Teleological Behaviorism

Howard Rachlin State University of New York at Stony Brook

A psychological science of efficient causes, using internalmechanisms to explain overt behavior, is distinguishedfrom another psychological science, based on Aristotelianfinal causes, using external objects and goals to explainovert behavior. Efficient-cause psychology is designed toanswer the question of how a particular act is emitted;final-cause psychology is designed to answer the questionof why a particular act is emitted. Physiological psychol-ogy, modern cognitive psychology, and some parts of be-haviorism including Skinnerian behaviorism are efficient-cause psychologies; final-cause psychology, a developmentof Skinnerian behaviorism, is here called teleological be-haviorism. Each of these two conceptions of causality inpsychology implies a different view of the mind, hence adifferent meaning of mental terms.

The object of this article is to present a view of finalcauses (teleological explanation) as conceived by Aristotle(ca. 350 B.C./1984), to show how a form of modern be-haviorism, which I call teleological behaviorism, makesuse of final causes, and to argue that this form of behav-iorism provides both a vehicle for prediction and controlof behavior and a potential meaning for mental terms atleast as useful as that potentially provided by physiologicalor cognitive psychology.

There is much disagreement among modern philos-ophers of science about what is a final cause and aboutwhat Aristotle meant by final cause or teleological expla-nation. See Davidson (1980), Donagan (1987), Dretske(1988), Ringen (1985), Taylor (1964), Staddon (1973),and Wright (1976) for discussions of causation in psy-chology and modern interpretations of Aristotle. The de-tails of the philosophical debate are far beyond the scopeof this article; therefore, the present exposition may beregarded less as an interpretation of Aristotle's actualviews than as a reconstruction of those views for the pur-poses of modern behaviorism.

According to at least one contemporary Aristotelianphilosopher (Randall, 1960, p. 66): "For Aristotle life orpsyche is the behavior of the organism as a whole in itsenvironment. Aristotle is thus a thoroughgoing behav-iorist." Kantor (1963, p. 161) considered behaviorismthe modern movement closest to Aristotle's "naturalistic"psychology. Thus, there is at least some basis in philosophyfor the present approach.

Final CausesAristotle's concept of cause was much wider than themodern one. For him, a causal explanation of a process

was an answer to a question about the process, whatevermight follow the word "because" in a sentence (Randall,1960, p. 124).

Aristotle referred to four types of causes: material,formal, efficient, and final. Material and formal causesexplain the nature of "substances" (static objects); efficientand final causes explain the dynamic behavior of objects(inanimate objects as well as organisms). Although weare mainly concerned here with dynamic processes—hence, efficient and final causes—it is worth noting therelation between the two static causes.

For Aristotle, a substance consisted of matter takingon a certain form. Thus, a circle as a substance might besaid to consist of a piece of cardboard taking on the al-gebraic form a2 + b2 = r2 (an example Randall, 1960,attributed to Spinoza). We may say "This piece of card-board is a circle because it has the form a2 + b2 = r2."However, no substance ever perfectly fits its form (no pieceof cardboard can be a perfect circle). For Aristotle, formswere actually categories into which various material ob-jects may be classified; science is the process of classifi-cation. Randall (1960, p. 51) quoted John Dewey as fol-lows: "Classification and division are [for Aristotle]counterparts of the intrinsic order of nature."

Now let us consider the two dynamic causes, efficientand final. Final causes are to efficient causes as formalcauses are to material causes. In our modern way ofthinking, causes precede their effects. Aristotle's efficientcauses do precede their effects. For instance, Aristotlestated in De Anima (Book II, chap. 12, 424a) that oursense organs are affected by forms of objects "the way inwhich a piece of wax [the organ] takes on the impress ofa signet ring [the form of the object] without the iron orgold [the matter]." When discussing the effects of objectson sense organs Aristotle did use efficient-cause expla-nations. However, most of De Anima (On the Soul) isdevoted not to this subject but to the relation betweenobjects and whole organisms. Such processes are labeledwith the familiar terms sensation, perception, imagina-tion, and thought and are explained in terms of final ratherthan efficient causes.

The relation of final causes to their effects parallelsthe relation of formal causes to their effects: A final cause

This article was prepared with the assistance of a National Institute ofMental Health grant. Hugh Lacey and Albert Silverstein provided ex-tremely helpful criticism of an earlier draft in a different form.

Correspondence concerning this article should be addressed toHoward Rachlin, Psychology Department, State University of New "Vbrk,Stony Brook, NY 11794.

November 1992 • American PsychologistCopyright 1992 by the American Psychological Association, Inc. 0003-066X/92/S2 00Vol. 47, No. II, 1371-1382

1371

is a form of movement (or classification of movement)abstractly conceived; its effect is a particular movement.

Relation of Final Causes to Their Effects

A final cause is to its effects as a wider concept (like adance) is to the particulars (like the steps) that make itup. Ackrill (1980) compared the act of buying a club toplay golf with putting to play golf. Buying a club is in-strumental in a way that putting is not; putting is part ofthe game itself. Thus, playing golf would not be an in-clusive final cause of buying a club but it would be ofputting. Similarly, playing a symphony would be a finalcause of playing the first movement but not of enteringthe concert hall. A final cause, such as a dance or playinggolf or playing a symphony, thus may be said to "em-brace" its particular effects (steps, shots, movements). In-strumental acts not embraced by the larger concept (suchas paying the fiddler, buying a club, or entering a concerthall) are not inclusive effects of that final cause, but theymay be effects of a still wider, still more abstract finalcause.

Analysis of final causes yields ends, ends that consistof abstract patterns of the movements that comprise them,ends that embrace those patterns. Final causes are notsimply efficient causes in reverse. An effect of an efficientcause follows its cause, but an effect of a final cause doesnot precede its cause; it fits into its cause. In some sense,a particular movement must occur first in order for apattern of movements to emerge just as the movementsof a symphony must be played before the symphony canbe said to have been played. In that sense and in thatsense only, a final cause follows its effects.

Loosely speaking, efficient causes are answers to thequestion, How does this or that movement occur? Analysisof efficient causes ultimately yields "mechanisms" thatmay range from simple billiard-ball-like interactions tocomplex computer circuits to complex neurochemicalprocesses (Staddon, 1973). Correspondingly final-causeanalyses are attempts to answer the question, Why doesthis or that movement occur? Of what more molar processdoes this particular movement form a part? Answers tothe "how" question, regardless of their completeness, donot automatically answer the "why" question. Accordingto Aristotle (Physics, Book II, chap. 7, 198a), it is thebusiness of physicists to know all the causes of their objectof study. Because psychology, according to Aristotle, is abranch of physics, he would have argued that psychologistsought to know all the causes, final as well as efficient, ofthe behavior of organisms. To illustrate how final causesexplain mental terms, let us consider Aristotle's accountsof sensation and imagination.

Sensation and Imagination in De Anima

A particular sensation was conceived by Aristotle as aparticular instance of a discriminative pattern. The dif-ference between the pattern of a person's actions (the per-son's soul) before sensation and afterward is not that the

soul contains something it did not have before but thatafterwards any complete description of the person's ac-tions must contain a description of aspects of the object(its sensory qualities).

