Behavior, Purpose and Teleology

Arturo Rosenblueth, Norbert Wiener and Julian Bigelow

in: Philosophy of Science, 10(1943), S. 18–24.

This essay has two goals. The first is to de-fine the behavioristic study of natural eventsand to classify behavior. The second is tostress the importance of the concept of pur-pose.

Given any object, relatively abstracted fromits surroundings for study, the behavioristicapproach consists in the examination of theoutput of the object and of the relations of thisoutput to the input. By output is meant anychange produced in the surroundings by theobject. By input, conversely, is meant anyevent external to the object that modifies thisobject in any manner.

The above statement of what is meant bythe behavioristic method of study omits thespecific structure and the instrinsic organi-zation of the object. This omission is fun-damental because on it is based the distinc-tion between the behavioristic and the alter-native functional method of study. In a func-tional analysis, as opposed to a behavioristicapproach, the main goal is the intrinsic orga-nization of the entity studied, its structure andits properties; the relations between the objectand the surroundings are relatively incidental.

From this definition of the behavioristicmethod a broad definition of behavior ensues.By behavior is meant any change of an en-tity with respect to its surroundings. Thischange may be largely an output from the ob-ject, the input being then minimal, remote orirrelevant; or else the change may be immedi-ately traceable to a certain input. Accordingly,any modification of an object, detectable exter-nally, may be denoted as behavior. The termwould be, therefore, too extensive for useful-ness were it not that it may be restricted byapposite adjectives — i.e., that behavior maybe classified.

The consideration of the changes of energyinvolved in behavior affords a basis for classi-fication. Active behavior is that in which the

object is the source of the output energy in-volved in a given specific reaction. The objectmay store energy supplied by a remote or rela-tively immediate input, but the input does notenergize the output directly. In passive behav-ior, on the contrary, the object is not a sourceof energy; all the energy in the output can betraced to the immediate input (e.g., the throw-ing of an object), or else the object may controlenergy which remains external to it through-out the reaction (e.g., the soaring flight of abird).

Active behavior may be subdivided into twoclasses: purposeless (or random) and purpose-ful. The term purposeful is meant to denotethat the act or behavior may be interpretedas directed to the attainment of a goal — i.e.,to a final condition in which the behaving ob-ject reaches a definite correlation in time or inspace with respect to another object or event.Purposeless behavior then is that which is notinterpreted as directed to a goal.

The vagueness of the words may be inter-

preted as used above might be considered sogreat that the distinction would be useless. Yetthe recognition that behavior may sometimesbe purposeful is unavoidable and useful, asfollows. — The basis of the concept of purposeis the awareness of voluntary activity. Now, thepurpose of voluntary acts is not a matter of ar-bitrary interpretation but a physiological fact.When we perform a voluntary action what weselect voluntarily is a specific purpose, not aspecific movement. Thus, if we decide to takea glass containing water and carry it to ourmouth we do not command certain muscles tocontract to a certain degree and in a certainsequence; we merely trip the purpose and thereaction follows automatically. Indeed, exper-imental physiology has so far been largely in-capable of explaining the mechanism of volun-tary activity. We submit that this failure is dueto the fact that when an experimenter stimu-

ROSENBLUETH, WIENER AND BIGELOW 1 Behavior, Purpose and Teleology

lates the motor regions of the cerebral cortexhe does not duplicate a voluntary reaction; hetrips efferent, output pathways, but does nottrip a purpose, as is done voluntarily.

The view has often been expressed that allmachines are purposeful. This view is un-tenable. First may be mentioned mechanicaldevices such as a roulette, designed preciselyfor purposelessness. Then may be considereddevices such as a clock, designed, it is true,with a purpose, but having a performancewhich, although orderly, is not purposeful —i.e., there is no specific final condition towardwhich the movement of the clock strives. Sim-ilarly, although a gun may be used for a def-inite purpose, the attainment of a goal is notintrinsic to the performance of the gun; ran-dom shooting can be made, deliberately pur-poseless.

Some machines, on the other hand, are in-trinsically purposeful. A torpedo with a target-seeking mechanism is an example. The termservo-mechanisms has been coined preciselyto designate machines with intrinsic purpose-ful behavior.

