation, but this is simply because the important work of characterizing that situation has already been done.
I can give a more systematic characterization of this issue, one which shows that it touches upon both simulation and theory.
In order to develop a prediction or explanation of another person's behavior, I need to have two things: a description of the other person's situation, and a description of the other person's behavior. By 'situation' I mean to capture those things which impact the person's current activity, such things as the other person's immediate physical environment, features of the other's personal history, or the other person's current goals and projects. 'Situation' cannot be given a more crisp definition than this, because the frame problem teaches us that there are many (perhaps infinitely many) features of a particular slice of space-time that can be singled out, and what we need when we are making a prediction or explanation are those features relevant for making a prediction or explanation of the other person. For example, while the weather and my friend's malfunctioning car are both features of the slice of space-time we currently inhabit, the car is a relevant part of her situation, in a way that the current weather is not, because the car will affect her immediate behavior in a way that the weather will not. The car counts as a part of the situation or context of her activity. (We can of course imagine situations in which the weather would be relevant, and this merely proves my point concerning the frame problem: we can find the connections between features of people's situation and their behavior which enable further predictions, even though spelling out the conditions of these connections remains a very difficult task.)
'Behavior' here simply denotes what the person is doing or trying to do: driving a car, talking with me about a book, ironing a shirt. As with situations, behaviors can be given any number of descriptions: ironing a shirt could equally be characterized more generally as "getting ready for work" or more mechanically as "pushing an electric appliance across an expanse of cotton to smooth it." As with situations, the frame problem shows that it is no easy matter to explain how, among the many possible descriptions of a behavior, one is settled upon that facilitates prediction and explanation.
A description of situation and behavior are both necessary conditions for prediction and explanation because both prediction and explanation require some initial conditions to get underway. I do not just predict what someone will do without some prior state to work from: I want to know what someone will do given some situation or behavior. According to its proponents, simulation involves putting oneself in the other's place, and that requires a characterization of the other's situation; it requires that there be a "place" to put myself into. Simulation is not the transformation or projection into a featureless and generic other. If it were this, it would not be able to generate predictions and explanations of a particular individual. What characterizes an individual as such are particularities of that individual - his/her behavior and her/his situation. Similarly, a theoretical explanation is always an explanation of something. Theorizing requires something to theorize about: I theorize about what would cause another to act as they do (as Fodor puts it). Explanation and prediction, whether they happen via simulation or theory, require a characterization of situation or behavior, and this condition is not only logically prior to the act of developing a prediction or explanation, in the way I just claimed, but also temporally prior. That is, whatever cognitive architecture develops predictions and explanations - theoretical or simulative - it can only go into action after the situation and behavior are characterized. This temporal priority follows directly from the conceptual or logical priority of situation and behavior in developing explanation and prediction: if I cannot predict or explain without first having situation and behavior, then I must first solve the frame problem to predict and explain.
(As a side point, I think that situation is typically more important than behavior for prediction, whereas behavior is more central to explanation. In prediction, having a description of situation is more important, because I must decide what the other will do, given a past history, a current environment, and some event or action that acts as a catalyst. On the other hand, I must have a characterization of what the other has just done, if I am to have something to explain at all, and so a description of behavior is more important for explanation.)
But why do I say that we must have a description of the situation or behavior? I insist on needing situation or behavior under a description because the frame problem teaches us that there are many ways of characterizing a situation or a behavior. Consider my friend and her malfunctioning car. Since her car is running (just running poorly) I could truthfully characterize her situation without reference to the car's performance at all. Such a correct characterization of her situation would fail to pick up on the salient feature of her situation, her car's performance. Similarly, behaviors can be given multiple true characterizations, even intentional characterizations. My friend could be described simply as operating the controls of her car, or as driving her car, or as driving to the store, or even as going to buy some aspirin. This is because we can and sometimes do characterize behaviors by their final goal or purpose, as well as by reference to some intermediate means of achieving that final goal.