What has the power of sensation [a whole living animal] is po-tentially like what the perceived object is actually; that is, whileat the beginning of the process of its being acted upon the twointeracting factors [the behavior of the animal and the qualitiesof the sensible object] are dissimilar, at the end the one actedupon [the behavior of the animal] is assimilated to the other[the color, smell, sound, and so on of the object] and is identicalin quality with it. {De Anima, Book II, chap. 5, 418a)

The process Aristotle described is behavioral discrimi-nation. Objects may be classified according to their colors,for example. The animal's discrimination enables an ob-server of the animal to classify its behavior in the sameway (by the colors of objects).

Sensing the color of an object is the act of discrim-inating between that color and others. If a driver generallystops at red lights and goes at green lights, the driver maybe said to discriminate between red and green. Giventhat this discrimination (and other discriminations) be-tween red and green do generally occur, then in a partic-ular instance, when the driver actually does stop at a redlight, we may say that the driver sensed the color of thelight. We can then classify that bit of behavior under thatparticular discrimination. Doing so (making the classi-fication) is an act {our act) of matching behavior and color.Stopping at red lights, saying "red," underlining thingsin red, and so on are discriminations of the same sort.The investigation and analysis of sensation is the inves-tigation and analysis of this process (of discrimination).

Aristotle's conception of imagination followed fromhis conception of sensation. Aristotle stated, "Imaginationmust be a movement resulting from an actual exercise ofa power of sense" {De Anima, Book III, chap. 3, 429a).That is, as far as the actual movements of the animal areconcerned, imagination is the same as sensation. The dif-ference is that the objects of sensation are present in theworld during sensation, whereas during imagination theobjects are not present in the world. Note that Aristotledid not say that the objects that are present in the worldduring sensation (which sensation discriminates among)are present inside the animal (as representations, internalimages, neural discharges, or anything else) during imag-ination. What he did say is that the movements the animalmakes during imagination with the objects absent are thesame as those the animal makes during sensation withthe objects present except in that one respect. If you gen-erally behave one way in the presence of, and anotherway in the absence of, red lights, you are discriminatingbetween red lights and other things. However, if, on oc-casion, you behave in the absence of a red light as younormally do in its presence, you are on that occasionimagining a red light.

For Aristotle, imagining was acting and not dream-ing: Vividness of imagination was not vividness of interiorimage but of overt behavior. Suppose two people are asked

1372 November 1992 • American Psychologist

to imagine a lion present in the room. One closes hereyes and says, "Yes, I see it, a mane and a tail, that'sright, it's walking around," and so on. The other runsscreaming for the door. For Aristotle, the first person isnot really imagining the lion but just striking a pose (asbad actors do). The second person is really imagining thelion. The location, intensity, orientation, or even the ex-istence of an image in that person's head would be entirelyirrelevant to the process. A good imagination is not justan aid or a tool in good acting. Rather, for Aristotle, goodacting is good imagining.

Aristotle's final-cause analysis of perception, mem-ory, and creative thought in DeAnima and his discussionof freedom, responsibility, and all of ethical behavior inNichomachian Ethics follows a similar form: A particularact is identified as a particular perception or thought, asfree or unfree, as good or bad not on the basis of theparticular internal acts (spiritual, cognitive, or physio-logical) that may efficiently cause it but rather on thebasis of the temporally extended pattern of overt behaviorinto which the particular act fits, that is, on the basis ofits final cause.

As regards introspective knowledge of one's ownmental states, one Aristotelian interpreter stated,

Aristotle has no reason to think that psychic states—perceptions,beliefs, desires—must be transparently accessible to the subject,and to him alone. Even if there are such states, this feature ofthem is not the feature that makes them psychic states. Psychicstates, for human souls as for others, are those that are causallyrelevant to a teleological explanation of the movements of aliving organism. (Irwin, 1980, p. 43)

Causation in PhysicsFor Aristotle, the motion of all physical objects was ex-plicable in terms of final as well as efficient causes. Astone, according to Aristotle, naturally moves toward thecenter of the earth unless hindered. As long as the stonedoes indeed move toward the center of the earth, itsmovement may be explained in terms of its own endswithout reference to other substances. However, if thestone should be thrown up in the air by a boy or a volcano,the movement of the stone would have to be explainedaccording to ends other than those ascribable to it alone:those of the boy or the volcano (in that sense the stonewas not "free"). Talking about the ends of a stone'smovement sounds odd to us because the success of Re-naissance physics with its rigorous insistence on efficientcauses has given us our modern conception of what doesand does not constitute proper scientific language.

According to Rorty (1982),

Galileo and his followers discovered, and subsequent centurieshave amply confirmed, that you get much better predictions bythinking of things as masses of particles blindly bumping eachother than by thinking of them as Aristotle thought of themanimistically, ideologically, and anthropomorphically. They alsodiscovered that you get a better handle on the universe by think-ing of it as infinite and cold and comfortless than by thinkingof it as finite, homey, planned, and relevant to human concerns.Finally, they discovered that if you view planets or missiles or

corpuscles as point-masses, you can get nice, simple predictivelaws by looking for nice simple mathematical ratios. These dis-coveries are the basis of modern technological civilization. Wecan hardly be too grateful for them. But they do not, pace Des-cartes and Kant, point out any epistemological moral. They donot tell us anything about the nature of science or rationality.In particular, they did not result from the use of, nor do theyexemplify, something called "scientific method." (p. 191)

In this passage, Rorty implied that in physics final-causeexplanations have less predictive power than efficient-cause explanations. Whatever their weakness in physics,this article argues that in psychology final-cause expla-nations may have as much or more predictive power thanefficient-cause explanations. Even in modern physics, finalcauses have a part to play.

According to Max Planck, the founder of quantumtheory:

The cause efficiens, which operates from the present into thefuture and makes future situations appear as determined byearlier ones, is joined by the cause finalis for which, inversely,the future—namely a definite goal—serves as the premise fromwhich there can be deduced the development of the processeswhich lead to this goal. (Cited in Yourgrau & Mandelstam, 1968,p. 165)

If efficient causes have proven to be inadequate to explainphysical phenomena at the most fundamental level, theymay be expected to be still less adequate in psychology.Therefore, an attempt at final-cause explanation in psy-chology is worthwhile. (See Silverstein, 1988, for a similarargument.)

Causation in Classical and ModernPsychologyJust as Renaissance physics replaced Aristotelian finalcauses with efficient causes, Descartes attempted to re-place final causes of human and nonhuman animal be-havior with efficient causes. According to Descartes, allnonhuman behavior is efficiently caused by externalstimuli. In Descartes's model, a stimulus acting througha sense organ is mechanically transmitted through thenerves to the brain, whereupon animal spirits (supposedto be a material cause of life) are released to flow backthrough the nerves to expand the muscles and causemovement. The difference between humans and otheranimals is that, in addition to this completely automaticmechanism that causes involuntary behavior, humanshave another way to direct animal spirits to variousnerves: Their will, which, acting through the pineal glandin the brain, can alter the direction of the flow of animalspirits. Whether a person's behavior is voluntary or in-voluntary, it is always, according to Descartes, efficientlycaused. Voluntary behavior is efficiently caused by thewill, whereas involuntary behavior is efficiently causeddirectly by external stimulation. Thus, Descartes broughtRenaissance psychology into line with Renaissancephysics.

