Everyone talks about stimulus-response psychology, but no one seems to do anything about it. This chapter constitutes a proactive effort to build a foundation for animal learning theories based upon an alternative to stimulus-response psychology, the psychology of action.
Disenchantment with traditional stimulus-response formulations in psychology has recently led a number of researchers (e.g., Harré, 1982; von Cranach, 1982; Lee, 1988; Oppenheimer, 1991) to advocate a view of psychology based on the notion of action. Among these, Lee's (1988) approach is distinct in that she begins from the radical behaviorist perspective of B. F. Skinner (1974).
Mandler & Kessen (1959) point out that various metatheoretical perspectives can be thought of as various languages used by psychologists in describing and explaining psychological phenomena. Thus, stimulus-response psychology can be understood as the linguistic practices of psychologists in describing and explaining psychological phenomena in terms of functional input-output relations. Alternatively, action psychologies involve the description and explanation of psychological phenomena in terms of things done by the organism. In this chapter, I will expand upon Lee's (1988) framework in order to see what using this new language to reexamine the phenomena of animal learning can tell us about the structure of learning.
This chapter will consist of three sections: First, I will show what sorts of structure are made evident by a radical behaviorist view of action. Second, I will argue that the action theory perspective brings into question the usual choice of the basic unit of analysis in animal learning theories and suggests an alternative. Finally, I will describe a typical example of operant learning phenomena, operant conditioning of the pigeon, in the action-theoretic language.
Stimulus-response psychologies, including methodological behaviorism and cognitive psychology, are notorious for failing to define their terms, including such basic terms as stimulus and response ( Koch, 1954; Mandler & Kessen, 1959; Gibson, 1960; Turner, 1967); Meehl, 1978). In attempting to define the basic terms of action theory, a number of authors ( Austin, 1962; Goldman, 1970; Davis, 1979; Brand, 1984; Hamblin, 1987; Lee, 1988) have chosen to examine, to some degree or another, the structure of action. I will explore that structure in detail in this section.
Lee's Metatheoretical Approach
Given the wide variety of action theories cited above, it is important to summarize the particular view followed here, that of Vicki L. Lee (1988). Lee comes from the perspective of radical behaviorism offering a language for psychology based on the notion of action. She contrasts her view with the traditional stimulus-response terminology of contemporary psychology as well as with other views of action. She suggests that the best language for treating behaviors as actions is a language based on a basic vocabulary of action verbs.
Basic terms. Lee avoids the word behavior as it "is ambiguous between activity, moment, and action" (p. 42). For instance, Ziff (1958) argues that the behaviorist inevitably means action when s/he uses the term behavior. However, Harré (1982) not a behaviorist, uses the term behaviour to mean mere bodily moment.
Lee prefers the term conduct to describe the general subject matter of psychology. She begins her inquiry by pointing to the sorts of events described in common English using action verbs, for example: "running, reading, writing, touching, holding, sitting, and driving" (p. 38). Unique, individual, particular event-tokens of this sort, Lee calls acts. General, broad classes of acts, Lee calls actions.
It is important at this point to note that Lee's terminology is quite distinct from the terminologies adopted in other action psychologies. Harré (1982) for instance, has a tripartite taxonomy of behaviour/action/act. As noted above, Harré uses the term behaviour to indicate mere bodily movement. He uses the term action to indicate action in the context of the person acting along with a supposed intention, possessed by that person, that is the proximate antecedent efficient cause of the action. He uses the term act to indicate action in the context of its broader social and cultural setting.
The problem of classification. Lee's argument also stands out from other psychologies of action due to her focus on the problem of classification. Lee (1988, chap. 4) details the type-token distinction, also known as the species-individual distinction, or the class-particular distinction. Lee prefers the last terminology. She defines a concrete particular as "anything we can regard as a single entity." By way of example, Lee specifies two sorts of concrete particulars: objects and events. She points out that concrete particulars are unique.
The other side of the distinction is exemplified by collections of particulars. Lee refers to these collections as classes. An older term for particular is token; for class is type. I will be using this older terminology in this chapter. The notion of classes allows the introduction of abstract entities into Lee's explanatory scheme.
Lee follows Skinner (1969, p. 131) in using the notion of a class as the basis for theoretical explanation in psychology. This strategy points up the motivation for Lee's focus on the problem of classification. At the same time, it points up an important distinction between Lee's view of action and other related views.
The differences in focus between Lee (1988) and Harré (1982) reflect -- to some degree -- the theoretical prejudices of the two authors. Harré focuses on the breadth of the context of interpretation as the primary distinction between levels of explanation in part because he believes that antecedents (i.e., intentions), as well as consequents, play a part in the explanation of conduct. Lee focuses on the token-type distinction in part because she does not believe it essential to assume the existence of intentions, mental states, or mental events in order to offer explanation of conduct.
If individual act-tokens are caused by unique, particular mental events or mental states (for example, intention-tokens), then types-of-action (action-classes) can be explained in terms of types-of-intentions in straightforward fashion. Thus, a detailed examination of the problem of classification is not necessary for the mentalist. As a radical behaviorist, Lee will not assume the existence of unseen mental entities such as intentions as part of an a priori scheme for explanation. As we will see later, in order to construct an explanatory strategy without postulating mental entities, the radical behaviorist uses a more intricate model of psychological explanation, relying centrally on the token-type distinction.
Awareness and action. Another difference is that Lee takes action as having very broad scope. She contrasts her own view with von Cranach's (1982) narrower view of action as "goal-directed, planned, intended, and conscious behavior" (p. 36). For Lee, beginning as she does with the language of action verbs, any sort of conduct describable using an action verb is to be considered an action until shown otherwise. In order to claim a measure of generality for his definition of action, von Cranach goes further, claiming that this specific, narrowly defined type of conduct is "the more common form," with other forms "found only in rare cases." I will have more to say on the rarity of non-conscious conduct later on.
Despite the fact that Lee (1988) wishes to deemphasize the distinction between conscious and non-conscious action, she agrees with Harré (1982) that the a priori elimination of verbal self-report from psychology on methodological grounds is an arbitrary practice and harmful to psychological inquiry. On the views of all these authors, self-awareness is a legitimate part of conduct.
Similarities to other action psychologies. There are other important similarities between Lee and other psychologists of action. Lee (1988), Harré (1982), and von Cranach (1982) all take it as essential that actions be construed as means-ends pairs. Harré (1982, p.10) expresses this sentiment quite eloquently: Action psychologies "depend upon the use of a single, overriding principle, namely that action should be seen in terms of a basic means-ends format. Every action points beyond itself, and, it seems, must be regarded for scientific purposes as a means towards some end."
