Beliefs, desires, and the ability to learn.
Underlying many of the conclusions I draw here is a distinction I argue for between behavior that is merely oriented toward a particular goal, and that which is directed toward that goal. Roughly, the idea behind this distinction is that some organisms are able to overcome many obstacles and achieve their goals, while for others an obstacle completely devastates any possibility of reaching the goal (unless evolution has already accounted for the possibility of that specific obstacle). I argue that this distinction rests on the structure of the representations in the various organisms: if the goal and the means to that goal are represented discretely, the organism is able to overcome obstacles (its behavior is goal-directed); if not, its behavior is goal-oriented.
This distinction illuminates the role that learning plays in allowing the behaver the flexibility to take advantage of having beliefs and desires rather than being able only to behave in a particular manner in a particular situation. I argue it is only the behaviors that are products of this flexibility (i.e., that are goal-directed) which are appropriate for folk psychological explanation. This ability to learn (relative to a particular goal), is necessary and sufficient for an actor to be able to behave in a goal-directed manner (relative to that goal). Therefore, being able to learn is a necessary condition for being a candidate for folk psychological explanation (again, relative to that goal).(3)
Because learning is central to the thesis advanced here about when folk psychological explanation is appropriate, it is important to be clear about how I use the term. Accordingly, the first section of the paper is a discussion of just that. In the second section I argue, first, for the distinction between goal-oriented and goal-directed behavior, and second, that only goal-directed behavior can be a proper candidate for folk psychological explanation. At the close of the second section it becomes apparent that goal-directed behavior is closely tied to the ability to learn; in the third and final section I examine this connection more closely, arguing for a more refined definition of learning, one which is necessary and sufficient for goal-directedness. The ability to learn in this way (relative to a particular goal), I conclude, is thus a prerequisite for folk psychological explanation of behavior directed toward that goal.
I. WHAT IS LEARNING?
A first approximation of a definition of learning is that a thing that learns knows something at some time that it did not know at an earlier time. Clearly, though, this will not do for the purpose of investigating the relationship between the ability to learn and the ability to have beliefs and desires. This definition presumes too much; knowledge is only possible where there is belief,(4) and to assume that learning essentially involves belief is to assume that only those organisms about whom folk psychology is a true theory are able to learn.
One way to focus on learning while avoiding intentional idioms is to use behavioral terminology: an agent has learned something if at some point in time it responds to a set of stimuli differently than it responded to a relevantly similar set of stimuli at an earlier time.(5) However, this definition falls short in two ways. First, it is too behavioristic. An alteration in behavior is not itself learning, it is simply evidence that the organism has learned (and not always good evidence). The learning itself could only be a change in disposition to behave; the actual change in behavior reflects that change in disposition. No change in observable behavior is necessary for there to be learning. Even so, this definition gets at something right: if there is no modification in behavior, that in itself gives some support to the hypothesis that no learning occurred; the learning that an organism does must be able to affect its future behavior, given the right circumstances. If the "learning" is unable to lead to any modification of behavior, if it is not even a change in disposition to behave, the organism cannot be said to have learned anything.(6) Learning is what causes, and explains, the change in disposition to behave.
Second, a change in response to a stimulus is not, by itself, a sufficient indication that learning has occurred: a female cat in heat will respond to a male cat significantly differently than she did when she was not in heat a short time earlier, but she certainly has not learned anything. What has actually happened is that there has been an internal physiological change in the cat so that it responds differently to the same stimulus. That is, the female cat's response is determined by her biological programming; at one time her biological programming dictated that she respond to the stimulus in one way, at another time the programming dictated that she respond in another. Similarly, there are other organisms which exhibit alterations in behavior which are not caused by learning, but by some process other than learning, such as evolution or design. When the changes in behavior are a result of some "hard-wired" disposition to change when a certain level of maturation is reached, or at a certain time of day, or time of year, etc., then they are not exhibitions of learning.(7) In order that the change in behavior be evidence of learning, the change itself must in some way be brought about by something (presumably some sort of information)(8) which had been present in that organism's environment.
This discussion can be summed up in the following motto. When it is the case that between two situations with relevantly similar stimuli (roughly speaking) an organism is presented with relevant new information,(9) and then it acts differently at the later presentation of the stimuli, we have evidence to say that it learned.
This is just the kind of evidence that leads to ordinary ascriptions of learning. The pigeon gets the information that when it pecks at a lighted circle it will get food, and afterwards it behaves accordingly--it pecks at the lighted circle. The rat is allowed to get the information that when it turns right at the first intersection and left at the second it will get to the cheese at the center of the maze, and afterwards it behaves accordingly--it turns right at the first intersection and left at the second. And so on. Of course, the learning is not limited to situations such as these; Angelina learns that bringing an umbrella with her to the beach is not necessary for satisfaction of her desire to stay dry, but it is a good idea if she wishes to avoid skin cancer. She continues to bring the umbrella, but she now utters different sentences as an explanation of her umbrella-carrying behavior. In each of these cases what licenses the ordinary ascription of learning is that the organism is presented with new information and alters its behavior accordingly.(10)
The evidence for learning in each of these cases is the change in behavior between different presentations of a similar stimulus. Of course, as noted above, this change in behavior is not itself the learning. What looks to be a more likely candidate for learning is that in each of these examples the organism is forming new pairs of representations of ends and representations of means to achieve these ends. When the pigeon learns to peck at the lighted circle it teams to associate a new means--pecking at the lighted circle--with an old end--obtaining food.(11) When the pigeon and the rat learn, in the ways described above, what they are learning is to associate a new means with an old end; when Angelina learns, she is associating a new end with an old means.
Here, then, is something that can serve as a rough characterization of learning: learning is the formation of new associations of goals and behaviors.(12)
The ability to form new associations of means and ends, or behaviors and goals, can only be present when the behavior is performed in order to achieve a goal. This is because the only thing that could cause the organism to form a new association is the failure of a behavior to completely satisfy a goal.(13) In order to understand the implications of this, it is worth examining goal-directed behavior more closely.