Contrary to Aristotle's conception of the nonprivacyof mental states (see previous quote from Irwin, 1980),for Descartes the essential fact about mental states was

November 1992 • American Psychologist 1373

their privacy. "I think therefore I am" is a pronouncementabout what is "clear and distinct" in the privacy of themind. The existence of anything else—one's own body,its behavior, and the bodies and behavior of all otherthings—rests, for Descartes, on this internal and privateclarity.

Since Descartes, psychology has, in various ways,attempted to demystify the action of the will. One methodhas been to incorporate all behavior, voluntary as well asinvoluntary, into more or less complicated reflex systems.Sechenov's (1863/1965) Reflexes of the Brain and Pavlov's(1927) Conditioned Reflexes represent such systems. Ac-cording to Sechenov, "The initial [efficient] cause of allbehavior always lies not in thought, but in external sensorystimulation, without which no thought is possible"(p. 321).

The American behaviorists Watson (1913), Guthrie(1935), Hull (1952), and Tolman (1949) differed stronglyon how to describe the internal efficient causes of behavior,but their ultimate objective, no less than that of Sechenovand Pavlov, was to explain behavior in terms of its internalefficient causes, to get behind behavior itself to discoverits underlying mechanisms. Disputes among these be-haviorists and between behaviorists and cognitive psy-chologists, as well as among various schools of cognitivepsychology, have centered around the question, What arethe internal efficient causes of behavior? At one extreme,Sechenov, Pavlov, and their modern descendants (e.g.,Rescorla, 1988) attempted to trace the "reflexes of thebrain." The connectionists Thorndike (1911) and Hull(1952) spoke more abstractly in terms of internal stim-ulus-response (S-R) connections or associations perhapseventually reducible to physiological reflexes.

More molar behaviorists like Tolman (1949) aban-doned the possibility that mental constructs could be re-duced to reflexes but still retained the goal of describinginternal constructs (if not mechanisms). In Tolman'shands, S-R psychology became S-O-R psychology, wherethe O was conceived as mediating as an efficient causebetween environment and behavior. The cognitive psy-chologist John R. Anderson (1991, p. 513) stated, "I havealways felt that something was lost when the cognitiverevolution abandoned behaviorism." Nonetheless, hewent on to say that "in doing this, however, I do not wantto lose the cognitive insight that there is a mind betweenthe environment and behavior."

For some modern philosophers of psychology, a givenmental state may be internally represented by the actionof a given computer mechanism consisting of individualcomponents, none of which may actually represent themental state. Dennett (1978) called such mechanisms"intentional systems." As a very simple example of sucha system, Dennett (1978, pp. 71-89) cited Thorndike'slaw of effect. In the operation of the law of effect, rein-forcement strengthens a modifiable connection causingbehavior that appears purposive, despite the fact that theorganism's purpose itself has no coherent internal rep-resentation. Another example of an intentional systemwould be a chess-playing computer that, say, had a ten-

dency to bring out its queen too soon even though nosuch tendency was explicitly programmed in the machine.Mental states, according to this sort of cognitive psy-chology, may be emergent qualities of behavior. Researchon "neural networks" (Grossberg, 1982) are perhaps theclearest example of how complex cognitive and behavioralprocesses may emerge from the concatenation of muchsimpler efficient causes.

Another sort of modern cognitive philosophy (Fodor,1981) insists that all mental states are internally repre-sented as such and interact with each other in the mindto cause behavior. According to Fodor (1981, p. 5),"mental causes typically have their overt effects in virtueof their interactions with one another." Psychology wouldconsist of the analysis of such interactions. Philosophersof this latter cognitive school frequently (and not withouta degree of justification) accuse those of the former ofbeing behaviorists.

The issue within classical behaviorism as well as be-tween classical behaviorism and modern cognitive psy-chology (of either school) is not whether psychology con-sists of the analysis of an efficient-cause mediating processbetween environment and overt behavior; all follow Des-cartes on the necessity of an efficient-cause analysis. Theissue is whether mental terms are necessary in such ananalysis.

In general, aside from Skinner, behaviorists andcognitivists alike have viewed psychological theory asabout efficient causes mediating between environmentand behavior. To Skinner, we owe the renaissance of theAristotelian focus on the behavior of whole organisms intheir environments.

Causation in Skinnerian PsychologyIn the more than half century between his first publishedarticle and his death in 1990, Skinner's views naturallyvaried considerably. It is not our object here to trace thatdevelopment. One thread that runs throughout histhought, however, is his ultimate concern with behaviorof the organism as a whole. The concepts of the operantas a class of individual movements of whole organismshaving a common effect, the causal power attributed tocontingencies of reinforcement of operants, and the weightassigned to an organism's and a species' history of rein-forcement are strains of Aristotelian thought in Skinner'spsychology.

A contingency of reinforcement (a broader term forschedule of reinforcement), for instance, may be con-ceived as an Aristotelian final cause of the individual actsof which it is composed. Note that an individual reinforcerconsidered in isolation could not be a final cause; a re-inforcer strictly follows the act it causes. A reinforcementcontingency includes the act it causes; thus, a contingencymay be a final cause.

However, although Skinner was never deeply con-cerned with the logic or philosophy of the concept of cau-sation, there are signs that he was not comfortable withteleological explanation as such. Skinner (1938, p. 69)held that an individual behavior-reinforcer contiguity

1374 November 1992 • American Psychologist

served as the (efficient) cause of subsequent response-rateincreases. His concept of superstition rests on adventitiouscontiguities as (efficient) causes of future behavior. EvenSkinner's most molar construct, history of reinforcement,is construed as existing wholly in the past; Skinner sawa person's reinforcement history as the efficient cause ofhis or her subsequent behavior rather than as the finalcause of the behavior comprising the history.

As part of his discussion of nutrition and reproduc-tion, Aristotle (De Anima, Book II, chap. 4, 415b) dis-tinguished between the soul as an efficient cause and thesoul as a final cause. The soul may be the efficient causeof behavior only in the sense that parents, by their repro-ductive activity, pass their behavioral characteristics onto their offspring. That is, the reproductive (the most ba-sic) part of the parents' souls efficiently causes their off-springs' souls. However, within a person's lifetime, his orher soul is the final, not the efficient, cause of behavior.An individual's history of reinforcement (Skinner's con-cept closest to Aristotle's soul) thus would have beenviewed differently by Aristotle and Skinner. For Skinner,it was an efficient cause of behavior. Aristotle would haveviewed it not as an efficient cause, preceding any partic-ular act, but as a final cause embracing the act.

Skinner's conception of reinforcement has seemedto some modern commentators (Ringen, 1985; Staddon,1973) to involve efficient causation but with temporalgaps. An individual response-reinforcer contiguity mayimmediately increase response rate, but then what standsbetween one response and another? Certainly, a historyof reinforcement, considered as an efficient cause of tem-porally remote behavior, must act through some (pre-sumably internal) mechanism. In referring to this pos-sibility, Staddon (1973) stated,

I suggest that the apparent simplicity of the relation betweenoperant behavior and its consequences implies not an absenceof mechanism, not that operant behavior is "undifferentiatedmaterial," in Skinner's phrase, but rather the existence of moreand richer mechanisms than have hitherto been seriously con-templated. . . . Explanations in terms of "purpose" or "finalcause" are always incomplete, (p. 55)

This is true as far as it goes. However, Staddon did notadd that explanations in terms of efficient cause (mech-anisms in Staddon's terms) are also incomplete. Goodautomobile mechanics (if such exist) are not necessarilygood drivers (may not be able to predict and control themovements of the car as a whole).