Both Lee and Harré emphasize the importance of the distinction between bodily movement (by whatever name) and action. Both insist that action consists in doings and not in happenings. As Harré puts it (1982, p.12): "Action is what people do as opposed to what happens to them." (See also Hamblin, 1987, p.140; Davis, 1979, pp.4-5; Goldman, 1970, pp.16-17; Brand, 1984, pp.3-6, for various views on the doing-happening distinction.)
Thus, while explanatory strategies differ, a focus on the essential differences between actions and events, along with a commitment to a holistic specification of action, make these psychologies of action quite distinct from more traditional stimulus-response psychologies. Indeed, even Lee's neglect of Harré's distinction between personal action and action-in-social-context is due to her commitment to the notion that all action must be individuated in terms of its environmental consequences. Lee's position is, in this small regard, even more holistic than Harré's, as she groups together all the environmental consequences, whether personal or social.
The language of psychology. One last, unique aspect of Lee's view is her concern with the language of psychology. As noted above, by beginning her inquiry with ordinary action verbs, she avoids a reductionistic, essentialist definition of action in terms of other, putatively more basic, notions. By focusing on verbs such as remembering, rather the nouns, such as memory, she avoids too-early imputation of unobservable internal mental events or states. This keeps her taxonomy parsimonious. For instance, actions are defined without introducing the concept of intentions. (Note that a sentence like: Sally intends to hit Joe. suggests that intending, itself, on Lee's view, very well might be an action.)
Stimulus-response psychologies, including methodological behaviorism and cognitive psychology, focus on nouns. In the study of learning, lights and bells lead to lever-press and key-peck, rewarded by food and water. Lee feels that this sort of language leads away from the true causes of conduct.
An Action Terminology
In order to get at structure by examining action, I propose a specific terminology for describing action, based on Lee's (1988) metatheoretical framework. Wittgenstein (1953) raised the question of the difference between an event, such as the raising of an arm, and an action, such as someone raising her or his arm. I begin with this question, hoping to specify a sequence of terms, beginning with event and concluding with action, that will clarify the nature of the difference between event and action.
I follow Lee (1988) in understanding events as entities extended in time: events begin and end. As Lee (1988, p.29) puts it: "Events are concrete particulars that occupy time, and objects are concrete particulars that occupy time and space."
Movements are a specific sort of event that involve objects. An event specified in terms of the motion of some object or objects through space for some fintie period of time is a movement. When at least one of the objects involved is the body of some animal or a part of that body, we call the movement a bodily movement.
Now, bodily movements are not the whole of the story. If my arm is at rest and is struck by some falling object, then my arm moves, but I did not move it. If I wire up the leg of a frog and send an electric current into the nerves, causing the frog's leg to jerk, then the leg moved, but the frog did not move it.
In order to deal with this further distinction, we need to add a new concept, that of person. Consider the behaving organism, not as an input-output device, not as an S-R automaton, but as a creative, active individual constantly acting upon its environment and being acted upon by it; a locus of behavior, its dispositions to act shifting constantly, changing, growing, and altering to cope with the changing environment around it. Not an organism at odds with its environment, but a person integrally a part of an ecosystem. The logical structure of this collection of dispositions is what Lee calls a person.
A deed, or doing, is a specific sort of bodily movement done by some person. It might be clearer to say that a deed is an ordered pair, consisting of a bodily movement and a person, where the person is the do-er or actor, who moves her or his own body. In either case, a deed or doing is that which a person does, in contrast to a happening, which is that which happens to a person. Lee (1988) uses the term activity (p.48) to refer to deeds.
Again, this is not the whole of the story, for actions, according to Lee, are more than mere doings. An action consists of a means and an end. A deed or activity is just the means. As Lee (1988, p.49) puts it: "Postural activities and movements provide the means by which we act upon the world and produce end results." 
If deeds are the means, what are the ends? Lee (1988) is not too clear, but seems to suggest that ends are consequent states of the environment. Harré (1982) claims that ends must be understood in terms of intentions possessed by the person. My hypothesis is that ends are further actions. We perform deeds in order to gain the opportunity for further action. The preceding sentence expresses my central claim as to the basic nature of the means-ends structure of action.
I do something (action1) in order that I later may do something else (goal=action2). I go to dinner (action) in order that I may eat (goal). I turn on a light in order that I might read a book.
Thus, an action is an ordered pair consisting of <deed,action> where the deed is the means and the action is the end. In computer science and linguistics, this sort of definition, where the term defined (action) also appears as a part of the definition, is called a recursive definition. Recursive definitions allow for an infinitely expansible structure. A disadvantage is that infinite regress is possible and should be avoided.
As I mentioned earlier, the token-type distinction plays a crucial role in a radical behaviorist view of psychological explanation. The proposed recursive definition of action provides a starting point for this explanatory strategy.
Two problems. Consider first of all that the definition is given in terms of types, not tokens. Action is defined, but not act. The specification of the type to which a particular token should be assigned, is called individuation. The individuation of acts is a difficult problem. On my view, an individual act-token is explained only in terms of the action-type(s) of which it is an instance. Thus the proper individuation of act-tokens is prerequisite to the explanation of individual acts.
Second, there is a venerable criticism of the individuation of acts in terms of their consequences. What of acts that do not result in their intended consequences? If I go to a restaurant in order to eat and the restaurant is closed, where is the act-of-eating that is the essential end of the means-end pair? The answer to this criticism is that the act-of-going-to-the-restaurant can only be understood as an act of going-to-the-restaurant by means of specifying that act-token as an instance of the action-type, goings-to-restaurants. The individual act-of-going did not result in an individual act-of-eating, but the individual act-of-going did belong to the general class of actions-of-going and the general class of actions-of-going has, as a general goal, the class of actions-of-eating.
Explanation in action terms. An act is an individual, concrete, particular event. Hence its consequences are also individual events. However, goals are generals. When an individual consequence is an instance of some goal, we call that consequence the intended result of that act. When a consequence is not an instance of any goal, then that consequence is an unintended side-effect of that act. (Recall that goals are further actions. This suggests that consequences are acts.)
I believe that it has been a fundamental mistake in psychology and philosophy to claim that actions can be individuated by identifying all the individual act-tokens that comprise them. If acts are collected according to the kinematic similarity of the physical bodily movement, the act-class or act-type is called a deed. Actions, like deeds, are type-level phenomena; but this does not mean that actions, like deeds, are mere collections of individual acts.