II. GOAL-DIRECTED AND GOAL-ORIENTED BEHAVIOR
In this section I argue that in order for an organism's behavior to be truly explicable in terms of beliefs and desires it must be actually directed toward a goal (roughly, performed in order to reach the goal), not merely oriented so that if the behavior is performed in the right circumstances the organism will achieve the goal (roughly, performed because the organism's biomechanical engineering causes it to act that way). One of the deciding factors in whether or not some behavior is goal-directed is the structure of the representations that cause the behavior. To see this, it is useful to look first at some lessons to be learned from the presumption that folk psychological explanation is sometimes correct,(14) and then to some examples of animal behavior.
The guiding premise in giving belief desire explanations of behavior is that beliefs and desires actually do cause behavior. If beliefs and desires are causes of behavior, if folk psychological explanation is at all accurate, both beliefs and desires must be present, and they must act in concert. Beliefs alone are not sufficient to cause action; they would fail to motivate the action. I may believe that I have a radio show to do today, but without any relevant desire (such as a desire to do the show) that belief is not going to get me to act. Desires alone are similarly inert. My desire to hear John Coltrane play live cannot get me to act in any way without some sort of belief about how to satisfy the desire.
In addition, because folk psychology is committed to the position that one belief (or desire) can be part of the cause of many different behaviors, there must be discrete structures which are the belief and the desire. Folk psychology implies that there is not one mental structure which constitutes both the belief and the desire, but that there is a different representation (or what-have-you) for each.(15) Consider the case in which there is but one mental structure representing both the belief and the desire: when asked why I ran from the tiger, I respond, "The tiger was hungry, and I wanted to not be its lunch." According to folk psychology we should be able to project from this state of affairs (assuming the explanation is correct) to various conclusions involving the same beliefs and desires. For example, I fed the tiger tuna fish in part because I continued to believe that it was hungry. However, if beliefs and desires are not separately constituted, there is no way in which my belief that the tiger is hungry could enter into any behavior-causing process other than the one it was already part of; it would be permanently linked to the desire to avoid being tiger-lunch. Folk psychology is committed to the thesis that beliefs and desires are individually active in many causal stories.(16) In order for this to be the case, they must be separate structures."
A. The Coastal Snail
When the coastal snail (Littorina) is upside-down and under water it crawls toward sunlight.(18) Whereas, if it is fight-side-up it will crawl away from the sunlight. However, this reversal of its phototaxis will only occur when the snail is underwater. If it is above water then it only moves away from sunlight: it has a negative phototaxis when above water, whether fight-side-up or upside-down. The coastal snail also has a negative geotaxis: it will crawl up, away from the earth, if given the opportunity. The combination of these conditionals ("if underwater and upside-down, then crawl toward sunlight," "if underwater and right-side-up, then crawl away from sunlight," "if above water, then crawl away from sunlight," and "if the opportunity presents itself, then crawl away from the gravitational pull") has the result that the coastal snail, in its natural habitat (near the water line of rocky coasts on the Mediterranean), will end up in dark, damp crevices just above the water line.
Watching the coastal snail in its natural habitat one might easily reach the conclusion that the snail is crawling in the direction it is because it wants to find for itself a dark, damp crevice. However, even when the snail is not in its natural habitat, and following these conditionals does not lead to the satisfaction of this goal, the coastal snail continues to follow the conditionals. The snail is unable to adapt when its behavior fails to lead to its goal. This should suggest that the folk psychological explanation proffered above is not an accurate explanation of the snail's movements.(19) The accurate explanation has something to do with the way natural selection has programmed the snail to move in order to take advantage of certain regularities in the snail's natural habitat.
The way natural selection prepares the snail to take advantage of these regularities is to equip it with conditionals that rigidly guide its movement. Instead of a conditional representing the relationship between the means to a goal and the goal (such as, "in order to reach a dark, damp crevice, do this"), the snail has representations merely of how to behave in a particular situation (e.g., "when underwater and upside-down, crawl toward sunlight"). The goal toward which the snail is moving is only implicitly represented in the conditionals that govern its behavior. One result of this is that if the snail is faced with a situation in which its behavior is not productive, it cannot adapt; it cannot team to relinquish that behavior in favor of other possible behaviors.
The inability of the snail to adapt its movements to a new environment, or, at minimum, to abandon its inefficacious movements, is what indicates to us that its behavior is not appropriately explained by appeal to folk psychology. This is because the inability to adapt indicates that the snail does not have separate representations of the goal and the means to achieve that goal. The behavior of the coastal snail is thus not actually directed toward a goal. Instead, the behavior is oriented toward a goal; it will reliably lead to a specific end when performed in the right setting, but this fact does not figure in the explanation of the snail's behavior. Even were these actions to cease to lead the snail to dark, damp crevices, the snail would continue to behave as it does. What causes the snail to perform these acts is not that they lead to the right sort of goal in the fight sort of circumstances, but that the servo-mechanisms which control its movements dictate that, for example, it move toward light when it is underwater and upside-down. (Of course, what caused these servo-mechanisms to be in place is that they reliably lead to the fight goal in the snail's natural environment. But this fact is irrelevant to this individual snail's behavior. This snail behaves as it does solely because it is pre-wired to do so. The ability, or lack thereof, of the servo-mechanisms to aid the snail in reaching its goal does not enter into the causal equation.)
While the goal may not figure in an immediate explanation of the snail's behavior, it does figure in a more mediate explanation: that they reliably lead the snail to dark, damp crevices is why the servo-mechanisms that cause the snail to behave as it does are present. Natural selection guarantees that the snail will move in such a way that if it starts in the fight environment it will end up in an appropriate crevice. The process of natural selection may be sufficient to orient an organism (such as the coastal snail) so that it will reliably find itself in an agreeable local environment, given that it started out in the right global environment. That is, natural selection may insure that there will be a correlation between the organism's actions and some goal state as long as the organism is in the proper environment. However, this does not make the actions of that organism goal-directed; when the actions are the result of natural selection, their performance is caused by the organism's biomechanical states, regardless of their ability to lead to the goal.(20)
In order for the goal to be causally relevant to an organism's behavior, the organism must have separate representations of the goal and the means to achieve that goal. Otherwise the organism will be unable to adapt when its behavior does not lead to the satisfaction of its goal; without separate representations (which the coastal snail clearly lacks) the actor could not abandon a particular means without also abandoning the goal toward which that means is directed.(21)
To understand the implications of this it would help to look at an example of an organism that might be goal-directed.