It is apparent from these considerationsthat although the definition of purposeful be-havior is relatively vague, and hence oper-ationally largely meaningless, the concept ofpurpose is useful and should, therefore, be re-tained.

Purposeful active behavior may be subdi-vided into two classes: feed-back (or teleologi-

cal) and non-feed-back (or non-teleological) Theexpression feed-back is used by engineers intwo different senses. In a broad sense it maydenote that some of the output energy of anapparatus or machine is returned as input;an example is an electrical amplifier with feed-back. The feed-back is in these cases positive— the fraction of the output which reentersthe object has the same sign as the originalinput signal. Positive feed-back adds to the in-put signals, it does not correct them. The termfeed-back is also employed in a more restrictedsense to signify that the behavior of an objectis controlled by the margin of error at whichthe object stands at a given time with refer-ence to a relatively specific goal. The feed-backis then negative, that is, the signals from thegoal are used to restrict outputs which wouldotherwise go beyond the goal. It is this secondmeaning of the term feed-back that is usedhere.

All purposeful behavior may be considered

to require negative feed-back. If a goal is to beattained, some signals from the goal are nec-essary at some time to direct the behavior. Bynon-feed-back behavior is meant that in whichthere are no signals from the goal which mod-ify the activity of the object in the course ofthe behavior. Thus, a machine may be setto impinge upon a luminous object althoughthe machine may be insensitive to light. Sim-ilarly, a snake may strike at a frog, or a frogat a fly, with no visual or other report fromthe prey after the movement has started. In-deed, the movement is in these cases so fastthat it is not likely that nerve impulses wouldhave time to arise at the retina, travel to thecentral nervous system and set up further im-pulses which would reach the muscles in timeto modify the movement effectively.

As opposed to the examples considered, thebehavior of some machines and some reac-tions of living organisms involve a continu-ous feed-back from the goal that modifies andguides the behaving object. This type of be-havior is more effective than that mentionedabove, particularly when the goal is not sta-tionary. But continuous feedback control maylead to very clumsy behavior if the feed-backis inadequately damped and becomes there-fore positive instead of negative for certain fre-quencies of oscillation. Suppose, for example,that a machine is designed with the purpose ofimpinging upon a moving luminous goal; thepath followed by the machine is controlled bythe direction and intensity of the light fromthe goal. Suppose further that the machineovershoots seriously when it follows a move-ment of the goal in a certain direction; an evenstronger stimulus will then be delivered whichwill turn the machine in the opposite direction.If that movement again overshoots a series ofincreasingly larger oscillations will ensue andthe machine will miss the goal.

This picture of the consequences of un-damped feed-back is strikingly similar to thatseen during the performance of a voluntary actby a cerebellar patient. At rest the subject ex-hibits no obvious motor disturbance. If he isasked to carry a glass of water from a tableto his mouth, however, the hand carrying theglass will execute a series of oscillatory mo-tions of increasing amplitude as the glass ap-proaches his mouth, so that the water will spilland the purpose will not be fulfilled. This testis typical of the disorderly motor performanceof patients with cerebellar disease. The anal-

ROSENBLUETH, WIENER AND BIGELOW 2 Behavior, Purpose and Teleology

ogy with the behavior of a machine with un-damped feed-back is so vivid that we ventureto suggest that the main function of the cere-bellum is the control of the feed-back nervousmechanisms involved in purposeful motor ac-tivity.

Feed-back purposeful behavior may againbe subdivided. It may be extrapolative (pre-dictive), or it may be non-extrapolative (non-predictive). The reactions of unicellular organ-isms known as tropisms are examples of non-predictive performances. The amoeba merelyfollows the source to which it reacts; there isno evidence that it extrapolates the path of amoving source. Predictive animal behavior, onthe other hand, is a commonplace. A cat start-ing to pursue a running mouse does not rundirectly toward the region where the mouse isat any given time, but moves toward an extrap-olated future position. Examples of both pre-dictive and non-predictive servo-mechanismsmay also be found readily.