The price Skinner paid for the rejection of teleologywas a simultaneous rejection of mental terms and accep-tance of private inner causes (Zuriff, 1979). Day (1969)referred to an exchange between Skinner and the philos-opher Michael Scriven: "Skinner is objecting . . . no t tothings that are private but to things that are mental" (p.38). Here is Skinner's (1953, p. 279) not atypical expla-nation of the causal status of an idea: "If the individualhimself reports 'I have had the idea for some time buthave only just recently acted upon it,' he is describing acovert response which preceded the overt." For Aristotle,

an idea would not be a covert response at all but a patternof wholly overt responses including the individual's verbalreport as one particular part of the pattern.

In relegating mental concepts such as ideas to theinterior of the organism (the same place where the or-ganism's physiology exists) Skinner is in line with moderncognitive psychology. Both Skinner, in the just cited pas-sage, and much of modern cognitive psychology placemental and physiological events together (cognitive neu-rophysiology) inside the organism. The difference is that,for Skinner, such internal events are unimportant simplybecause they are internal (they are not controlling vari-ables). Skinner stated (1953),

The objection to inner states is not that they do not exist, butthat they are not relevant in a functional analysis. We cannotaccount for the behavior of any system while staying whollyinside it; eventually we must turn to forces operating upon theorganism from without. Unless there is a weak spot in our causalchain so that the second link is not lawfully determined by thefirst, or the third by the second, then the first and third linksmust be lawfully related. If we must always go back beyond thesecond link for prediction and control, we may avoid manytiresome and exhausting digressions by examining the third linkas a function of the first. Valid information about the secondlink may throw light upon this relationship but can in no wayalter it. (p. 35)

For modern cognitive psychologists on the otherhand, internal events are critically important; their studyis the essence of psychology itself. Consider the followingfrom Dennett (1978) regarding the philosophers Ryle andWittgenstein, who, like Aristotle, identified mental termswith molar actions of whole organisms:

Ryle and Wittgenstein are the preeminent modern theorists ofthe personal [whole organism] level. In fact, in their differentways they invent the enterprise, by showing that there is workto be done, that there are questions that arise purely at thepersonal level, and that one misconceives the question if oneoffers sub-personal [i.e., cognitive] hypotheses or theories as an-swers. Typically readers who do not understand, or accept, thesedifficult claims see them as evading or missing the point, andcomplain that neither Ryle nor Wittgenstein has any positivepsychological theory to offer at all. That is true: the personallevel "theory" of persons is not a psychological theory, (p. 154)

Skinner (1938) claimed that there are only two basickinds of behavior: respondents, which are classes of be-havior (like a person's pupillary dilation or a dog's sali-vation) elicited by immediate antecedent stimulation, andoperants, which are classes of behavior correlated withimmediate environmental consequences. Examples ofoperants are rats' lever presses, pigeons' key pecks, andall animal behavior normally considered voluntary. As-pects of the environment crucial for respondent and op-erant dynamics are antecedent stimuli of respondents:reinforcers of operants, consequences that generally in-crease an operant's rate of emission (operant conditioningbeing the study of the relationship of operants and rein-forcers); and discriminative stimuli, in the presence ofwhich a given operant-reinforcer relation obtains. An ex-ample of a discriminative stimulus is the open-closed

November 1992 • American Psychologist 1375

sign on the door of a shop, signaling a given relation be-tween door pushing (the operant) and door opening (thereinforcer).

Nowhere here are mental terms. Sometimes Skinner(1953) offered "interpretations" in which the use of men-talistic vocabulary in everyday speech was explained interms of operants, reinforcers, and discriminative stimuli.Self-control, for instance, was held to be nothing but op-erant avoidance of certain discriminative stimuli ("getthee behind me Satan") like crossing the street to avoidthe enticing smell of a bakery.

Skinner's nonmentalistic terminology has servedvery well in the analysis of discrete operants like pigeons'key pecks, rats' lever presses, and human button pushes(Honig & Staddon, 1977). Furthermore, the patterns ofbehavior discovered in one situation with one species oftenappear in other situations with other species. These pat-terns change in systematic ways with motivational vari-ables like reinforcer deprivation and drug dosage. Skin-nerian techniques have been very successful in areas ofbehavior therapy ranging from treatment of severe psy-choses to weight control; their great advantage in theseapplications is their resolute focus on consequences andcontingencies of reinforcement. For example, manywomen (and some men) suffer from agoraphobia; theyare house bound, have panic attacks in public places, andconsequently refuse to leave home. In searching for causesand treatment of such behavior, the Skinnerian behaviortherapist considers not just its antecedents but also itsconsequences: avoidance of work, avoidance of sexualtemptation, attention from relatives and friends, and soon. Focusing on the actual consequences of dysfunctionalbehavior has led in many cases to the development ofsuccessful treatment by substitution of less dysfunctionalbehavior to achieve equivalent ends. Skinnerian tech-niques have been successfully applied also in businessmanagement and in areas normally considered cognitive,such as the teaching of reading and mathematics to chil-dren and college-level courses as diverse as anatomy andforeign languages.

However, despite this success, it has not been possibleeither in the operant laboratory or in the many areas ofapplication of Skinnerian behaviorism to divide all be-havior neatly into specific respondents and specific op-erants. A respondent must be correlated with an ante-cedent stimulus and an operant with a consequent rein-forcer. What, for instance, reinforces the act of refusingan offered cigarette by a smoker trying to quit? Havingto deal with and talk about such obviously importantacts, behavior therapists have taken two roads, neither ofwhich are satisfactory.

Some, like Homme (1965), developed an operantpsychology of the hidden organism, speaking of inner(covert) respondents, inner operants (coverants), and innerdiscriminative stimuli. According to these psychologists,people who refuse an offered cigarette can just reinforcethe act themselves (pat themselves on the back so tospeak). This conception has both logical and empiricalproblems. Logically, if people can reinforce their own

actions, why should they ever withhold reinforcement ofany action? What reinforces the giving and withholdingof self-reinforcement (Catania, 1975)? Empirically, thereis just no evidence that self-reinforcement works and someevidence that it does not work (Castro & Rachlin, 1980).

The other road taken by behavior therapists has ledto cognitive-behavioral therapy (Mahoney, 1974). Cog-nitive-behavioral therapists retain Skinnerian techniquesfor acts that are clearly reinforced. However, where en-vironmental reinforcers are not obvious or immediate,cognitive-behavioral therapy abandons behaviorism en-tirely and refers to mental states as inner causes. Thus,people who refuse the cigarette may do so because theybelieve it is better for their health and because they wantto be healthy. A therapist might then try to strengthenthose peoples' beliefs and desires by logical argument, byasking them to repeat a statement of their beliefs overand over, or by reinforcing the statement of the belief.Even this last procedure is cognitive, not behavioral, be-cause it rests on the assumption that the statement ismerely evidence of an internal state and that the rein-forcement acts not only on the external statement butalso on the inner belief. After all, it is the refusal of cig-arettes (what the belief is said to cause) rather than theverbal statement that the therapist is ultimately trying tostrengthen. In principle, there is nothing wrong with cog-nitive-behavioral therapy. If people do have beliefs as in-ner states and if beliefs can cause specific actions, thenchanging the belief will change the action.