B. Rats and Cognitive Maps
Rats have considerably greater skill in overcoming obstacles and reaching their goals than do coastal snails. One possible explanation for their success is that rats construct a cognitive map of their environment which allows them to achieve their goals, even when this requires the ability to negotiate obstacles.(22) This map consists in a set of linked nodes (representations of landmarks) formed during the rat's exploration of its environment. When the rat is hungry, for example, the node which represents the location of food is potentiated (that is, the node is activated, increasing the likelihood of activation in the nodes connected to the potentiated node). This starts a radiating pattern of potentiation; each node potentiates those connected to it to a lesser degree. The hungry animal follows the trail represented by the nodes with the strongest potentiations to the food. The rat would eat because it is hungry, the hunger potentiates the node representing food, which potentiates the nodes which lead the rat to the food. And when the rat eats the food this results in the destruction of the motivational state; there is no hunger acting to potentiate any nodes, and so the rat stops eating and following the trail.
If the cognitive map model is correct or nearly correct, a case can be made that the rats' feeding behavior is goal-directed rather than goal-oriented. To see this we need to notice one aspect of the rats' behavior: several experiments have shown the ability of the rats to update their maps. In one the rats found, when they were hungry, that their usual (and the easiest) path to the food had been cut off. They very quickly assimilated this information and directly took the next best route. They accomplished this by abandoning one means to their goal (reaching the food) and adopting another, all the while retaining the goal.
The difference between the rats and the coastal snails is that the coastal snails are unable to relinquish some behavior without also relinquishing the goal toward which that behavior is aimed. (This is evidence that the coastal snails' behavior is merely goal-oriented.) However, the rats' behavior continues toward the goal even though they give up a particular means of reaching that goal. As I point out above, there is no separate representation of the goal which is driving the coastal snails' behavior; the goal toward which the coastal snails are oriented appears only when the conditionals are considered as a whole. This makes the snails unable to adopt new behavior for the purpose of achieving the goal. The snail is unable to alter its behavior because th, behavior is just the way it behaves; it is not actively linked to any goal. Contrast this with the rat. Instead of a series of rigid conditionals the rat is equipped with a representational system that represents both its goal and the means to the goal (assuming something like the cognitive map story is correct). The rat is able to adapt its behavior to achieve the goal because the goal is what drives the behavior. In short, unlike the snail, the rat has separate representations of the goal and the means to achieve that goal, allowing it to be sensitive to both its successes and failures relative to that goal.
This ability to continue to move toward satisfying a goal despite obstacles which would otherwise prevent satisfaction of that goal, which the rat possesses and the coastal snail lacks, is what separates goal-directed actors from those which are only goal-oriented. The goal-oriented actor is able to achieve its goal only when it finds itself in the right environment; the goal-directed actor, which can adjust its behavior in ways that take the environment into account, is able to achieve its goal in a wider variety of environments.(23)
It is worth taking a moment to take stock here. I've said several different things about the distinction between goal-directed and goal-oriented behavior: that goal-directed behavior involves the flexibility to overcome obstacles when moving toward a goal, while for goal-oriented behavior, an obstacle devastates the organism's ability to reach that goal; that goal-directed behavior requires separate representations of the goal and the means to that goal, while goal-oriented behavior indicates that there is no distinctly represented goal; and that goal-directed behavior involves having the goal explicitly represented, while in goal-oriented behavior, the goal is only implicit in the conditionals which direct the organism's behavior. How are all these distinctions related?
Allow that the central distinction between goal-directed and goal-oriented behavior is a distinction in the ability to overcome obstacles. Then we can ask about the representational structure that allows an organism to reach its goal despite obstacles, and the representational structure that prevents an organism from overcoming an obstacle. In order that an organism overcome an obstacle and remain directed toward its goal, it must have the flexibility to relinquish the behavior that led to that obstacle while retaining the goal as a goal. The representation of the goal, therefore, must be distinct from the representation of the means to that goal; otherwise, the organism could not abandon one without abandoning the other.(24) When an obstacle removes the possibility of reaching a goal, this suggests that, as is the case with the coastal snail, the means to the goal and the goal are not distinctly represented; the organism cannot abandon the means without also abandoning the goal. Finally, when the organism merely has representations of how to behave in particular circumstances, without a separately represented goal, the goal only appears when considering the end result of implementing the means; the goal is not explicitly represented. But if the representations of the means and the goal are distinct, then there is an explicit representation of the goal present.(25)
There are two aspects of this discussion worth highlighting. Remember that folk psychology requires that beliefs and desires be represented discretely so that one can be abandoned while the other is retained. Goal-oriented behavior is behavior which is not caused by discrete representations of goals and the means to achieve those goals. Therefore goal-oriented behavior is not properly explained by reference to folk psychological states.(26) On the other hand, goal-directed behavior does satisfy the demand for separate representational structures, and is thus at least a necessary condition for folk psychological explanation. In order to be appropriate for folk psychological explanation a behavior must be more than goal-oriented; it must be at least goal-directed.(27) Here, then, is an answer to one of the questions raised at the start of this paper: a behavior is properly explained by reference to folk psychology only if it is goal-directed.(28)
Second, because the representation of the goal is separate from the representation of the means to that goal, the actor is not only able to abandon an inefficacious behavior without abandoning the goal toward which it is directed, but is also able to adopt a new behavior directed toward that goal. That is, being goal-directed is intimately associated with being able to learn.
III. LEARNING REVISITED
The clearest counter-examples to the claim that the ability to learn is intimately connected to being goal-directed are of organisms that appear to learn, but not to be goal-directed. Examining these examples will help to refine the characterization of learning offered above as well as to reaffirm the connection between goal-directedness and learning.