In all cases, we need to understand our ability to see situations and behaviors as connected in meaningful patterns which project definite possibilities for future actions and which engender sensible explanations of past behavior. We seek these connections when we seek salient descriptions of contexts and behaviors. From the nearly infinite available descriptions of "where someone is at" we select one or a few possible descriptions of situations and behaviors that enable predictions that track with someone's behavior and explanations that accurately capture another's motivations and beliefs.
Moreover, how I characterize the situation and behavior - which description I settle upon - will determine the prediction or explanation I develop. This follows directly from the fact that explanation and prediction depend on having a situation and behavior in hand. If I characterize the situation and behavior differently, then I will have a different matter to explain, or a different starting point for prediction, and so my prediction will be different. Returning again to my friend and her car, it will change completely what prediction I will generate if I include how the car runs along with the other elements of her situation. Similarly, if my friend is making coffee, I can think of this as making a tasty warm beverage or as satisfying a caffeine addiction, and how I view this activity will affect what prediction or explanation I will make if the electric kettle breaks and the person is unable to finish his task. Viewing it as addiction will lead me to expect a furious or frantic response, characterizing it as making something tasty may lead me simply to expect the other to be disappointed.
I conclude, therefore, that generating a prediction for, or an explanation of, another person's behavior requires generating some description of the person's current situation and behavior. Furthermore, understanding how we do this requires understanding how we solve a particular version of the frame problem, according to which the salient features of a person's behavior and situation are found and characterized. This conclusion, of course, does not amount to a claim against either theory or simulation accounts of psychological prediction and explanation. It does, however, place both of these accounts in a new context. For it now seems that neither theory nor simulation are sufficient as explanations of the folk psychological ability to understand other people. There must be some prior, "interpretive moment" in which the conditions for simulating or theorizing - the characterization of situation and behavior - are found.
Bringing the outside in.
So far, I have argued that both theory-theory and simulation theory face a general frame problem because characterizations of behavior and situation are necessary pre-condition for simulating or theorizing. This claim certainly carries implications for cognitive architecture. Among other things, it leads us to look at "where" and "when" within the process of psychological interpretation theorizing or simulating actually takes place. If my arguments are correct, it would seem that either theorization or simulation take place after a great deal of work has already been done: the interpretation of behavior and situation is already settled before either a theoretical reasoning process or a simulation takes hold of the "results" of this interpretive process and generates predictions and explanations.
Now I do not propose, in this paper, to engage in speculations about what actual cognitive "mechanisms" could enable this interpretation of situation and behavior, except to claim that whatever these "mechanisms" are, they are at work very quickly and prior to any of the architecture typically discussed in the theory and simulation debate.(2) Rather than going "down" into architecture, I will go "out" towards the world and see what background information and competencies are necessary to find a characterization of the situation and behavior that finds salient features of the world. It is my contention that grasping the other's behavior and situation requires a great deal of general knowledge and competence about the world that I and the other share, general knowledge that is not simply knowledge about the psychological states of others.(3)
In order to support this contention, I would like to look at three of the examples I have discussed to find the background knowledge at work, and then see if some summary conclusions about the character of this background knowledge can be given.
In the case of my friend and her poorly running car, I need to be able to perceive, with my ears and eyes, that the car is running poorly. If I think that her poorly running car will be a significant event, this in part because b/c I know that cars are vital to succeeding day-to-day in modern America. I understand the place of cars in the fulfillment of our needs, tasks, purposes, career and so forth.
In the second example Bill comes through the door with the coat in his hand and I want to know whether he will put it on the chair. Again, I need to perceive the current state of the chair - does it have a punch bowl on it, or some other thing? But even needing to look at the chair to see if something rests on it shows a lot of background knowledge. I only look at the chair because I have some general idea that Bill does not want his coat ruined, but this presupposes further knowledge which is not about Bill's psychological state such as: repairing/replacing coats costs money, bright red punch ruins coats, and so forth.