I wish to build a formal barrier to that reductive mistake into the action language itself. Let us say that individual acts (act-tokens) are members of deeds (act-classes) in the usual set-theoretic sense, but that acts are instances of actions, in the following sense:
An individual act is an instance of some action if (a) the act is an element of the deed-class (type of bodily movement) that is the means of that action and (b) the goal of that act is the end of that action. How a goal (a general) may be a goal of some particular act is a matter for the particular psychological theory. In cognitivist terms, the goal might be a part of the plan in the mind of the actor. In Skinnerian terms, the goal is the action-class of which previous consequences have been instances. In either case, the structure of the relation is the same. The instance-relation between act and action is dependent upon other instance-relations between consequences and goals. Hence, this definition is also a recursive one.
Individual acts are mere doings. Action, properly understood, always involves some notion of purpose. Purpose can be modeled only with respect to action-classes, not individual acts. This is the technique by which the radical behaviorist explains purpose.
The Macrostructure of Conduct
The recursive definition of action allows for the individuation of actions, though not acts. This individuation is determined with respect to what I call the macrostructure of conduct. Lee (1988, p.57) says that any action "is embedded in a network of overlapping and interlocking actions." This suggests that the structure of conduct can be explicated using a language based on action verbs. Various relations between persons (actors), actions, and bodily movements (deeds) can be used as models of various ways that people behave.
The organization of conduct. A single act (e.g., Ned crooks his finger) may have many goals (having the finger bend, pulling the trigger, firing the gun, shooting the King, killing the King, starting the revolution, etc.) and thus be part of many simultaneous actions. Thus, many actions may be nested together by virtue of including the same deed as the means to each, but with each action having a different goal (end). These goals may be partially ordered as to being more or less proximal or distal to the deed in question. In the philosophy of action, this ordering is considered to be central to the problem of individuating action ( Austin, 1962; Anscombe, 1963; Davidson, 1963; Goldman, 1970); Brand, 1984). W. G. Lycan (personal communication, November, 1990) refers to this ordering as the rich-thin continuum.
Another way actions may be ordered is as proper subparts or components of one another. I walk to school (action1 = <walking, arrive at school>). In order to achieve this, I step forward (action2 = <put one foot forward, step with the other foot>). Here two actions are nested having no parts in common, but one subgoal (step with other foot) is a precondition to a superordinate goal (arrive at school).
A third way actions may be related is that their component deeds may overlap in time. While walking to school, you chew some gum. The actions are not functionally related: Either goal may be achieved (or abandoned) before the other. One deed may start before (or after) the other.
Fourth, a single goal might be achieved by any number of deeds (or deed-sequences). I may go to school by walking along any of a number of paths. I may take a car or a bus. The deeds vary drastically, but the goal is the same. This relation is usually called equifinality.
Finally, the component subaction of a larger action may be performed by a different actor than is the larger action. This is most common in social action. The team wins the game, due in part to John's scoring a goal. Another possible application involves Lycan's (1988, chap.1) notion of homuncular functionalism. The homuncular functionalist espouses the view that proper subparts of a person, say, that person's visual system, or dynamic balance subsystem might be considered capable of belief or action. The homuncular relation is one between actors; one actor is a proper subpart of another actor.
All of these structural relations among actions and the components of actions have been described elsewhere. The point here is that the notion of an action as the pairing of an event (or, more specifically, a deed or activity) with a further action is a formal way of characterizing all of the above relations.
The Microstructure of Actions
Lee (1988) concentrates on molar description in psychology. However, in order to address my specific concerns about the language of animal learning theory, we need to examine things at a molecular level as well. This raises the question of microstructure.
The notion of microstructure has received a good deal of attention of late, if only because the word microstructure appears in the subtitle of the new "bible" of the newest wave in cognitive science, connectionism ( Rumelhart, D.E., McClelland, J.L., and The PDP Research Group, 1986; McClelland, J.L., Rumelhart, D.E., and The PDP Research Group,1986). In philosophy, related issues are discussed under the rubric of subdoxastic states (Stich, 1978).
The essential notion of microstructure is that larger units, of whatever sort, seem to be composed of smaller units that act as functional intermediaries in the processes involving those larger units. As trivial as this notion may seem, we shall see that it has been ignored in many reputable theories of animal learning. By examining the microstructure of action, we will find a candidate for the basic unit of analysis for learning theories.
Types of microstructure. Cognitivists, of course, worry mostly about cognitive units: thoughts and beliefs and the like. If we wish to examine the structure of learning, we need to look at behavioral units: actions. Roughly speaking, the notion of microstucture can be modeled using the macrostructural relation of one action as a component subpart of another.
Consider the microstructure of response. A prototypical response such as a lever-press consists of a sequence of microactions, each subordinate components of the whole action -- atomic or molecular activities (deeds), each with the goal of making the next microaction possible. (This is what behaviorists call precurrent responding, or chaining.) Just as I step forward now with one foot in order to step again immediately afterward with the other -- my eventual (superordinate) goal being my arrival at school, so the rat must step up to the lever, raise its paw, lower its paw, etc., etc. Mechner (1992) distinguishes between an operant, such as a lever-press, and the sequence of sub-operants of which the operant is composed.
Further, the expression "the rat pressed the lever" is ambiguous as between the richer description of the entire sequence of activities culminating in the paw actually moving against the lever and the thinner description of that ultimate micromovement alone. When does Ned's action of killing the King begin? When the trigger begins to move? When the finger begins to move? When Ned raises the gun? When he steps forward to shoot? When he plans the assassination?
In the real world, however, the environment constantly shifts. The cat chasing the rat in order to obtain food will find the rat far less cooperative than the rat finds the lever. Indeed, in the wild, the rat only occasionally finds grain or garbage in the same location and never finds it in exactly the same location as does the rat finding the lever in the cage.
Even in the laboratory, it is easier to press a fixed lever with one's eyes open. Choose an object in front of you, just slightly out of easy reach. Reach out and touch it. Withdraw your hand. Now, close your eyes and repeat the procedure. Even simple actions are easier when we rely on sensory information.
The reason that the sensory information is helpful is that hand and eye are never in exactly the same position relative to the external object. Even with a fixed environment, the organism's body constantly shifts its orientation. The exact positional relation between the actor's body and the object(s) acted upon is never repeated. Action must be understood in this context. This is a principal argument for equifinality (Brunswick, 1955).
The dynamics of microstructure. In general, we do not blindly execute subactions followed by further subactions, repeating until the superordinate goal is achieved. Rather, we act (i.e., perform a microaction) and then, depending upon sensory feedback, choose among a small set of possible next microactions. In short, we use sensory information to make microscopic mid-course corrections as we head toward our goal.