A. A Versatile Protozoan
Stentor is a one-celled pond-dwelling protozoan. It consists of a tube attached to some stationary object in the pond, and a disk with cilia at the top of the tube. It feeds by drawing water through the disk with the cilia. When a poison cloud is introduced into its vicinity it responds in a remarkably adaptive manner: it performs a series of behaviors which are increasingly costly. First, for a short time, it merely continues to feed as usual. If the water current fails to carry the poison cloud past it in this time, it then turns away from the current for a brief interval. It will do this a few more times. If this fails to alleviate the problem, it then reverses its ciliary movement--Stentor pushes the water away from it for an instant. If, when the ciliary movement returns to normal, poison is still being carried into its tube, Stentor then reverses its ciliary direction several times in rapid succession. If this still fails to relieve it, it will contract into its tube. The first time it contracts it stays for a mere 30 seconds. However, as with the other poison-avoidance behaviors, it repeats the contraction behavior, each time more drastically; eventually Stentor starts to contract violently while in the tube, breaking free of the object to which it was attached, and leaving its tube as well, to swim away, grow another tube, and feed somewhere else.(29)
What may seem clear about the behavior of Stentor is that it involves associating new means to an old end. Stentor retains the goal of avoiding the poison while abandoning various means which fail to lead to the satisfaction of that goal. However, what remains unclear is whether this is an example of behavior which is truly goal-directed, rather than merely goal-oriented (that is, does it truly learn?, does it truly have separate structures representing its end and the various means to that end?). What suggests that it is goal-directed is that it is able to abandon one sort of behavior while retaining the goal to which that behavior was directed. Unlike the coastal snail, Stentor is apparently not merely following a series of conditionals which, in the right circumstances, lead it to its goal. Stentor is acting in a way which would presumably allow it to reach its goal in any circumstances in which its goal is satisfiable. (If the entire pond in which it lives were filled with some noxious chemical, it would not be able to avoid that cloud, however, no possible movement could carry it out of range of that substance.) Nonetheless, we balk at calling the behavior of a protozoan "goal-directed."
There are two facets of Stentor's behavior that are worth noting in this regard. First, while the behavior appears to be plastic (especially in comparison to that of the coastal snail), it really is not. It may be able to exhibit flexibility in the short run, by abandoning one behavior for another, but this pattern of behavior is rigidly determined; Stentor has a series of responses to a noxious substance in its environment, all of which are triggered invariably.(30) Each episode of exposure to some undesirable material in its environment will trigger an identical series of avoidance patterns. Second, Stentor's later behavior is not affected by its experience with the poison. While Stentor appears to be adopting new means in order to achieve an old end, it is failing to adopt any new behavior for more than a short period, if at all. Learning, I suggest, is more than the mere short-term adoption of a new means; the new behavior must become part of the organism's behavioral repertoire (at least until this behavior is supplanted by another).(31) Nothing of this sort occurs in Stentor's case, this is seen in the absence of effect the earlier incidents with the poison has on later incidents. Careful consideration of Stentor's behavior, that is, shows us that it is as rigid in its response to the poison as the coastal snail is in its pursuit of dark, damp crevices. The obvious conclusion is that Stentor is unable to learn from its previous experience.(32) As with the coastal snail, the process of evolution has oriented Stentor so that the actions it is invariably caused to perform under certain conditions will reliably lead to the goal of being in the right local environment. And, like the coastal snail, Stentor is unable to recruit new means to achieve its old goals.
Learning involves more than mere adaptation to solve a particular immediate problem: future behavior in similar situations must reflect that adaptation. (For example, the learned behavior might surface with greater ease in the future. This is what we would expect if Stentor had indeed learned.) However, it must not become a rigid part of the organism's repertoire. This is seen by looking at the feeding behavior of Blue Jays.
B. Birds and Butterflies
Blue Jays feed on, among other things, butterflies. As a result, in order to protect themselves, Monarch butterflies have evolved to be both poisonous and distinctively-marked. When a Jay eats one of these butterflies it gets very ill, and thereafter does not eat other butterflies with similar markings (including nonpoisonous butterflies of different species which have evolved similar markings for their protection).(33) It certainly looks as if the Jay learns; it represents new conditions which are called into play when it is engaged in eating behavior. The bird's behavior is modified because it has eaten something that made it sick, and it seems as if the correct explanation of this change in behavior is that the bird represents certain things (those that resemble the butterfly) as being bad to eat, and consequently it does not eat them. It may seem perfectly natural to say of the bird that the reason it does not eat butterflies with such patterns, even though it is hungry, and they are present, is that it believes that they are not good to eat (and it has the goal of eating only nourishing things).
Despite this evidence, I do not think that the birds' feeding behavior can be accurately said to be goal-directed. Therefore this behavior is not properly explained by reference to folk psychological states. To illustrate this I want to highlight some facts about this story. What provides evidence for the hypothesis that the birds are learning is that they clearly exhibit some plasticity in their behavior. They get the information that certain butterflies are bad to eat, and they modify their behavior accordingly. However, such modification is not plastic. Once the birds learn that these butterflies are bad to eat "they will never eat the food again."(34) (In fact, they can be taught to never eat something that did not make them sick; apparently there is a preset amount of time that a Jay waits to see if some new food makes it sick, and if the bird does get sick within this time, no matter what the cause, it will avoid that food.)