Predictive behavior may be subdivided intodifferent orders. The cat chasing the mouseis an instance of first-order prediction; thecat merely predicts the path of the mouse.Throwing a stone at a moving target requires asecond-order prediction; the paths of the tar-get and of the stone should be foreseen. Exam-ples of predictions of higher order are shootingwith a sling or with a bow and arrow.

Predictive behavior requires the discrimina-tion of at least two coordinates, a temporal andat least one spatial axis. Prediction will bemore effective and flexible, however, if the be-having object can respond to changes in morethan one spatial coordinate. The sensory re-ceptors of an organism, or the correspondingelements of a machine, may therefore limit thepredictive behavior. Thus, a bloodhound fol-lows a trail, that is, it does not show any pre-dictive behavior in trailing, because a chemi-cal, olfactory input reports only spatial infor-mation: distance, as indicated by intensity.The external changes capable of affecting au-ditory, or, even better, visual receptors, per-mit more accurate spatial localization; hencethe possibility of more effective predictive reac-tions when the input affects those receptors.

In addition to the limitations imposed bythe receptors upon the ability to perform ex-trapolative actions, limitations may also occurthat are due to the internal organization ofthe behaving object. Thus, a machine whichis to trail predictively a moving luminous ob-

ject should not only be sensitive to light (e.g.,by the possession of a photoelectric cell), butshould also have the structure adequate forinterpreting the luminous input. It is prob-able that limitations of internal organization,particularly of the organization of the centralnervous system, determine the complexity ofpredictive behavior which a mammal may at-tain. Thus, it is likely that the nervous systemof a rat or dog is such that it does not permitthe integration of input and output necessaryfor the performance of a predictive reaction ofthe third or fourth order. Indeed, it is pos-sible that one of the features of the disconti-nuity of behavior observable when comparinghumans with other high mammals may lie inthat the other mammals are limited to predic-tive behavior of a low order, whereas man maybe capable potentially of quite high orders ofprediction.

The classification of behavior suggested sofar is tabulated here:

It is apparent that each of the dichotomiesestablished singles out arbitrarily one feature,deemed interesting, leaving an amorphous re-mainder: the non-class. It is also apparentthat the criteria for the several dichotomies areheterogeneous. It is obvious, therefore, thatmany other lines of classification are avail-able, which are independent of that developedabove. Thus, behavior in general, or any ofthe groups in the table, could be divided intolinear (i.e., output proportional to input) andnon-linear. A division into continuous anddiscontinuous might be useful for many pur-poses. The several degrees of freedom whichbehavior may exhibit could also be employedas a basis of systematization.

The classification tabulated above wasadopted for several reasons. It leads to the sin-

ROSENBLUETH, WIENER AND BIGELOW 3 Behavior, Purpose and Teleology

gling out of the class of predictive behavior, aclass particularly interesting since it suggeststhe possibility of systematizing increasinglymore complex tests of the behavior of organ-isms. It emphasizes the concepts of purposeand of teleology, concepts which, althoughrather discredited at present, are shown to beimportant. Finally, it reveals that a uniformbehavioristic analysis is applicable to both ma-chines and living organisms, regardless of thecomplexity of the behavior.

It has sometimes been stated that the de-signers of machines merely attempt to du-plicate the performances of living organisms.This statement is uncritical. That the grossbehavior of some machines should be similarto the reactions of organisms is not surprising.Animal behavior includes many varieties of allthe possible modes of behavior and the ma-chines devised so far have far from exhaustedall those possible modes. There is, there-fore a considerable overlap of the two realmsof behavior. Examples, however, are readilyfound of man-made machines with behaviorthat transcends human behavior. A machinewith an electrical output is an instance; formen, unlike the electric fishes, are incapableof emitting electricity. Radio transmission isperhaps an even better instance, for no ani-mal is known with the ability to generate shortwaves, even if so-called experiments on telepa-thy are considered seriously.

A further comparison of living organismsand machines leads to the following infer-ences. The methods of study for the twogroups are at present similar. Whether theyshould always be the same may depend onwhether or not there are one or more qualita-tively distinct, unique characteristics presentin one group and absent in the other. Suchqualitative differences have not appeared sofar.