In his book, Cybernetics, Wiener (1947) gives the classic picture of the microstructure of action:
When we desire a motion to follow a given pattern, the difference between this pattern and the actually performed motion is used as a new input to cause the part regulated to move in such a way as to bring its motion closer to that given by the pattern. For example, one form of steering engine of a ship carries the reading of the wheel to an offset from the tiller which so regulates the valves of the steering engine as to move the tiller in such a way as to turn these valves off. ...
Now, suppose that I pick up a lead-pencil. To do this I have to move certain muscles. However, for all of us but a few expert anatomists, we do not know what these muscles are; and even among the anatomists, there are few if any who can perform the act by a conscious willing in succession of the contraction of each muscle concerned. On the contrary, what we will is to pick the pencil up. Once we have determined on this, our motion proceeds in such a way that we may say roughly that the amount by which the pencil is not yet picked up is decreased at each stage. This part of the action is not in full consciousness.
To perform an action in such a manner, there must be a report to the nervous system, conscious or unconscious, of the amount by which we have failed to pick the pencil up at each instant. If we have our eye on the pencil, this report may be visual, at least in part, but it is more generally kinaesthetic, or to use a term now in vogue, proprioceptive.
It may seem odd that I quote from this hoary old text. But there are important lessons in this particular passage.
First, behaviorists of all stripes have been reluctant (for a number of reasons) to incorporate cybernetic notions into their work. A principal reason is that concern with microstructure is thought to be a step backward to the molecular behaviorism of Watson. I believe that there is an essential role for molecular analysis in the explanation of conduct.
Second, despite an initial flirtation with cybernetic structure (Miller, Galanter, & Pribram, 1960), cognitive psychologists have ignored the holistic, cyclic, dynamic, interactionist implications of that structure and continued in the "linear" tradition of stimulus-response psychology. Indeed, one of von Cranach's (1982) main interests is to reestablish concern for hierarchical cycles in the study of cognition.
von Cranach's effort is laudable. However, he, like most contemporary advocates of a cybernetic approach, misses a key point of the above passage. As noted earlier, von Cranach (1982) restricts his study of action to conscious acts. He claims that other sorts of acts are "found only in rare cases" (p. 36).
In the above passage, Wiener claims that every "goal-directed, planned, intended and conscious behaviour" (to use von Cranach's own phrasing), even one so simple as the picking up of a lead pencil, is composed of many microacts, many of which, if not all, are not conscious. On Wiener's account, for each conscious act, there are dozens, if not hundreds, of non-conscious acts.
A serious consideration of the microstructure of action leads one to the conclusion that it is conscious acts that are proportionately rare. If such is the case, an explanatory strategy that does not make overmuch of the issue of awareness, such as Lee's, may be preferred to an explanatory strategy that requires introspective access, such as Harré's (1982, p. 18) or von Cranach's (1982, pp. 47-48).
Most importantly, as we shall see in the next section, a proper consideration of microstructure will point out the appropriate basic unit of analysis for the study of animal learning. Traditional theories of learning, both behavioral and cognitive, have failed to explain learning by taking their basic unit at the wrong level of organization.
The argument that a unique, basic act-description exists and that the specification of such actions is essential to proper explanations of learning will depend in great part upon an analogy between action and perception. In particular, the argument depends upon an analogy to Stich's (1978) distinction between beliefs and subdoxastic states. I claim that an equivalent distinction exists between actions and what I call subintentional acts.
It may be useful to examine how subdoxastic states differ from beliefs in order to get an idea of how subintentional acts might differ from actions. One of Stich's (1978) examples is that of a person perceiving objects as being at different distances, the process of depth perception. Many different features of the changing visual field, many different stimulations of the retina, are involved in the process that results in the person's belief that object A is behind object B. Stich argues persuasively that none of these subcomponents of belief-producing processes are themselves beliefs.
Intuitively, this makes sense. Suppose Sally believes that the lamp is further from her than the chair. It may very well be that, somewhere deep in her visual subsystem, the disparity between the image of the lamp on one retina and the image of the lamp on her other retina, is calculated. Further, her ultimate belief about the closeness of the lamp may be due, in part, to this calculation within her visual subsystem. Indeed, a proper understanding of the microstructure of perception may require an understanding of those sorts of calculations. Despite all this, it makes little or no sense to say that Sally believes that there is a binocular disparity between the two images on the two retinas.
Subdoxastic states differ from beliefs in two ways, according to Stich. They are inaccessible to consciousness and they are less "inferentially integrated" with the rest of the beliefs (other than the beliefs actually produced). Stich also mentions that the process from one doxastic state to another, or to a terminal belief, lacks the usual equifinality associated with the more ordinary sort of inference.
I suggest we can add a linguistic distinction to the list. We are distinguishing here between the claim that Sally believes the lamp to be behind the chair, and the claim that Sally believes that there is a binocular disparity between the pattern on her left retina and the pattern on her right. The terms lamp and chair, which refer to the objects about which Sally has a belief, are, grammatically speaking, the predicate objects of the belief sentence. Those two terms refer to things outside Sally's body. In contrast, 'pattern on the retina' is an expression that refers only to things within the body, or, more properly, to states of the body. I call this linguistic distinction the predicate object distinction.
As Dan Hunter (personal communication, July, 1991) points out, this distinction suggests that subdoxastic states are epistemologically more basic than perceptual beliefs, since we come to knowledge of the outside world through the acquisition of subdoxa regarding states of our own bodies.
Furthermore, from an evolutionary standpoint, knowledge about the nature of external objects cannot be hardwired into the brain to the same degree as knowledge about the nature of internal bodily states. Lamps and chairs have been around far too short a time for human brains to have evolved specific processes for recognizing them. We must learn about lamps and chairs.
However, the stimulation of rods and cones in the retina and the reorientations of eyeballs, have been around for as long as retinas and eyeballs. There has been plenty of time for the brain to evolve so as to discern important differences in these bodily states. From an evolutionary perspective, an important difference is one that provides an evolutionary advantage. The evolutionary advantage of the ability to acquire knowledge of the outside world is obvious. Hence, it is quite reasonable to assume that brains would evolve to be able to detect exactly those changes in bodily states that provide information about the outside world.
In sum, it appears to matter a good deal whether or not the predicate object of a belief sentence makes reference to something outside the subjects body.
Are microactions actions?