These attributes of the birds' (apparent) learning ability tells us that the birds are not true learners. Once the bird forms some means-end association (in which the end has to do with feeding) it is unable to modify it; if the behavior is not appropriate (e.g., if it was not the butterfly that and made it sick) the bird will nevertheless continue to act as it has "learned" to act. While the Jays are initially able to modify their behavior, such modification becomes permanent. This inability to continue learning illustrates why the Jays are not goal-directed, but are only goal-oriented. The goal of not eating things that are bad for them, that they do achieve, they achieve because the kinds of actions that their biomechanical programming dictates lead them to that goal when they are in the right environment, much like the coastal snail.(35)
For an organism to be able to learn, it must be able to do more than exhibit plasticity of behavior in response to some stimulus; it must be able to continue to do so. In order to be a learner (in the sense in which I am using the term) an organism must be able to associate new behaviors with an old goal when such an association is warranted. The problem with the Jays is that they are able to do this only once. The requirement that an organism be able to continue to learn relative to some goal rules out behavior that is explicable without reference to a proximal goal, that is, without reference to a goal acting directly on the individual organism's actions. This requirement ensures that an organism which receives the information that its actions are not sufficient to achieve its goal will always be able to abandon those actions.
In sum, to learn relative to some goal an organism must satisfy these criteria:
(1) The organism is able to form new associations of that goal and actions (as a result of a new stimulus presented to the organism between the occasion of two similar situations),
(2) This new association of the goal and an action must be able to affect future actions of the organism,
(3) The organism must be able to continue to form such associations while retaining the same goal, i.e., it must not lose its ability to satisfy condition (1) relative to a goal.(36,37)
I have been arguing that the satisfaction of these conditions is necessary for an organism to be goal-directed, that an organism which fails to satisfy these conditions relative to a particular goal is not goal-directed relative to that goal. What remains is to argue that satisfaction of these conditions relative to a goal is sufficient for goal-directedness relative to that goal.
Suppose satisfaction of these conditions is not sufficient for a particular behavior's being goal-directed, then it is possible for some organism to satisfy these criteria relative to some goal, and be merely oriented toward that goal. But if this organism is able to exhibit the kind of plasticity that (3) requires, it must be able to jettison any behavior that does not lead to the goal and yet remain oriented toward that goal. If it is the case that the behavior that this organism exhibits is not truly goal-directed, then it must be the case that there is only one representational level that is governing its actions; there is not a representation of both a means conditional and a goal state, but only of the one conditional, or set of conditionals. (It is likely that this latter possibility describes the protozoan Stentor.) It must be the case, for this organism to exhibit the kind of plasticity that is imagined, that the set of conditionals that governs its behavior is large enough to account for the many possible obstacles over which the organism may need to exhibit plasticity. However, that set of obstacles is so large and varied that preparing for all of them in advance is not possible. The kind of plasticity that (3) requires involves an ability to adapt in order to continually overcome obstacles. But this would require true adaptability; it would not be possible given a preset series of different representations of what to do in each specific circumstance (even were there many, many different representations). I conclude that satisfaction of conditions (1), (2), and (3) relative to some goal is sufficient as well as necessary for being directed toward that goal.
I argue here that folk psychological explanation, when it is the proper sort of explanation to give of some behavior, implies that there are two distinct structures operating in the brain of the agent behaving. One is a belief, or set of beliefs, about the state of the world and how to reach the goal state. The other is a representation of the goal state. Furthermore, only when these two structures are actually discrete(38) is goal-directedness accurately predicated of the agent. If they are not discrete, then the actions which look to be goal-directed, in virtue of which we claim that the agent is doing something because of her beliefs and desires, are not really goal-directed, and thus are not truly open to folk psychological explanation. They are merely oriented towards the goal, and in less than optimal conditions they will not result in the satisfaction of that goal. The distinction between goal-oriented and goal-directed behavior can be seen in the plasticity that an organism exhibits with regard to its goal and the actions undertaken to achieve that goal. There are two basic levels of this plasticity: the organism may not be able to abandon a set of behaviors without abandoning the goal toward which they lead, this clearly indicates that the organism is only goal-oriented; or the organism may appear to abandon one set of behaviors while adopting another which better lead it to the goal, in this case we may be tempted to claim that the organism is exhibiting goal-directedness. However, in order to be truly goal-directed an organism must be able to learn robustly: it must be able to abandon previously adopted behaviors that do not lead to its goal, and to do so whenever the behavior is not leading it toward its goal. That is, it must be able to modify its actions when they are inefficacious in achieving its goals. This ability is evidence that the organism is able to do what I call "learning," and it is this ability that distinguishes goal-directed from non-goal-directed actors.
The ability to learn relative to a goal is the ability to satisfy (1), (2), and (3) relative to that goal. Possession of this ability is both necessary and sufficient for being goal-directed. Being goal-directed is itself necessary for agency; therefore, the ability to satisfy (1), (2), and (3) is also necessary for agency. Folk psychological explanation of a behavior is thus appropriate only when the agent is able to learn relative to the goal to which that behavior is directed; if the agent is unable to learn relative to a particular goal, then her behavior is not caused by beliefs and desires about that goal.
To return to Angelina: if she is a kleptomaniac, and that means that she is unable to learn not to steal the umbrella when stealing the umbrella is a poor way of satisfying her desire to stay dry, then a folk psychological explanation of her umbrella-stealing behavior which appeals to her desire to stay dry is not the proper explanation of her behavior.(39)
(1.) This question has received significant attention lately from both philosophers and scientists. See Bekoff and Jamieson (1996) and Allen and Bekoff (1997). For a more general overview of the issues involved in attributing minds to nonhuman animals, written by a scientist for an audience of nonscientists, see Dawkins (1993).
(2.) For other, recent, discussions of folk psychology that center on learning, see Dretske (1988), and McLaughlin (ed.) (1991), (see especially Dretske's reply on learning); for a conceptual analysis of folk psychology that also discusses learning ("educability" in his terms), see Bennett (ms); for a discussion of folk psychological explanation that also focusses on the plasticity necessary for the applicability of such an explanation, see Enc (ms); and for discussions of the criteria for applying folk psychological explanations to the behavior of nonhuman animals, see Allen and Hauser (1991), Bennett (1991), and Davidson (1984, 1985).
(3.) As will become obvious in the body of the paper, the learning under discussion here does not subsume all varieties of learning. In particular, most psychological studies of learning focus on some sort of associationist paradigm (be it classical conditioning, operant conditioning, or some other sort of learning that involves the association of proximally presented stimuli); see, for example, the table of contents in Staddon (1983) for a nice overview of the varieties of learning studied most frequently by psychologists. Only lately have psychologists turned away from a strict associationism (see Bolles , and Gallistel [19901). However, my concerns here are isomorphic to these issues; I am interested in noting the connection between learning in a particular manner, which I define below, and folk psychological explanation.