From a viewpoint that sees mental terms as descrip-tions of final causes, however, the cognitive-behavioraltherapists are making what Ryle (1949) called a categorymistake. If a belief is nothing but a pattern of actions,then a statement of a belief is merely one of those actions.Altering the statement would affect the belief only to theextent that it affects one part of the pattern, not the centralsource of all parts. Beethoven's fifth symphony, for in-stance, is a pattern of notes with four very familiar notesat the beginning. When you hear those notes from anorchestra, you can be fairly certain that you are about tohear the rest. To alter those four notes is to seriously alterBeethoven's fifth symphony (perhaps to the extent thatyou would want to call it something else) but not nec-essarily to have any effect on the other notes.

Then again, there may indeed be a central state, moreor less coherently represented in the nervous system, moreor less innate, that controls all the behavior that an outsideobserver would call evidence of a person's belief. Alteringa belief (for example, substituting the score of Schubert'sninth for Beethoven's fifth) would then alter all of its be-havioral effects. Cognitive-behavioral therapists are thustrying to get at the central antecedents, the efficient cause,the core, the nub, the nut, the origin, the control roomof belief behavior, the very essence of a person's belief.The problem is that they have abandoned the aspect ofSkinner's program that made it so successful: its concen-tration on consequences rather than antecedents. A ther-apist who focuses on the central efficient causes (the how)of a person's belief tends to lose sight of the reinforcers

1376 November 1992 • American Psychologist

(what a person is getting by behaving in this way), whateffects the person's belief has on relations with family,friends, with the environment in general, the why of thebelief.

The failure of behavioral psychology to deal withmentalistic concepts such as beliefs (which philosopherscall propositional attitudes or intentional states) and pains(which they call raw feels) has, by and large, left treatmentof strictly mental dysfunction to cognitive and physio-logical psychology. A comprehensive behavioral analysisof such concepts does not currently exist, but such anal-yses are possible and may be highly useful.

For instance, according to Skinner (1945), a pain isan internal private stimulus; purely behavioral analysisand treatment can never get at pain itself, only at painbehavior. Thus, Skinner essentially left the field to criticsof behaviorism like Putnam (1980), who argue that be-haviorism is inadequate to deal with pain itself. Nev-ertheless, from a teleological viewpoint, pain and painbehavior may be construed as one thing. Rachlin (1985)argued that pain treatment based on the equality of painand pain behavior (widely construed) has been at least aseffective (and if generally accepted might be more effec-tive) than treatment based on pain as an internal state.

Teleological BehaviorismGuthrie (1935) distinguished between acts and move-ments. A movement, according to Guthrie, is a particularset of muscular contractions resulting in a particular locusof bodily displacement. (Guthrie's definition of movementis thus much narrower than Aristotle's.) An act, accordingto Guthrie, is a coordinated pattern of movements leadingto some definable result. Waving good-bye, for example,would be an act. Ten different instances of a person'swaving good-bye might involve 10 wholly different setsof movements but constitute a single act repeated 10times. Guthrie claimed that the apparent learning of actswas essentially an accident of the learning of particularmovements.

Skinner's (1938) concept of operant conditioningreversed Guthrie's claim. An operant (an act as Guthriedefined it) may be directly reinforced, according to Skin-ner, without regard to particular movements.

Teleological behaviorism expands Skinner's originalconcept of reinforcement from a single event dependenton a single operant (for instance, a single food-pellet de-livery immediately after a single lever press) to a patternof environmental events perhaps only vaguely contingenton an overlapping pattern of operants. In Herrnstein's(1969) conception of avoidance, for instance, the effectivecontingency is a negative correlation imposed betweenthe rate of aversive events and the rate of operant emission,not any individual consequence (hypothetical or real)contingent on any particular operant.

Even so abstract a concept as belief is such a patternof overt observable acts and consequences. Take away theobservable acts, and the belief goes away as well; it is justas if you were to take away the lights and shadows on themovie screen. Without lights and shadows, the action, the

characters, their motives, and their beliefs all go away aswell.

Teleological Behaviorism and Modern Philosophyof Psychology

At this point, it may fairly be asked why, if Aristotle wasreally a behaviorist, are modern philosophers, particularlythose who consider themselves Aristotelians, so dead setagainst behaviorism. The answer is not that these phi-losophers misunderstand Aristotle; nonetheless, they domisunderstand behaviorism, especially its molar char-acter, and they completely ignore teleological behavior-ism. However, one point of agreement between modernAristotelian philosophy and behaviorism is a mutual re-jection of introspection and the private nature of mentalevents. Donagan (1987) stated,

A false and deeply confused doctrine that was philosophicallyfashionable is still encountered: namely, that taking propositionalattitudes [like beliefs] and persisting in them are items in one'sflow of private consciousness, which are named by private os-tensive definition, and which have complex causal relations withone's bodily states, (p. 54)

Donagan seemed to be arguing against introspection asa method in psychology. Aristotelian philosophers in gen-eral do indeed deny the privileged status of introspection.However, Donagan went on to say that "it does not followthat description of actions in terms of their doers' prop-ositional attitudes can be analyzed without residue interms of patterns of their surface behavior." Let us discussthat residue. Donagan did not say that if you removedall of the surface behavior you could still have the residueleft. Rather, when we look at the surface behavior, weperceive it and describe it in such a way that our percep-tions and descriptions cannot be analyzed into strictlybehavioral terms, just as you could not analyze the per-sonality of the characters in a movie into lights and shad-ows. Then, one must ask, how do we perceive beliefs andother propositional attitudes? The answer, according toDonagan and most contemporary philosophers, Aristo-telians, and others, is that we perceive them as arisingfrom inside of the person, ourselves or someone else, whohas them.

Dretske (1988, p. 3) stated that "behavior is endog-enously produced movement, movement that has itscausal origin within the system whose parts are moving."Thus, according to Dretske, acts to which no internalcause is attributable are not even behavior, let alone vol-untary behavior. For Dretske, overt behavior as we per-ceive it comes with a sort of (efficient) causal tail. A rat'slever press, if it is perceived as behavior, is perceived asbeing caused by something inside the rat. The lever pressitself plus the cause make up the rat's act. Correspond-ingly, when we perceive a person's actions as arising fromthat person's belief, we also perceive behavior with a tailextending inside the person. However, this time the tailwags the person, so to speak. The part of our perceptionsthat reaches inside the person contains not just a singleinternal efficient cause as in the case of a rat's lever press

November 1992 • American Psychologist 1377

but an entire representational system. The person's cog-nitive state consists of the overt act and the internal rep-resentational system together. (Dretske's analysis of be-havior is thus essentially cognitive.) The representationalsystem (the reason for the act) like the causal tail on therat's lever press is the "residue" inevitably left over fromany conceivable analysis of surface behavior alone. Thisis how modern philosophy justifies modern cognitivepsychology: internal states as either efficient causes of orless determinate "reasons for" external behavior.

However, the present discussion of Aristotle's phi-losophy suggests that the residue after the most pene-trating finite behavioral analysis is just more behavior,like the residue of uncertainty about the probability of acoin being unbiased after x number of tosses. There isinevitably a residue in this case, but it exists because anideal analysis would take an infinite amount of time, notbecause the residue exists in another place.