All in all, dividing actions into functional subparts seems less of a problem than dividing the process of belief formation/acquisition into the processing of subdoxa. However, just as Stich (1978) raises the question of whether those cognitive intermediaries are truly beliefs, the analogous question needs to be considered. Are these behavioral intermediaries, the microactions, actually actions? Or do we need to distinguish a separate classification, that of subintentional act, for some of these microactions? If so, where do we draw the line between the two?
Austin (1962) makes the clearest statement of this problem. In discussing the rich-thin continuum of action noted above, he concludes (with some fervor) that, at some level, we must decide that such an intermediary is, in fact, not a proper action in some sense:
With physical actions we nearly always naturally name the action not in terms of what we are here calling the minimum physical act, but in terms which embrace a greater or less but indefinitely extensive range of what might be called its natural consequences (or, looking at it another way, the intention with which it was done).
We not merely do not use the notion of a minimum physical act (which is in any case doubtful) but we do not seem to have any class of names which distinguish physical acts from consequences ....
Note that if we suppose the minimum physical act to be movement of the body when we say 'I moved my finger', the fact that the object moved is part of my body does in fact introduce a new sense of 'moved'. Thus I may be able to waggle my ears as a schoolboy does, or by grasping them between my finger and thumb, or move my foot either in the ordinary way or by manipulating with my hand when I have pins and needles. The ordinary use of 'move' in such examples as 'I moved my finger' is ultimate. We must not seek to go back behind it to 'pulling on my muscles' and the like.
In the above passage, Austin makes two important points. In terms of the macrostructure of conduct, he notes that we have no usual way in common parlance English to identify a bodily movement distinct from its intended consequences. This supports Lee's view that the essential components of action are means and end. More importantly, in terms of animal learning, this lacuna in English has given rise to some confusion about what should count as a 'response' in learning theories.
However, it is Austin's second point that concerns us here. That point involves the microstructure of actions. Austin claims that the absence of any way (again, in common parlance English) of talking sensibly about a person as the do-er of microactivities (consider Wiener's "conscious willing in succession of the contraction of each muscle" above) means we must reject the notion that such subintentional acts are actions by the person.
Austin's conclusion leads to a problem. If subintentional acts are not proper actions, then they must, on my theory of action, lack some essential feature of being an action. On the analysis given, they certainly do not lack goals. And they certainly involve bodily movements. Just as Lycan (1988) suggests that the subdoxastic states of the person may be the beliefs of one of that person's proper subparts, I claim that subintentional acts are not proper actions because they are not performed by the person. Rather, they are performed by some subpart of that person. Returning to Austin's example above, I am the one who crooks my finger, but it is not me, rather it is some motor-control subsystem of mine, that pulls on the muscles.
Finally, consider the predicate object distinction. It seems to make sense for both beliefs and actions. Austin makes use of precisely this distinction in defining the "minimum physical act." In Austin's view of a minimum physical act, we note that the "thinnest" construal of the action, the limiting case, that of crooking the finger, is describable solely in terms of the body; whereas pulling the trigger, firing the gun, shooting the King, and killing the King (etc., etc.) all involve the body in relation to external objects.
I claim that the distinction regarding predicate objects is the crucial one both for determining what is an action (in the behavioral case) and for determining what is a belief (in the cognitive case). Other distinctions have merit, but a critical comparison will have to wait for another time.
The Basic Atomic Action
The way is now paved for the promised specification of a basic unit of analysis, a basic atomic action. The basic atomic action we are searching for should be the sort of microelement that composes the microprocesses that Wiener speaks of in the quoted passage. In order to find these basic elements of microstructure, we begin with the macrostructure and work our way down the most likely looking continuum, the richness-thinness continuum.
Travelling down this continuum, we find a sudden barrier to common sense. For the very next thinner act-description, the person does not appear to be the actor, the performer of the act. Taking this as our cue, we search nearby for a unique sort of act-description. We discover that the thinnest proper description of an act has three unique characteristics: (a) The predicate object of the action sentence (the thing done to) is always a part of one's body. (b) As Austin notes, the sentence describing the act is thoroughly ambiguous, admitting of two interpretations. Either the body part is moved "in the ordinary way," or by means of manipulation (directly or indirectly) by another body part. (c) When we isolate the first of these two alternatives, that is, moving one's body part "in the ordinary way," we discover a final mark. "The ordinary way" is only one particular way. Unlike almost any other act-description, it admits of almost no equifinality.
Returning to our recursive definition of action as a means-end pair, we see that, in the new formulation, the thinnest action looks mighty strange:
"Ned crooked his finger" (action1) =
<"Ned's finger crooking" (deed), "Ned has his finger crooked" (action2) ? >
This is the most extreme example of what Lee (1988, p.59) refers to as "end results bound closely to behavior." In short, Ned crooked his finger in order that his finger be crooked. It is exactly the strangeness of this formulation that tells us that we have found the elemental act. (More precisely, this is how we can use the action theoretic notation to specify an act as elemental.) Further, it is a way of eliminating the threat of infinite regress, a problem for any formulation using a recursive definition. 
The most important moral we can take from Austin is that English leads us astray at a critical moment. In English, we have no ordinary way of distinguishing the action wherein Ned crooks Ned's finger from the bodily movement that instantiates that action. The closest we can come is to designate the expression Ned crooks his finger as the action and the expression the crooking of Ned's finger as the bodily movement (deed).
However, this transformation between present indicative active and gerund is more notational convenience than genuine ordinary English usage. The language remains unclear. As does Lee, I begin with common parlance distinctions. But grammar alone cannot determine the nature of psychological phenomena. Theoretical commitments are unavoidable.
The Inadequacy of Molar Terms
It remains to show that not only are the usual act-descriptions of molar behaviorism and cognitive psychology not basic, but that those descriptions are inadequate for the purpose of psychological explanation.
Molar terms. The molar behaviorist takes the normal distal description of an act, like a rat's pressing a lever or a pigeon's pecking a key, as both the basic unit of empirical measurement and as the basic unit of theoretical analysis. This unit is called the molar response. Mainstream cognitivists have accepted molar specification of basic terms. The contemporary learning theorist takes the rate of occurrence for some molar response as the empirical measure of learning. This measure is called response rate. The basic theoretical construct said to correlate with response rate is called response strength.
Molar behaviorism, so-called, is the thesis that the normal distal decription of an action is the only one that need concern the psychologist. By restricting descriptions of stimulus and reponse to a single position along the richness-thinness continuum, the neobehaviorists hoped to establish a more or less unambiguous language for recording the behavioral regularities they hoped to discover in the laboratory. The claim was made that the appropriate position along the richness-thinness continuum could always be distinguished in terms of the precise degree of equifinality just broad enough to encompass all the bodily movements of interest and to exclude all those not of interest.