In our final example of the hikers the question is: why do I see my friend turning around as salient? Because, prior to this we had most likely agreed to hike to some further destination, and I know, as a very general principle, that one cannot get towards point A by moving in a direction away from point A. Again, this very general principle would not be a part of any folk psychology. (It certainly is not knowledge about psychological states, and if folk psychology is thought to hold such general propositions as this, it might as well be taken to hold just about everything.)
In all of these cases, I believe that two things come forward as necessary conditions for finding a characterization of the situation. The first and most obvious condition is a shared perceptual world. I and the other must be able to perceive the same features of our shared world, and when I notice a feature of the environment as salient, I am implicitly assuming that the other person can also notice this same feature, and often, has already noticed this same feature. The sharing of biologically given sense organs, while necessary, is itself insufficient to explain how two people can key off the same environmental features. The case of the malfunctioning car illustrates this. While I and my friend ride in her car together, there can be a variety of stimuli: the radio, the traffic, our conversation, the scenery, and finally the noises, sounds and vibrations of the car itself. When I notice that the car is running poorly, I foreground the sounds and vibrations of the car in the context of all the other noises and features, not just because of my innate perceptual abilities, but because I have experience with and knowledge about how cars should run. The background expectations and knowledge that I have about cars enables me to notice my friend's car running poorly. My friend may have similar knowledge, or she just may be very familiar with her own car, and this enables her to find, within the stream of experience, the problem with her car.
Andy Clark (1997) discusses more striking data about the importance of background context for perception. Because human eyes actually have a very small field of high-resolution, they achieve a broader field of vision by saccades - quick moves from one point of the environment to another at the rate of several a second. Yet, which aspects of the distal environment the eye picks out can be changed simply by setting perceptual tasks in advance of entering the scene. So the eyes will settle upon different objects depending on what the person is actually engaged in finding or doing. As another, non-visual example of this phenomenon, consider trained musicians judging an audition versus an inexperienced auditor. Such things as tonal quality, familiarity with musical traditions, and the ability to note key changes, all play a role in the judging, but noticing these within the stream of music requires considerable training and is not the product of mere passive perception.
Visual and auditory perception of an environment is not simply taking in the stimuli passively, but often actively "meeting" the stimuli relative to a pre-set task, training or some assumption about what one will meet in the world. So, mere sameness of perceptual mechanisms, even joint visual attention mechanisms, is not sufficient without further sharing the same general sense of which features of an environment are those that are important. Maurice Merleau-Ponty (1942) was right to claim that our tasks, projects and interests "polarize" our world into the important foreground features and the less important background.
The first condition for finding a characterization of situation and behavior is sharing abilities to single out features of an environment. These abilities cover both the capacities of sense organs and the "training" of these organs. (4) The second condition is harder to characterize, but it is a general knowledge of "how the world works and what people do in the world to get by." I and the other person share knowledge of facts about how the world that we both live in functions. These facts include such things as: going from A to B (typically) does not involve turning back towards A; red punch ruins coats; cars should run smoothly and without giving off clouds of smoke; fixing things cost money. It is hard, in fact, to give a general characterization of this background knowledge because almost anything could be relevant in determining behavior. For example, this knowledge ranges from things such as truths about "physics" (the point about going from point A to point B) to "biology" - people have various needs like satisfying hunger and so forth, to various kinds of "social" knowledge about the function and use of various objects and tools, how our contemporary society functions (just about everything requires money). All of this knowledge plays a role in determining saliency, that is, in narrowing down to the right description of situation and behavior.
If we ask why we need this background knowledge to solve the frame problem, the answer certainly has to do with the rationality of human action. Rationality in humans and in human action is a far larger topic than I can adequately deal with here. However, I believe I can elucidate a fairly obvious sense in which we are rational that would grant some insight into why background knowledge is a part of solving the frame problem.