There is no question that sometimes, in our society,in our linguistic environment, we attach inner causes toexternal behavior. Such attachment may harmlessly becalled a branch of "folk psychology," which then may besaid to have developed into modern cognitive science(Stich, 1983). However, as Kantor (1963) argued, thereis another branch of folk psychology, a sort of folk be-haviorism, that is satisfied only when an act is explainedin terms of its consequences. Our folk psychology tells usto be satisfied with an explanation of a murder, for in-stance, in terms of consequences such as money or sexrather than in terms of inner compulsions. When bothsorts of explanation are available, we tend to choose theformer. It is only when there is no clear and direct envi-ronmental consequence that we are driven to the latter.However, because in such cases we try to invent a plausibleinternal representational system, as cognitive psychologyis set up to do, why is it not just as valid to try to discovera plausible environmental consequence even if unclearand indirect? This, it seems, is exactly what Aristotle didin corresponding circumstances when he looked for finalcauses as behavioral explanations.

Kantor (1963) claimed that a behavioristic folk psy-chology is the true one, the only one on which a scientificpsychology could be built. Why then, it may fairly beasked, would we ever, in our everyday mutual interactionsor our technical discourse, talk as though mental eventswere internal efficient causes of our behavior? Taylor(1964) blamed it on "atomism," which he defined (p. 11)as "the notion . . . that the ultimate evidence for anylaws we frame about the world is in the form of discreteunits of information, each of which could be as it is evenif all others were different." On this basis, Taylor criticizedboth modern cognitive psychology ("centrism") andclassical behavioral psychology ("peripheralism"). Taylorstated that "teleological explanation is, as has often beenremarked, connected with some form of holism, or anti-atomistic doctrine" (p. 12).

However, the sort of holism Taylor (1964) advocateddiffers from the behavioristic holism advocated here.Taylor's (and Dretske's, 1988) holistic approach embraces

an organism's current internal as well as its current overtbehavior (peripheralism plus centrism). The teleologicalbehaviorist's holistic approach, alternatively, embraces anorganisms' past and future overt behavior as well as itscurrent overt behavior. The former is a holism of space,the latter, a holism of time.

Utility Functions as Final Causes

The utility functions discoverable by behavioristic meth-ods (Rachlin, Battalio, Kagel, & Green, 1981) are ex-amples of final causes. Utility functions are essentiallymathematical summaries of observed behavior. Theystand to a particular choice under particular constraintsas the equation of a circle stands to particular circles.

The empirical procedure for investigation of utilityfunctions is as follows:

1. Observe behavioral patterns under a selected set ofenvironmental constraints.

2. Infer a utility function from the observed patterns as-suming that utility is maximized within the set of con-straints observed or imposed in Step 1.

3. Given the utility function and the assumption of max-imization, predict behavior under a new (previouslyunobserved) set of constraints.

4. Revise the utility function on the basis of deviationbetween predicted and actual behavior.

Behavior under a set of imposed constraints thusreveals the subject's preferences (Samuelson, 1973). Thebetter the utility functions are known, the better behavioris predicted and controlled. Utility functions convergenaturally onto parts of the intentional idiom: the conceptsdesire, appetite, will, self-control, and value. Whetherutility functions will ever provide a precise analysis ofother intentional idioms such as belief, knowledge, andintelligence, terms that seem to refer to verbal as well asnonverbal behavior, is still (as Lacey & Rachlin, 1978,argue) an open question. If a behavioral analysis ulti-mately fails, the failure will be due to the complexity ofthe task. It will be like the failure to precisely predict andcontrol the weather rather than any intrinsic inaccessi-bility or opacity of its subject.

The present analysis, therefore, agrees with modernAristotelian philosophers that mental terms (propositionalattitudes, dispositional states) form at least part of ourfundamental "data language." The issue is what to dowith those data. Should they be analyzed into currentovert behavior plus internal causes and reasons as somepart of our folk psychology seems to dictate, or shouldthey be analyzed wholly on the level of overt behaviorand its consequences over time as some other part of ourfolk psychology seems to dictate? Aristotle might haveargued that both avenues need to be pursued: We shouldknow all of the causes of our object of study; success ineither, far from blocking the other, will illuminate its path.

Teleological analyses, because of their molar nature,are necessarily less precise than efficient-cause analyses.A given act may be truly understood only at some time,perhaps a considerable time, after it occurs because the

1378 November 1992 • American Psychologist

context (the final cause) of an act extends into the futureas well as into the past. An individual lever press is un-caused in exactly the same sense that an individual eventhas no probability. A probability (conceived as a relativefrequency) may be known as accurately as you want itto be known if you are willing to wait. The same goes forthe causes of an individual lever press, and, arguably, thesame goes for a person's belief (including one's own be-lief). The difference between cognitive (efficient-cause)psychology and teleological (final-cause) behaviorism isperhaps most clearly illustrated in terms of their separateconceptions of probability.

ProbabilityFigure 1, complicated as it is, is a simplified picture ofhow cognitive and behavioral models deal with identicalconcepts. (For the sake of brevity, we drop the label te-leological and just say behaviorism, behavioral, and soon.) The double vertical line represents the boundary be-tween a person and the world. Note that five horizontallines—three solid and two dotted—cross the boundary.The solid lines are the three critical behavioral variables.The top solid line heading into the person represents dataor information of some kind, including such stimuli astrain whistles, red and green traffic lights, or the instruc-tions a psychologist might give to an experimental subject.These informative stimuli may function in two ways. First,they may signal significant outcomes (like trains signaledby whistles or food signaled by tones in Pavlov's experi-ments). In such cases, these signaling events are calledconditional stimuli (CSs). The relation of CSs to signifi-cant environmental events {unconditional stimuli, or USs)is independent of an animal's behavior. In Skinnerian

Figure 1A Model of Probabilistic Choice

THE WORLD

Information

THE PERSON

respondentcontingency

(CSor 8 ' )

varbalreport

nonverbalovartbehavior

operantcontingency

outcome

repreaentatlonof Information

declelonmechanlam

choicemechanlam

( US or relnforcer )

Note. A behavioral model (left) and a cognitive model (to the right) of proba-bilistic choice. (CS = conditional stimulus; S° = discriminative stimulus; US =unconditional stimulus.)

terms, this relation is a respondent contingency. The wholeprocess, including its effects on behavior, is called re-spondent conditioning.

Second, the information could signal not a signifi-cant event as such but the relation between behavior andoutcome, a relation called, again using Skinner's terms,an operant contingency. Informative stimuli that signalthe presence of an operant contingency are called dis-criminative stimuli. The red and green traffic lights thatsignal the relation between crossing the street and thelikelihood of having an accident (or getting a ticket) arediscriminative stimuli. This whole process including itseffects on behavior is called operant conditioning. (Thereis much debate among behaviorists, which we shall com-pletely ignore here, as to whether respondent conditioningis really a form of operant conditioning or vice versa.)

The two dashed arrows crossing the world-personboundary represent verbal reports (of representations anddecisions in the illustration). Verbal reports may be com-plex and temporally extended (as are many nonverbalactions such as playing a saxophone), in which case (likeother complex overt actions) they have a complex syn-tactical structure. Alternatively, they may be simple anddiscrete (grunting or saying ouch or yes or no), in whichcase they may have a simple structure. In either case,verbal reports, like other overt behavior, may be elicitedby an unconditioned stimulus or entail significant con-sequences, thus taking part, like other overt behavior, inrespondent or operant conditioning.