However, molar responses are not basic actions. As Austin (1962) notes above, the normal distal description of an act all but inevitably involves reference to physical objects outside the animal's body. The rat presses a lever and the pigeon pecks a key.
The problem of response strength. Just as humans have not had time to evolve mechanisms for recognizing lamps and chairs, neither have we had time to evolve mechanisms for turning on lamps or sitting in chairs. We have had time to evolve mechanisms for extending our arms, crooking our fingers, bending our legs, adjusting our balance, etc.
What we have not evolved to do, we must perforce learn to do. On the structural view advocated here, what we learn must be composed of new relations amongst those things we already know. Hence, learned actions are composed ultimately of basic innate actions, that is, minimum physical acts.
In short, molar responses are themselves learned entities. Whatever the true nature of molar responses, any theory that purports to be a theory of learning must explain where molar responses come from. That is something that traditional learning theories do not explain. From this perspective, any learning theory that makes predictions solely in terms of increases or decreases in rates of molar responding is not a theory of learning per se, but rather a model of one aspect of learning.
The traditional reply of the molar behaviorist to the above criticism is that all of behavior is ultimately understandable in terms of changing rates of responding across all behaviors. Even if this claim is true (as I believe it is), the molarist reply fails. The examination of microstructure, above, indicates that molar responses are quite unlike minimum physical acts. We have no reason to assume that rates of molar responding are governed by the same laws as rates of basic actions.
Molar responses are highly structured complexes consisting (at least in part) of basic actions. Hence, the rates of basic actions we are concerned with are rates conditional on the occurrence of previous basic acts. Molarists might claim that rates of molar responding conditional only on the occurrence of stimuli can tell us about rates of basic actions conditional upon other basic actions as well as stimuli. Any such claim should be supported by mathematical argument. Presently, it is not.
Given these assumptions, proper explanation of any psychological phenomenon consists in showing how microactional sequences of basic atomic actions account for that phenomenon.
Indeed, the molarist case is in worse shape than the above suggests. Consider the collection of all possible microaction sequences. We might think that a specific molar response class, such as a lever-press, consists of a specific group of such microsequences. This is not the case. If the rat is located at one particular spot, say to the left of the lever, there does exist a specific (albeit large) group of microsequences, any of which could effect a successful lever-press. On the other hand, if the rat is located elsewhere, say to the right, none of those microsequence will do. A distinct group of microsequences is required. Further, as was noted above, in the wild, a rat is never in exactly the same place twice.
Finally, if we refocus our attention on the notion that behavior is that which the subject does, there is a sense in which a molar response is not something done by the subject at all. Molar terms describe basic actions taken in the context of specific environmental affordances. Said affordances are produced, in part, by previous basic actions. Despite this, environmental affordances are simply not anything the subject does.
All of the above makes for a rather discouraging picture of the prospect of explanatorily adequate theories of learning. Nevertheless, I am optimistic. A detailed proposal for a computational theory of learning must wait for another time. However, the example of the pigeon given below should both help to clarify the above issues and suggest the eventual form of such a computational theory.
In this section, I will take a specific set of well-known data and reformulate it in terms of the new language proposed. This reformulation will show why certain problems have arisen in prior treatments of the phenomena. It will also give a practical illustration of the new problems that come with the new language.
I have chosen the specific case of the operant conditioning of the pigeon. I choose this data because it is, in some sense, straightforward and well understood. (I hesitate to describe any behavior as "simple.") Despite its apparent simplicity, a good theoretical account is unavailable at present.
The Experimental Paradigm
To begin with, a description of the procedure is in order. For the experiment, the pigeon is taken from its home cage and placed in the operant chamber. There is usually a light that can illuminate the entire chamber. Typically, one wall of the chamber contains the apparatus -- usually, one or more keys and a food-dispensing tray.
A key is a button on the wall that covers an electrical switch. Contact is made when the key is pressed. The button is usually made of a translucent material and can be lit from behind. By action of a mechanism behind the wall, a food pellet can be dropped into the food-dispensing tray.
All of the apparatus described, buttons, chamber lights, key lights, food dispenser, etc., are operated by a mechanism behind the wall of the cage. A fixed logical procedure or program is set up in that mechanism to operate each part of the apparatus under specified conditions. For instance, food might be dispensed on every third peck of the key; or one for every peck, given that the key is lit from behind with a red light; or only when the chamber is lit throughout.
The various logical relations in the program or procedure are called contingencies. They are intended to model the contingent demands and opportunities of a real-world environment. In practice, the procedure or program, called a schedule, often becomes very complex. The basic result, however, is simple. The pigeon comes to conduct itself so as to acquire the food. Within the limits of its capacity for learning, it can adapt to a wide variety of complex contingencies, learning to cope with those demands and thus gaining access to food provided by the food dispenser.
Finding the Basic Unit of Analysis
As in the general case, I believe that it is an essential first step to the construction of any theory to identify the basic unit of analysis. As a means of finding the basic unit, my current prejudice is for what I have called the predicate object test. The predicate object test involves finding the richest act-description wherein that which is acted upon, is a part of the actor's body. That description is the description of the basic action.
Applying the predicate object test. In the case of pigeon learning, traditional learning theories take the keypeck as the basic unit of analysis. The rate of keypecking is the principle empirical measure. The strength of the keypeck response is the fundamental theoretical construct. By keypeck, I take these theoreticians to mean the pigeon thrusting its head forward so that the beak depresses the key, closing the electrical circuit behind. Of course, the action sentence: The pigeon pecks the key. has a predicate object, the key, outside the pigeon's body.
Herein lies the dilemma. The critical measure of the pigeon's conduct, as well as its learning, is the closing of the electrical switch behind the key. The precise, and indeed, the only molar action description that captures exactly the right portions of the pigeon's behavior is the keypeck. The keypeck is the bit of behavior, the action, that the experimenter has experimented upon. Predictions about the keypeck are the only sort of predictions in which the theoretician could have the slightest interest. Yet the keypeck is not a basic act description.
The solution is simple. Examine the microstructure of the keypeck. Decompose it into basic acts. Only two types of action are needed. (a) The pigeon reorients its head. (b) The pigeon pecks. (That is, the pigeon thrusts its head forward.) Both of these actions are basic by the predicate object test. The predicate object in both cases is the head. Every keypeck can be thought of as a sequence of head orientations (until the key is directly in front of the beak) followed by a peck. In Mechner's (1992) terms, the keypeck is an operant with the peck as its terminating sub-operant. The microstructure of behavior is revealed by analysis of the sub-operants.