A minimal characterization of rationality is simply the ability "to see what follows from what" and to respond to reasons given in favor of a proposition by grasping the meaning of these reasons and their relation to that proposition (see Heal, 1998b). This is certainly an ability we have, even if we do not have it in perfect form - even, that is, if we are prone to errors. However, it seems to me that we have this ability, not just with respect to propositions and arguments - not just with statements - but also with features of our natural and social world. We have an appreciation of the function and purpose of various objects in our world, what they are used for and what they can do. We know how to employ these means in achieving tasks, for example, using a car to get to work or a computer to write an essay. We further understand, within the limits of our experience, the possible consequences of various actions that we undertake, and we even have the ability to appreciate how events beyond our control may impact our own ability to reach whatever goals we have set for ourselves. Whether we succeed or fail at something, we can learn what caused our success or failure and learn from this cause how to succeed better in the future. We have an understanding of our own possible roles in our society - roles such as teacher, parent, citizen - and what these roles demand of us and what we can expect others to do given their roles. In short, we humans have the capacity to learn about our world, its rules, limits, and possibilities. Most importantly, we react to this world and to tailor our responses and action to this world as a means of achieving our goals or current tasks. The sense of rationality I am discussing here is a rationality which appreciates which consequences follow from which actions, and how we can function within these constraints and even act to change them.
Given that human action is rational in this way, understanding human action will involve reference to the facts and character of the world to which rational human action is responding. Hence, any characterization of the situation and behavior of another person will need to key off of these facts about the world if it is to have a chance of fitting that behavior within a stream of human activity that will engender predictions and explanations. Knowledge about the current state of the world and environment, which is gained through perception, and background knowledge about how the world is, are necessary conditions for interpreting another person. This conclusion may seem trivial. Obviously I use knowledge about the world when I interpret the behavior of another person. But, my point is that this knowledge is "ground level" - it is required to even get the most basic necessary conditions of prediction and explanation. And this further implies that knowing our way around others overlaps extensively with knowing our way around the world.(5)
Works Cited.
Carruthers, P and Smith, P eds. 1996 Theories of Theories of Mind. Cambridge: Cambridge University Press.
Clark, Andy 1997 Being There. Cambridge, MA: MIT Press.
Davies M. and Stone T. eds, 1995a Folk Psychology. Cambridge: Blackwell.
--- 1995b Mental Simulation. Cambridge: Blackwell.
Fodor, Jerry 1987 Psychosemantics. Cambridge: MIT Press.
Goldman, Alvin I. 1995a "Interpretation Psychologized" in Davies and Stone 1995a.
--- 1995b "Empathy, Mind and Morals" in Davies and Stone 1995b.
Gordon, Robert. 1995a "Folk Psychology as Simulation" in Davies and Stone 1995a.
--- 1995b "The Simulation Theory: Objections and Misconceptions" in Davies and Stone 1995a.
--- 1995c "Simulation Without Introspection or Inference from Me to You" in Davies and Stone 199b.
Heal, Jane 1996 "Simulation, Theory and Content" in Carruthers and Smith.
--- 1998a "Co-Cognition and Off-Line Simulation: Two Ways of Understanding the Simulation Approach" in Mind & Language: 13, 477-98.
--- 1998b "Understanding Other Minds from the Inside" in Current Issues in Philosophy of Mind, ed. Anthony O'Hear. Cambridge: Cambridge University Press.
Merleau-Ponty, Maurice 1942 The Phenomenology of Perception, trans. Colin Smith. London: Routledge and Kegan Paul.
Pylyshyn, Zenon 1987 (ed.) The Robot's Dilemma: The Frame Problem in Artificial Intelligence. Norwood, New Jersey: Ablex Publishing company.
1. I am grateful to Angela Arkway for clarifying this distinction.
2. Although I will point the reader to my earlier paper, arguing that there may not be much need for interpretation of action, because actions might be directly perceived.
3. I am grateful to Jim Garson for helpful discussion on this issue.
4. Such a condition is only necessary in cases where two people are interacting directly. I do not need to share perceptions with someone I am merely hearing a story about.
5. Writing of this paper was greatly aided by Robert Gordon's NEH Seminar on Theory and Simulation.