(4.) One variety of knowledge, "knowledge how," may be possible without belief, but the knowledge implicit in this definition is "knowledge that," which is not possible without belief. The locus classicus for the distinction between knowing how and knowing that is Ryle (1949).
(5.) Clearly this is very rough. I have no precise criteria to indicate what would make two situations relevantly similar. However, such precision is not necessary for this essay; the idea is simply that whatever difference there might be in environmental stimuli is not able to wholly account for the change in behavior. This characterization should suffice to direct our intuitions here.
(6.) Note the modal nature of this formulation. The claim is not that there must be an immediate change in behavior, or even that the organism must have a chance to fulfill its changed disposition. All that is necessary is that the organism would be able to incorporate the new information into its behavioral repertoire if it had the chance. How it might do so is also not important. So, if I learn that you do not like being woken before seven, I may modify my behavior by ceasing to wake you before seven (given the opportunity-I can learn this, and modify my disposition to behave, even if I get hit by a bus this afternoon and never have the chance to not wake you before seven again), or I may modify my behavior by giving different explanations of what I'm doing as I continue to wake you before seven. The difference between me and a rooster, for example, is that the rooster will never learn to modify its behavior and not wake you before seven (although it might learn not to crow before seven).
Note, also, that my interest here is not in when learning takes place, just in whether it is possible for learning to take place. Given this, I need not worry about when the modification in disposition to behave actually takes hold, just whether it is possible for the organism to modify its dispositions to behave.
(7.) What about, say, a songbird learning its song; isn't this learning? Certainly this is a change in disposition to behave, and just as certainly, many scientists who study learning study just this sort of change in disposition to behave. Still, this sort of learning, brought about as it is by maturation and not information, is not the sort of learning I am interested in here; it will not help to determine when folk psychological explanation is appropriate and when it is not.
(8.) Exactly what counts as information is not important here. The rough idea is that the only way to explain the change in behavior is by reference to some fact that the organism has had the opportunity to learn.
(9.) The information need only be new to the organism behaving. And, of course, new information includes information available to the organism previously, but which had not yet been absorbed by the organism.
(10.) It may be the case that the organism is not learning, that more detailed examination of any given situation would show that the modification in the organism's behavior is not due to the information presented to it, but to some other factor. However, such counter-examples do not argue against this characterization of learning, merely against relying on ordinary judgments.
(11.) I do not mean to suggest that the pigeon has an active mental life which includes (mental) sentences such as, "If I want food, I should peck here." I do mean to suggest that there must be some sort of link between its motivational state (hunger) and its behavior; what goes on when it learns is that a different behavior gets associated with the motivational state. One natural way to think of this is as a conditional, "if S is the situation, do A," where S may describe the motivational state, a crude representation of the environment, or a more fully mental representation of the environment. Learning would then be either the recruitment of a new A for the old S, or a new S for the old A. My discussion below is framed in terms of this conditional, although nothing rests on this aspect of the discussion.
(12.) This rough characterization of learning will be refined below (in section 3).
(13.) This appears to be incorrect; one could form a new association of a means and an end as a result, for example, of recognizing that the new means satisfies the end at a smaller cost. However, this is merely a less dramatic example of a behavior that no longer completely satisfies the goal.
(14.) The points I make here are old hat. I make them only because they are important for the argument that follows.
(15.) I do not wish to commit myself here to any particular position on exactly what desires are. What I have said so far, and what I say below, is consistent with desires being merely activations of the nervous system, as Dretske (1988) suggests; quasi-perceptual states that represent something as being good, as Stampe (1986, 1987) suggests; or ordinary representations in a desire-box, as standard functionalist doctrine suggests.
(16.) One may read this as a rejection of Davidsonian holism, however it is not meant as such. Davidson's holism is, as I understand it, a position with regard to the possibility of attributing just one belief to an organism, and with regard to finding the neural structures that correspond to particular beliefs and desires (see his ). Given Davidson's adherence to the supervenience of the mental on the physical, however, he must believe that folk psychological states are realized in physical structures, and given the grounds for his holism--the multifarious events in which any mental state can be causally involved (see, again, his , especially pp. 254-255)--he must hold that beliefs and desires are distinctly realized, in the sense intended here (see note 17). Despite this agreement about the distinct structures which realize mental states, the conclusions I reach with regard to folk psychological explanations of behavior do seem to be inconsistent with those that Davidson draws from his holism (see his [1984, 1985])-that mental states are appropriately predicated only of language users and interpreters. Rather than dispute Davidson's claims directly, for it may be the case that our accounts converge (it may be that the sort of learning ability that I argue an organism must have in order that its behavior be appropriately explained by reference to folk psychological states is only possible if the organism is a language user and interpreter), I concentrate on reasons to believe that the account offered here is correct. If that means that Davidson's criteria are overly strong, so be it.
(For direct arguments against Davidson's account, see Jeffrey  and Dardis [ms].)
(17.) Some read this commitment very strongly; for example, Ramsey, Stich, and Garon (1991) argue that the causal modality that beliefs enjoy implies a representational modality. That is, different beliefs must be encoded in physically different structures. See Forster and Saidel (1994) for an argument against this claim. When I say that beliefs and desires must be separate structures, I mean merely that they must be able to enter into different causal interactions. How this may be realized neurally is unimportant for the purposes of this paper. (It may help one to understand this claim to think of a green cube. The object is able to enter certain causal interactions on account of being green and others on account of being a cube. Furthermore, it could lose one property while still retaining the other. It is to this extent that beliefs and desires must be separate structures.)
(18.) See Gallistel (1980), pp. 143-165.
(19.) Daniel Dennett (1978a, pp. 60-65; 1978b, pp. 244-5) draws the same conclusion from the behavior of the wasp Sphex.