In the process of uncovering the basic actions underlying the keypeck, we have described the sequence leading up to the actual pecking of the key in terms of Wiener's (1947) cycle of alternations between micromotion and environmental assessment. This picture also corresponds to Gibson's (1972) view of vision as an active process: retinal inputs leading to ocular adjustments leading to altered retinal inputs, etc. Each reorientation of the head also moves the eyes with respect to the external environment, thus altering any subsequent retinal input. Any new information from this altered retinal input is available for the pigeon to use in the determination of the next microaction, be it another head reorientation, a peck, or some other act irrelevant to the experiment. Each bit of new information constitutes a new opportunity for the pigeon to reach the key.
The goals. Now, consider the macrostructure of this conduct. Due to the peculiarity of English, noted earlier, in individuating the minimum physical acts, we have also picked out the basic bodily movements involved. These are the deeds, the means. At the thinnest level of analysis, each head orientation is done so that the head be reoriented and each peck so that the beak move forward.
The head orientations are also done in order to provide new retinal inputs (information opportunities) to the pigeon. They are also done in order to move the beak into a position in front of the key. The peck probably is done to satisfy some innate propensity to peck in the pigeon as well as another to peck bright (or brightly lit) objects. (This propensity is revealed in the phenomenon of animal learning known as autoshaping.) However, as learning progresses, the peck is done in order to press the key, in order that food be provided, in order that the pigeon may eat. (We may stop with the eating as this may be treated, for our purposes here, as a terminal, irreducible goal.)
Simply put: every keypeck is a peck, but not every peck is a keypeck. In order for a keypeck to occur, at least the following must be true: a key must exist, the animal must be capable of pecking (a snail just will not do), the animal must be positioned with its beak in front of the key, the key must be peckable by the animal. Peckable means that the key is not held rigid by a clamp or covered by a peck-proof shield, and that the animal is not harnassed so as to be unable to thrust its head forward, etc. Peckability is an affordance.
The Question for Action Theories of Learning
In stimulus-response psychology, the question to be answered by any learning theory is clear. What is the functional relation between environmental events and the rate at which molar responses appear? Since I claim that a theory based on these molar responses is necessarily inadequate, the question must be respecified for the new sort of theory proposed here.
Linguistic issues. As suggested above, there is a sense in which the keypeck is not something the pigeon does. In English, we use terms such as keypeck to describe the pigeon's conduct. As Austin's analysis shows, terms such as keypeck inevitably confuse the subject's deeds per se with those same deeds understood in the broader context of action. Logically speaking, a keypeck occurs if and only if two things are true: a) the environment provides the appropriate affordance, and b) the pigeon pecks. While the environmental affordance is due, in part, to the previous conduct (i.e., precurrent responding) by the pigeon, the environmental affordance is, notwithstanding, not anything that the pigeon does. The keypeck is an interaction between organism and environment.
Learning, in essence, is the process by which the organism is changed by the demands of the environment. Our only measure of the state of the organism and, thus, our only measure of any change in that state, is the doings of the organism. If common-parlance descriptions of conduct cannot distinguish between the doings that measure learning and the interactions that cause further learning, then those self-same common parlance descriptions cannot serve in theories of learning.
In sum, normal distal act descriptions (NDADs) pick out molar responses, which are, or should be, the basic empirical unit of measurement. Molar responses and the rates thereof are our only data for the study of behavior and learning. Toponymic act descriptions (TADs, expressions making reference only to parts of the body) pick out basic actions, which are, or should be, the basic theoretical unit of analysis. Basic actions are the only true measure of the underlying process we call learning. Basic actions explain molar responses.
As with all psychological processes, the relevant phenomena provide only an indirect measure. The challenge to psychological theory is to analyze the underlying processes given only the available data.
Psychometric issues. The intuitive appeal of traditional learning theories is due to the same thing that renders those theories useless as psychological explanation: NDADs are powerful linguistic devices that capture behavior with respect to the precise context of interest. Traditional learning theories, expressed using NDADs, appealingly suggest that all of conduct can be segmented neatly into molar responses and their rates. They do this at the cost of conflating environmental affordances and organismic deeds. That conflation leads the molarist to perpetually misattribute the environmental contributions to conduct to some source within the organism. This fundamental attribution error (Jones & Nisbett, 1992) is the mistake of mentalist and cognitivist alike.
All the same, to the degree that traditional learning theories can be said to explain learning at all, they do so using a powerful language unavailable to the proponent of a microstructural approach like the one advocated here. The molarist is obliged to show why laws governing molar responses also apply to other sorts of behavioral units. Similarly, a proponent of the present approach is obliged to show how the weaker language based on TADs can explain conduct understood within the broader contexts of personal goals, social settings, and cultural surround (Harré, 1982).
The solution is to consider learning as akin to a process of statistical approximation (Dennett, 1981). Assume that, for any organism, there is an ideal behavioral function, expressible in molar stimulus-response terms, that best allows the organism to cope with its environment. Learning is best understood as the processes whereby the behavioral propensities of the organism, espressed in terms of microsequences of basic actions, come to approxiamte that ideal.
In the new view of learning, conduct is a constant cyclic interaction between environment and organism. The environment is disposed to react in various ways to various acts. These environmental dispositions are called contingencies. The organism is also disposed to act in various ways to various environmental events. These organismic dispositions are called propensities or repetoire.
The question for learning theorists is to see what function can model the changes in repetoire actually occurring in the animal. Modeling the function must begin with a realistic model of the environment-organism interaction along with a realistic model of environmental contingencies. The function must gradually alter the model organism's repetoire in such a manner that the pattern of changing conduct matches, in all of the manifold ways possible, the pattern of changing conduct of animals in the laboratory and ultimately in the real world.
In conclusion, I am a molar behaviorist as regards description and observation, but a molecular behaviorist as regards theoretical explanation. To require the specification of a particular sequence of microacts in explaining composite action is to abandon equifinality in a futile search for reductive explanation of action. Instead, the proper account of action involves postulating a statistical sample from a population of microaction sequences that could account for acts within the given action class in principle.
The potential for Lee's action theory extends far beyond the application made here. Planned directions include arguments for the elimination of the stimulus-response distinction in theoretical accounts of learning and a computer simulation technique for solving the psychometric problem of comparing competing theories of animal learning.
Anscombe, G. E. M. (1963). Intention. (2nd Edition.) Ithaca, NY: Cornell University Press.