The broad classes of behavior are the samein machines and in living organisms. Specific,narrow classes may be found exclusively inone or the other. Thus, no machine is availableyet that can write a Sanscrit-Mandarin dictio-nary. Thus, also, no living organism is knownthat rolls on wheels — imagine what the re-sult would have been if engineers had insistedon copying living organisms and had thereforeput legs and feet in their locomotives, insteadof wheels.

While the behavioristic analysis of ma-chines and living organisms is largely uniform,

their functional study reveals deep differences.Structurally, organisms are mainly colloidal,and include prominently protein molecules,large, — complex and anisotropic; machinesare chiefly metallic and include mainly sim-ple molecules. From the standpoint of theirenergetics, machines usually exhibit relativelylarge differences of potential, which permitrapid mobilization of energy; in organisms theenergy is more uniformly distributed, it is notvery mobile. Thus, in electric machines con-duction is mainly electronic, whereas in organ-isms electric changes are usually ionic.

Scope and flexibility are achieved in ma-chines largely by temporal multiplication of ef-fects; frequencies of one million per secondor more are readily obtained and utilized. Inorganisms, spatial multiplication, rather thantemporal, is the rule; the temporal achieve-ments are poor — the fastest nerve fibers canonly conduct about one thousand impulsesper second; spatial multiplication is on theother hand abundant and admirable in itscompactness. This difference is well illus-trated by the comparison of a television re-ceiver and the eye. The television receiver maybe described as a single cone retina; the im-ages are formed by scanning — i.e. by orderlysuccessive detection of the signal with a rateof about 20 million per second. Scanning is aprocess which seldom or never occurs in or-ganisms, since it requires fast frequencies foreffective performance. The eye uses a spatial,rather than a temporal multiplier. Instead ofthe one cone of the television receiver a humaneye has about 6.5 million cones and about 115million rods.

If an engineer were to design a robot,roughly similar in behavior to an animal or-ganism, he would not attempt at present tomake it out of proteins and other colloids. Hewould probably build it out of metallic parts,some dielectrics and many vacuum tubes. Themovements of the robot could readily be muchfaster and more powerful than those of theoriginal organism. Learning and memory,however, would be quite rudimentary. In fu-ture years, as the knowledge of colloids andproteins increases, future engineers may at-tempt the design of robots not only with a be-havior, but also with a structure similar to thatof a mammal. The ultimate model of a cat is ofcourse another cat, whether it be born of stillanother cat or synthesized in a laboratory.

In classifying behavior the term teleology

ROSENBLUETH, WIENER AND BIGELOW 4 Behavior, Purpose and Teleology

was used as synonymous with purpose con-

trolled by feed-back. Teleology has been inter-preted in the past to imply purpose and thevague concept of a final cause has been of-ten added. This concept of final causes hasled to the opposition of teleology to determin-ism. A discussion of causality, determinismand final causes is beyond the scope of thisessay. It may be pointed out, however, thatpurposefulness, as defined here, is quite inde-pendent of causality, initial or final. Teleologyhas been discredited chiefly because it was de-fined to imply a cause subsequent in time to agiven effect. When this aspect of teleology wasdismissed, however, the associated recognitionof the importance of purpose was also unfor-tunately discarded. Since we consider pur-posefulness a concept necessary for the un-derstanding of certain modes of behavior wesuggest that a teleological study is useful if itavoids problems of causality and concerns it-self merely with an investigation of purpose.

We have restricted the connotation of teleo-

logical behavior by applying this designationonly to purposeful reactions which are con-trolled by the error of the reaction — i.e., bythe difference between the state of the behav-ing object at any time and the final state inter-preted as the purpose. Teleological behaviorthus becomes synonymous with behavior con-trolled by negative feedback, and gains there-fore in precision by a sufficiently restrictedconnotation.

According to this limited definition, tele-ology is not opposed to determinism, butto non-teleology. Both teleological and non-teleological systems are deterministic whenthe behavior considered belongs to the realmwhere determinism applies. The concept ofteleology shares only one thing with the con-cept of causality: a time axis. But causalityimplies a one-way, relatively irreversible func-tional relationship, whereas teleology is con-cerned with behavior, not with functional rela-tionships.

ROSENBLUETH, WIENER AND BIGELOW 5 Behavior, Purpose and Teleology