20. One way to think about this is that the snail's actions are pre-wired in such a way that it will "recognize" when it has reached the goal (the antecedents of the conditionals will fail to be satisfied, and it will cease movement), but that at no time will its actions be directed in any way by a comparison of its current state with its goal state. See McFarland and Bosser (1993) for a discussion of goal-directed behavior that's presented in these terms.
(21.) But what of the situation in which the organism has two distinct representations of means-end pairings which each involve the same end? Surely, that organism is able to abandon one of those representations while retaining the goal. The protozoan Stentor is a good example of such an organism. See section 3.A for a discussion of its behavior, and an argument that it is not goal-directed.
(22.) This suggestion is due to J. A. Deutsch. See Gallistel (1980, pp. 334-359) for a more comprehensive discussion of the cognitive map and Deutsch's model of the role representations play in animal behavior.
(23.) An illustrative way of thinking about this distinction is that goal-oriented actors are given a push (by evolution) in the direction of their goal. If the push proves to be inadequate, they cannot repair their position. Goal-directed actors, on the other hand, have the ability to redirect themselves.
24. Again, I do not mean to suggest that one must have semantic representations of the goal and the means to the goal; merely that goal-directed behavior is only possible when several different behaviors can be linked to one motivational state. See note 11 above. Nor do I mean to suggest that the representations must be realized in physically distinct structures; see note 17.
(25.) McFarland and Bosser (1993) draw a distinction between goal-directed and goal-seeking behavior similar to the one I draw between goal-directed and goal-oriented behavior. On their account, goal-directed behavior is behavior that is controlled by a representation of the goal, while goal-seeking behavior is behavior in which no goal representation is operative, but which will reach a goal if performed in the right circumstances. Our aims are different, however. They are interested in the implications for behavior that follow from the assumption that for some organisms the goal is represented explicitly, while I start by looking at the plasticity of the behavior, and make inferences about the representational structure underlying that behavior.
(26.) It may look as if we can tell a story about goal-oriented behavior in which the behaver does have both means and ends representations, and they are discrete. All we need do is imagine that the behaver is unable to acquire new means representations. Then the behavior performed will retain the stagnation characteristic of goal-oriented behavior, even though the behaver does separately represent its ends and the means to achieve those ends. What this characterization of a goal-oriented behaver misses is that means representations are supposed to be representations of the world; if they are unable to change, if the behaver is unable to acquire new means representations, then it does not truly have representations of a means to achieve a goal. Instead, its rigidity is so deep that all it is able to represent (much less act on) is a world that looks a particular way.
(27.) It may seem that my discussion of goal-directed behavior assumes that goal-directed behavers have beliefs and desires. The goals toward which the actor is directed are its desires; it is only these that will cause it to change its behavioral course. However, all I have argued for is that goal-directed behavior essentially involves a separate representation of the goal. This could come in the form of a desire, as it does with human beings (and then, if we are well- functioning, our desires coincide with our biological goals); or it could come in the form of some sort of need state which is not fully a desire, say for food (simpliciter), or sex (simpliciter), as it presumably would in the case of animals which are goal-directed but not fully intentional.
(28.) But suppose an actor represents a goal in such detail that the goal includes the means to achieve it. Not only does Ramona want a glass of Laphroaig, but she wants it to come from this bottle, and she wants you to pay for it, etc. Ramona cannot abandon the means to the goal without abandoning the goal, for the goal includes the means. According to what I say here, folk psychological explanation is not the right sort of explanation to give for Ramona's Laphroaig-pursuing behavior. This should seem intuitively correct: if we cannot separate Ramona's desire for Laphroaig from her means of achieving that desire, then the real desire is for Laphroaig-by-that-means, not for Laphroaig simpliciter. Any relevant beliefs she has will be about Laphroaig-by-that-means, not about Laphroaig simpliciter. We make this distinction regularly in descriptions of much more mundane behavior: we talk about the desire for ice cream, rather than the desire for something to eat; or the desire for scotch, rather than a drink. The desire for scotch cannot be satisfied without also getting something to drink; however, the goal toward which the relevant beliefs are directed is the goal of acquiring scotch. If we try to redescribe the behavior as drink-directed, then it may look as if the behavior is not goal-directed. "Ramona has a glass of water in front of her, why does she continue to ask the bartender to bring her a Laphroaig?" Her behavior may not be directed toward the goal of acquiring a drink, only oriented toward that goal, while it is directed toward the goal of acquiring a scotch (when informed that there is no Laphroaig to be had, she asks for Lagaluvin).
(29.) Stentor's avoidance behavior is discussed in depth in Staddon (1983), pp. 92-7, and 114-115.
(30.) Staddon discusses two possible explanations of Stentor's behavior. He suggests that either there is a serial set of actions organized so that the failure of one avoidance method triggers the next method, each method being more drastic (costly for the protozoan) until Stentor detaches itself from its foothold. Or the presence of the noxious cloud triggers several behaviors in parallel; in this case the time of exposure is the key variable, the longer the exposure to the cloud the more drastic the avoidance method that is caused by crossing some temporal threshold.
(31.) McFarland and Bosser (1993, p. 257) make a similar claim: "When an animal learns ... its behavioral repertoire undergoes permanent alteration."
(32.) Stentor is in a position such that it need not learn from its experiences. It has already evolved a complex series of behaviors which allow it to avoid harmful substances in its feeding area at a minimal cost; what more could it want? However, this is beside the point. I am merely claiming that it cannot learn from its experiences, and that this fact indicates that it is not goal-directed.
(33.) See Gould (1982), pp. 263-264.
(34.) Gould, p. 264. Despite what Gould claims, it may seem unlikely that the food aversion Jays acquire as a result of eating the Monarch is actually permanent. However, my point here is merely about what conclusions we could draw were the "learned" behavior permanent.