Austin, J. L. (1962). How to Do Things with Words (2nd ed.; J. O. Urmson and M. Sbisà, Eds.). Cambridge, MA: Harvard University Press.
Brand, M. (1984). Intending and Acting. Cambridge, MA: MIT Press/Bradford Books.
Brunswick, E. (1955). The conceptual framework of psychology. In O. Neurath, R. Carnap, & C. Morris (Eds.), International Encyclopedia of Unified Science, 1, (pp.655-760, also published separately, 1952). Chicago: University of Chicago Press.
Cranach, M. von (1982). The psychological study of goal-directed action: basic issues. In M. von Cranach & R. Harré (Eds.), The Analysis of Action, (pp.35-73). Cambridge: Cambridge University Press.
Davidson, D. (1963). Actions, reasons, and causes. Journal of Philosophy, 60, 685-700.
Davis, L. (1979). Theory of Action. Englewood Cliffs, NJ: Prentice Hall.
Dennett, D. C. (1981). Three kinds of intentional psychology. In R. A. Healey (Ed.), Mind, Psychology, and Reductionism. Cambridge: Cambridge University Press.
Gibson, J. J. (1960). The concept of stimulus in psychology. American Psychologist, 15, 694-703.
Gibson, J. J. (1972). A theory of direct visual perception. In J. R. Royce and Wm. W. Rozeboom (Eds.), The Psychology of Knowing (pp. 215-232). New York: Gordon and Breach.
Goldman, A. I. (1970). A Theory of Human Action. Englewood Cliffs, NJ: Prentice Hall.
Hamblin, C. L. (1987). Imperatives. Oxford: Basil Blackwell.
Harré, R. (1982). Theoretical preliminaries to the study of action. In M. von Cranach & R. Harré (Eds.), The Analysis of Action (pp.5-33). Cambridge: Cambridge University Press.
Jones, E. E. & Nisbett, R. E. (1972). The actor and the observer: Divergent perceptions in the causes of behavior. In E. E. Jones, D. E. Kanouse, H. H. Kelley, R. E. Nisbett, S. Valins, & B. W. Weiner (Eds.), Attribution: Perceiving the causes of behavior (pp. 37-52). Morristown, NJ: General Learning Press.
Koch, S. (1954). Clark L. Hull. In W. K. Estes, S. Koch, K. MacCorquodale, P. E. Meehl, C. G. Mueller, Jr., W. N. Schoenfeld, & W. S. Verplanck (Eds.), Modern Learning Theory (pp. 1-176). New York: Appleton-Century-Crofts.
Lee, V. L. (1988). Beyond Behaviorism. Hillsdale, NJ: Lawrence Erlbaum Associates.
Lycan, W. G. (1988). Judgement and Justification. Cambridge: Cambridge University Press.
McClelland, J.L., Rumelhart, D.E., & The PDP Research Group (1986). Parallel Distributed Processing: Explorations in the microstructure of cognition, Volume II: Psychological and biological models. (Cambridge, MA: Bradford Books/MIT Press)
Mandler, G. & Kessen, W. (1959). The Language of Psychology. New York: John Wiley & Sons.
Mechner, F. (1992). The revealed operant: A way to study the characteristics of individual occurrences of operant responses. Cambridge, MA: Cambridge Center for Behavioral Studies.
Meehl, P. E. (1978). Theoretical risks and tabular asterisks: Sir Karl, Sir Ronald, and the slow progress of soft psychology. Journal of Clinical and Consulting Psychology, 46, 806-834.
Miller, G. A., Galanter, E., & Pribram, K. H. (1960). Plans and the Structure of Behavior. New York: Henry Holt.
Oppenheimer, L. (1991). The concept of action: A historical perspective. In L. Oppenheimer & J. Valsiner (Eds.), The Origins of Action: Interdisciplinary and international perspectives, (pp.1-35). New York: Springer-Verlag.
Palya, W. L. (1989). Research Methods Lecture Notes. Unpublished manuscript. Jacksonville State University, Southeastern Behavioral Analysis Center, Jacksonville, AL.
Rumelhart, D.E., McClelland, J.L., & The PDP Research Group (1986). Parallel Distributed Processing: Explorations in the microstructure of cognition, Volume I: Foundations. Cambridge, MA: Bradford Books/MIT Press.
Skinner, B.F. (1969). Contingencies of Reinforcement: a theoretical analysis. New York: Appleton-Century-Crofts.
Skinner, B. F. (1974). About Behaviorism. New York: Alfred A. Knopf.
Stich, S. P. (1978). Beliefs and subdoxastic states. Philosophy of Science, 45, 499-518.
Turner, M. B. (1967). Philosophy and the Science of Behavior. New York: Appleton-Century-Crofts.
Wiener, N. (1947). Cybernetics: or control and communication in the animal and the machine. New York: John Wiley & Sons.
Wittgenstein, L. (1953). Philosophical Investigations, translated by G. E. M. Anscombe (3rd Edition, 1973). New York: MacMillan.
Ziff, P. (1958) "About behaviorism." Analysis, XVIII. (Reprinted in Philosophical Turnings: Essays in conceptual appreciation by Paul Ziff, pp. 155-160. Ithaca, NY: University of Cornell Press, 1966.)
Thanks to Dave Eckerman, Bill Lycan, Danna Cornick, Stella Sutter, Jaan Valsiner, Bill Knorpp, Eve Segal, George Skelly, Dick Smyth, Dan Hunter, and others too numerous to mention, who have patiently listened to my mutterings on this topic for the last fifteen years or so. And to the even better friends who have refused to listen. The reader should also note a less personal, but equally important debt that I owe to the writings of Gregory Bateson and C. S. Peirce. Thanks most of all to my mom, Mrs. Edie Kemp, for her support during my long illness, said support making this paper possible.
This paper originally appeared as Chapter 11 (pp. 306-328) of Valsiner, J. & Voss, H-G. (Eds.). (1996). The Structure of Learning Processes. Norwood, NJ: Ablex.
 V. L. Lee (personal communication, August 19, 1991) comments that "means can be actions... and not just bodily activities." This constitutes a difference between my view and Lee's.
Actually, the problem of infinite regress is quite a bit
more complex than this. For instance, in response to a question: Why did
Ned crook his finger? a legitimate answer might be: In order to start
the revolution. Thus, reasons and intended consequences might lead us
outward to the richest description instead of inward to the thinnest. The
problem of infinite regress is, I believe, soluble, but a detailed exposition
of this must wait for another time.