The behaviorist studies the relation between theseboundary-crossing events from the left in Figure 1. Be-havioral inferences and models are inferences and modelsabout respondent and operant contingencies that may notbe present at the moment but may serve as the contextfor current actions (Staddon, 1973). A grocer, Mr. Jones,for instance, may give a poverty-stricken customer a freeloaf of bread because he is at heart a sympathetic andgenerous person or as part of a promotional campaign.The grocer's differing motives form differing potentialcontexts for (and causes of) his act. From a behavioralpoint of view, these contexts differ by virtue of differingrespondent and operant contingencies operating over pe-riods of time much wider than the present moment. Thequestion of which motive the grocer really has can beanswered, from a behavioral viewpoint, only by referenceto such overt contextual events: the existence or nonex-istence of a promotional campaign and the grocer's be-havior when no such promotional campaign is beingwaged.

One important feature of the behavioral viewpointbears emphasis: The question may be settled by referenceto future as well as to past events, because the temporalcontext of a brief event extends into the future as well asthe past. To decide what the grocer's motives really were,you could perform an experiment post hoc: for example,sending poor customers into the store on different occa-sions. (In technical terms, you would be trying to deter-mine a utility function for the grocer to predict behaviorin other circumstances.) If the grocer should die imme-

November 1992 • American Psychologist 1379

diately after his generous act (and historical records arelost or nonexistent), there is no way, even in principle, todetermine its behavioral context (the grocer's motives).Supposing contrary to fact that Beethoven's fifth sym-phony and Schubert's ninth symphony had three identicalcontiguous notes, it would be like trying to determinewhich symphony you were listening to from those threenotes alone.

Again, from the behavioral viewpoint, a mental actsuch as a motive has meaning only over an extended spanof time. To sample only a piece of that time is to makea guess as to the motive. As wider and wider samples aretaken the grocer's motive becomes more and more know-able, but it will never be 100% knowable; in principle,its context is infinite. This is true even with respect tothe grocer's knowledge of himself because, from the be-havioral viewpoint (from the left in the diagram), thegrocer is in a privileged position only by virtue of thequantity rather than the quality of his information: Hehas a larger sample of his own behavior than anyone elsebut by no means a better sample. Indeed his informationas to his own motives may be much worse than that ofan outsider because he can observe his own behavior (asa whole person) only by reflection (through its variousoutcomes), whereas outsiders may view his behavior di-rectly. Thus, the behaviorist turns on its head the commonmentalist notion that people have privileged access to theirown mental states. In this respect, teleological behavior-ism is like Freudian psychology (although, of course, dif-ferent in many other respects).

Cognitive psychology takes the same sort of data asdoes behavioristic psychology—the five arrows crossingthe world-person boundary—and infers a context of thoseevents not in past and future contingencies but in thepresent, inside the behaving organism. Cognitive psy-chology looks at the diagram of Figure 1 from the right.The cognitive context of the grocer's generous act, there-fore, consists of events inside the grocer (or states of in-ternal mechanisms) contemporaneous with (or imme-diately preceding) that act: its efficient causes.

The right section of Figure 1 illustrates in highlyschematic terms how a cognitive model (Kahneman &Tversky's, 1979, prospect theory) handles probabilisticinformation. In a typical experiment, each of a large groupof subjects may be asked a series of hypothetical questionsof the following form: "Which would you prefer: a 50%chance of winning $10,000 or $5,000 for sure?" (By far,a greater number of subjects say they would prefer thesure thing.) The first stage of the model consists of "ed-iting" the problem and transforming its four elements(probabilities of 1.0 and .5, amounts of $5,000 and$10,000) into internal representations. The internal rep-resentations may be reported directly (upper dashed ar-row), but their creation is only the first step of the decisionprocess. They are then combined in some way specifiedby the theory to arrive at a decision, and the decision isreported (lower dashed arrow). Investigators have foundverbal reports of representations of probability to be gen-erally unreliable predictors of decisions (the effective in-

ternal representations of probability are termed decisionweights in prospect theory). However, verbal reports ofdecisions are quite reliable predictors of actual choiceswhen (occasionally) subjects are asked to choose amongactual rather than hypothetical probabilistic outcomes.(Of course, money amounts are much smaller in the ac-tual choice experiments.) The ultimate object of a cog-nitive theory is to predict decisions and choices in a widevariety of real-life and laboratory situations. Arguably,prospect theory does a good job.

The important point from our perspective is thatprobability is conceived fundamentally as an internalstate. The experimenter who says "The probability of$ 10,000 is .5" is, according to prospect theory (and mostcognitive theories), merely activating an internal repre-sentation. The true probability is that representation. Thisreflects a subjectivist view of probability: as a degree ofconfidence or degree of certainty about coming events.That confidence or certainty then may determine sub-sequent choices (Lucas, 1970). When a meteorologist says"The chance of rain is 90%," the confidence in rain thusaroused may cause the listener to take his or her umbrellato work. The meteorologist's words are efficient causesof the confidence, which is (part of) the efficient causeof the decision to take an umbrella, which in turn is (partof) the efficient cause of actually taking the umbrella.The critical mechanism may involve various feedbackloops (as Figure 1 shows) and may well be much morecomplex than this (as even a knee-jerk reflex is muchmore complicated than a simple series of S-R connec-tions), but it is nonetheless a mechanism, one that involvesa series of efficiently caused acts.

For a behaviorist, the experimenter's assertion that"the probability of $10,000 is .5" does not elicit an in-ternal representation. It already is a representation of aclass of wholly external probabilistic events. The functionserved by the verbal statement is to place the presentnarrow situation (the psychology experiment) in the con-text of those events. In other words, the probability state-ment is a discriminative stimulus for a particular historyof reinforcement. Thus, to say to a person that the prob-ability of x is .5 is to say something like "Behave in thisexperiment as if a coin was being flipped, and you wererewarded with x if the coin came up heads." Unless theperson has had that sort of reinforcement history, theexperimenter's statement is meaningless. The true prob-abilities, from a behavioral viewpoint, are the relativefrequencies of events themselves. This is an objectivistview of probability. The meteorologist's statement "Theprobability of rain is 90%" is itself, for the behaviorist, adiscriminative stimulus for taking an umbrella to workand is established in the same way that any discriminativestimulus is established: by reliably having signaled a givenset of operant contingencies. (On previous occasions whenthis sort of weather was predicted, taking an umbrella towork was reinforced.)

Thus, for the behaviorist, the essential meaning ofa probability statement is what it represents externally(its discriminative function), whereas for the cognitivist,

1380 November 1992 • American Psychologist

the essential meaning of a probability statement is howit is represented internally. This is a difference betweenbehavioral and cognitive semantics. Note, however, thatneither the behavioral nor the cognitive approach dealswith or accounts for syntactic structures. It is no moreclear why several syntactically equivalent statementsshould arouse the same internal representation than it iswhy several syntactically equivalent statements shouldserve as discriminative stimuli for the same overt choicebehavior.