(35.) If the birds were in an environment in which there was also an insect which would cause them to be ill several hours after eating it (i.e., after the waiting period has elapsed), the birds would stop eating all those things which, in reality, did not make them sick. The result could easily be that they eat only the insect that makes them sick (or perhaps even that they starve). Of course, it is very improbable that such a situation would evolve. (If we were in a similar situation, we would eventually discover what it was that made us sick, and relearn to eat those things that were good for us. Perhaps the same would be true of the Jays; my point is merely that if the Jays are unable to relearn, then they are merely goal-oriented.)
(36.) Of course, it may eventually give up; however, my claim is that it must continue to have the ability to satisfy condition (1). If it loses that ability with regard to some goal, it will no longer be appropriate to explain its acts, with respect to that goal, by reference to folk psychological states. If I continually boil eggs in order to make the phone ring and am unable to abandon that behavior despite being convinced that such behavior cannot lead to the satisfaction of my goal, we would rightly hesitate to refer to folk psychological states in explaining my behavior.
(37.) What if the organism is able to satisfy (1) and (2) relative to some goal, but not (3)? It seems that some learning is still going on. Perhaps so, yet that organism is not a robust enough learner with respect to that goal that a folk psychological explanation of its behavior is warranted. For example, Gould and Gould (1994) discuss the manner in which young finches learn about predators: they hear older birds' mobbing call (finches respond to certain predators by mobbing them) and "learn" that the object they are currently viewing is a predator. Experimental manipulations resulted in young finches "learning" to mob things that were not predators, including a bottle of laundry detergent. This sort of learning, despite satisfying (1) and (2), seems best explained by classical conditioning; it is the sort of learning that might support a goal-oriented behavior, but not a goal-directed behavior.
(38.) In the minimal sense discussed above; see note 17.
(39.) Earlier versions of this paper were presented to the cognitive science group at the University of Southwest Louisiana, and to the SESAME/EMST colloquium series at the University of California at Berkeley. I am indebted to those audiences for their comments. I am also indebted to Colin Allen, Fred Dretske, Berent Enc, Malcolm Forster, Heather Gert, LaVerne Shelton, Leora Weitzman, and two anonymous referees for this journal for their comments on earlier drafts (in some cases distant ancestors) of this paper.
Allen, Colin, and Bekoff, Marc (1997) Species of Mind: The Philosophy and Biology of Cognitive Ethology (Cambridge: MIT Press).
Allen, Colin, and Hauser, Marc (1991) "Concept Attribution in Nonhuman Animals: Theoretical and Methodological Problems in Ascribing Complex Mental Processes," Philosophy of Science 58, 221-240.
Bekoff, Marc, and Jamieson, Dale (1996) Readings in Animal Cognition (Cambridge: MIT Press).
Bennett, Jonathan (ms) "Analysis without Noise."
--.(1991) "How is Cognitive Ethology Possible?" in Ristau, Carolyn A., ed., Cognitive Ethology: The Minds of Other Animals (Hillsdale, NJ: LEA), 35-49.
Bolles, Robert C. (1988) "Nativism, Naturalism, and Niches," in Bolles, Robert C. and Beecher, Michael D., eds., Evolution and Learning (Hillsdale, NJ: LEA), 1-16.
Dardis, Anthony (ms) "Kitty Thoughts."
Davidson, Donald (1973) "The Material Mind" reprinted in Essays on Actions and Events (Oxford: Oxford University Press, 1980), 245-259.
--.(1984) "Thought and Talk" in Inquiries into Truth and Interpretation (Oxford: Oxford University Press), 155-170.
--.(1985) "Rational Animals" in LePore, Ernest and McLaughlin, Brian, eds., Actions and Events: Perspectives on the Philosophy of Donald Davidson (Oxford: Basil Blackwell), 473-480.
Dawkins, Marian Stamp (1993) Through Our Eyes Only? (Oxford: W. H. Freeman).
Dennett, Daniel (1978a) "Skinner Skinned," in Brainstorms (Cambridge: MIT Press, 1978), 53-70.
--.(1978b) "Mechanism and Responsibility," in Brainstorms (Cambridge: MIT Press, 1978), 233-255.
Dretske, Fred (1988) Explaining Behavior (Cambridge: MIT Press).
Enc, Berent (ms) "Rules and Reasons."
Forster, Malcolm, and Saidel, Eric (1994) "Connectionism and the Fate of Folk Psychology: A Reply to Ramsey, Stich, and Garon," Philosophical Psychology 7, 437-452.
Gallistel, C. R. (1980) Organization of Action: A New Synthesis (Hillsdale, NJ: LEA).
--.(1990) The Organization of Learning (Cambridge: MIT Press).
Gould, James L. (1982) Ethology (New York: Norton).
Gould, James L., and Gould, Carol Grant (1994) The Animal Mind (New York: Scientific American Library).
Jeffrey, Richard (1985) "Animal Interpretation" in LePore, Ernest and McLaughlin, Brian, eds., Actions and Events: Perspectives on the Philosophy of Donald Davidson (Oxford: Basil Blackwell), 481-487.
McFarland, David, and Bosser, Thomas (1993) Intelligent Behavior in Animals and Robots (Cambridge: MIT Press).
McLaughlin, Brian (1991) Dretske and His Critics (Cambridge: Basil Blackwell).
Ramsey, William, Stich, Stephen, and Garon, Joseph (1991) "Connectionism, Eliminativism, and the Future of Folk Psychology," in W. Ramsey, S. Stich, and D. Rumelhart, eds., Philosophy and Connectionist Theory (Hillsdale, NJ: LEA, 1991), 199-228.
Ryle, Gilbert (1949) The Concept of Mind (Chicago: The University of Chicago Press).
Staddon, J. E. R. (1983) Adaptive Behavior and Learning (Cambridge: Cambridge University Press).
Stampe, Dennis (1986) "Defining Desire," in J. Marks, ed., The Ways of Desire, (Chicago: Precedent Pub.).
--.(1987) "The Authority of Desire," The Philosophical Review, 96, 335-381.
|Printer friendly Cite/link Email Feedback|
|Publication:||American Philosophical Quarterly|
|Date:||Jan 1, 1998|
|Previous Article:||Prescriptions are assertions: an essay on moral syntax.|