One reason for discussing behavioral and cognitiveapproaches to probability here is that experimental workwith both approaches seems to have converged on cor-responding models (Rachlin, 1989; Rachlin, Logue, Gib-bon, & Frankel, 1986). We have no space to trace out allof the various correspondences, but to obtain their flavorthe reader might imagine how one would present a persondirectly with a probabilistic event. It is impossible topresent a probability in an instant. A coin has to be flippedmany times for the probability of .5 of heads to emergeas a property of the coin; this takes time. Thus, a prob-ability itself must be presented as a relative frequency ofactual events. As it happens, the choice behavior of hu-mans and nonhumans in the face of various relative fre-quencies of actual events corresponds to human verbalreports of decisions in the face of verbally presentedprobabilities; the form of the functions relating proba-bility to behavior is the same for humans and nonhumans,accounting for the same sorts of apparent irrationalities,whereas the parameters of those functions differ widelybetween species.

Because the air is currently so full of antibehavioristpolemics from psychologists of mentalistic, cognitive, andphysiological orientations as well as from philosophers ofall orientations, it may seem that, in defending the be-havioristic approach, I am attacking all of the others. Onthe contrary, progress in answering either question, whyor how, awaits progress in answering the other.'

1 For a detailed exposition of current work in the areas discussedin this article and its relation to language, consciousness, and other psy-chological issues, see Rachlin (1989, 1991, in press).

REFERENCES

Ackrill, J. L. (1980). Aristotle on eudaimonia. In A. Q Rorty (Ed.),Essays on Aristotle's ethics (pp. 15-34). Berkeley, CA: University ofCalifornia Press.

Anderson, J. R. (1991). More on rational analysis. Behavioral and BrainSciences, 14, 508-513.

Aristotle (1984). The complete works of Aristotle (J. Barnes, Ed.).Princeton, NJ: Princeton University Press. (Original work publishedca. 350 B.C.)

Castro, L., & Rachlin, H. (1980). Self-reward, self-monitoring, and self-punishment as feedback in weight control. Behavior Therapy, 11, 38-48.

Catania, A. C. (1975). The myth of self-reinforcement. Behaviorism, 3,192-199.

Davidson, D. (1980). Essays on actions and events. Oxford, England:Clarendon Press.

Day, W. (1969). Radical behaviorism in reconciliation with phenome-nology. Journal of the Experimental Analysis of Behavior, 12, 315-328.

Dennett, D. (1978). Brainstorms: Philosophical essays on mind andpsychology. Montgomery, VT: Bradford Books.

Donagan, A. (1987). Choice: The essential element in human action.London: Routledge & Regan Paul.

Dretske, F. (1988). Explaining behavior: Reasons in a world of causes.Cambridge, MA: MIT Press.

Fodor, J. A. (1981). Representations: Philosophical essays on the foun-dations of cognitive science. Cambridge, MA: MIT Press.

Grossberg, S. (1982). Studies of mind and brain: Neural principles oflearning, perception, development, cognition, and motor control. Am-sterdam: Reidel Press.

Guthrie, E. R. (1935). The psychology of learning. New York: Harper.Herrnstein, R. J. (1969). Method and theory in the study of avoidance.

Psychological Review, 76, 49-69.Homme, L. E. (1965). Perspectives in psychology XXIV. Control of

coverants, the operants of the mind. Psychological Record, 15, 501-511.

Honig, W. K., & Staddon, J. R. (Eds.). (1977). Handbook of operantbehavior. New York: Prentice-Hall.

Hull, C. L. (1952). A behavior system: An introduction to behavior therapyconcerning the individual organism. New Haven, CT: Yale UniversityPress.

Irwin, T. N. (1980). The metaphysical and psychological basis of Aris-totle's ethics. In A. O. Rorty (Ed.), Essays on Aristotle's ethics (pp.35-53). Berkeley: University of California Press.

Kahneman, D., & Tversky, A. (1979). Prospect theory: An analysis ofdecisions under risk. Economelrica, 47, 263-291.

Kantor, J. R. (1963). The scientific evolution of psychology, volume 1.Chicago: Principia Press.

Lacey, H. M., & Rachlin, H. (1978). Behavior, cognition and theoriesof choice. Behaviorism, 6, 177-202.

Lucas, J. R. (1970). The concept of probability. Oxford, England: Cla-rendon Press.

Mahoney, M. (1974). Cognitive behavior modification. Cambridge, MA:Ballinger.

Pavlov, I. P. (1927). Conditioned reflexes (G. V. Anrep, Trans.). London:Oxford University Press.

Putnam, H. (1980). Brains and behavior. In N. Block (Ed.), Readingsin philosophy of psychology (Vol. 1, pp. 24-36). Cambridge, MA:Harvard University Press.

Rachlin, H. (1985). Pain and behavior. The Behavioral and Brain Sci-ences, 8, 43-83.

Rachlin, H. (1989). Judgment, decision and choice. New York: Freeman.Rachlin, H. (1991). Introduction to modern behaviorism (3rd ed.). New

York: Freeman.Rachlin, H. (in press). Behavior and mind: Two psychologies. New York:

Oxford University Press.Rachlin, H., Battalio, R., Kagel, J., & Green, L. (1981). Maximization

theory in behavioral psychology. The Behavioral and Brain Sciences,4, 371-388.

Rachlin, H., Logue, A. W., Gibbon, J., & Frankel, M. (1986). Cognitionand behavior in studies of choice. Psychological Review, 93, 33-45.

Randall, J. H., Jr. (1960). Aristotle. New York: Columbia UniversityPress.

Rescorla, R. A. (1988). Pavlovian conditioning: It's not what you thinkit is. American Psychologist, 43, 151-160.

Ringen, J. (1985). Operant conditioning and a paradox of teleology.Philosophy of Science, 52, 565-577.

Rorty, R. (1982). The consequences of pragmatism. Minneapolis: Uni-versity of Minnesota Press.

Ryle, G. (1949). The concept of mind. London: Hutchinson House.Samuelson, P. A. (1973). Economics: An introductory analysis (9th ed.).

New York: McGraw-Hill.Sechenov, I. M. (1965). Reflexes of the brain. In R. J. Herrnstein &

E. G. Boring (Eds.), A source book in the history of psychology (pp.308-320). Cambridge, MA: Harvard University Press. (Original workpublished 1863)

Silverstein, A. (1988). An Aristotelian resolution of the ideographic versusnomothetic tension. American Psychologist, 43, 425-430.

November 1992 • American Psychologist 1381

Skinner, B. F. (1938). The behavior of organisms. New York: Appleton-Century.

Skinner, B. F. (1945). The operational analysis of psychological terms.Psychological Review, 52, 270-277, 291-294.

Skinner, B. F. (195 3). Science and human behavior. New York: Macmillan.Staddon, J. E. R. (1973). On the notion of cause with applications to

behaviorism. Behaviorism, I, 25-64.Stich, S. (1983). From folk psychology to cognitive science. Cambridge,

MA: MIT Press.Taylor, C. (1964). The explanation of behavior. London: Routledge &

Kegen Paul.

Thorndike, E. L. (1911). Animal intelligence. New York: Macmillan.Tolman, E. C. (1949). There is more than one kind of learning. Psycho-

logical Review, 56, 144-155.Watson, J. B. (1913). Psychology as the behaviorist views it. Psychological

Review, 20, 158-177.Wright, L. (1976). Teleological explanations. Berkeley: University of

California Press.Yourgrau, W., & Mandelstam, S. (1968). Variational principles in dy-

namics and quantum theory. New York: Dover.Zuriff, G. E. (1979). Ten inner causes. Behaviorism, 7, 1-8.

1382 November 1992 • American Psychologist