Control theory has equally serious implications for sociology. In particular,although the "control of perception" described by Powers differs subtly fromthe kind of interpersonal control discussed by most sociologists, I will arguethat control theory challenges our basic assumption that one person can havecontrol over the behavior of another. My objective in this article is toexplore the implications of control theory for our understanding ofinterpersonal control and social power. Because sociologists are generallyunfamiliar with the control-theory perspective, and because it challenges manyof our usual ways of thinking about human behavior, I will start by describingthe theory in some detail. In my discussion, I will refer to the theory as"perceptual control theory" (PCT), a designation which has the advantage ofemphasizing the control of perception as its distinguishing feature. Linking the input and output functions in the PCT model is a comparator, whichsubtracts the perceptual signal from a preset standard signal, calledthe reference signal. The difference between the perceptual signal andreference signal becomes an error signal, which by way of theoutput function (involving muscle tensions, if the system is an animal) isconverted into an output quantity and thus produces physical changes in thesystem's environment. These physical changes may affect the environment in avariety of ways, some of which are completely irrelevant to the purposes of theorganism, but some of the physical changes become a feedback function whichserves to counteract any environmental disturbances occurring at thesame time. The resulting input quantity, which is translated by theinput function into a perceptual signal, remains controlled at a level whichmatches the reference signal. This closes the feedback loop. In this example, the speedometer serves as the input function by transforminginformation about the rate of revolution of the car's wheels into a physicalsignal indicating the car's speed (in the form of some mechanical or electronicsetting of a meter), i.e., a perceptual signal. The cruise control mechanismitself performs as a comparator, subtracting the speedometer reading from theset cruising speed, or the reference signal. The resulting difference is theerror signal. The larger the error signal, the greater the change theoutput function produces in the accelerator setting, and thus thegreater the change in the output quantity of engine speed. Bychanging the amount of power to the wheels, and thus the feedback, the cruisecontrol system adjusts the speed of the car (as measured by the perceptualinstrument, the speedometer), quickly bringing it back to the reference value.(For other mechanical examples of control systems, see Powers 1979a). Note that the cruise-control mechanism performs exactly the same function asthe human driver when the cruise control is not in operation. This similarityis hardly surprising, because engineers originally designed servomechanisms, ofwhich cruise control is an example, in imitation of useful human capabilities(Powers 1978:418; 1989a:256ff.). If a driver wants her car to cruise ata certain speed but is not driving a car equipped with cruise control, she mustfirst accelerate the car to her chosen speed and then keep a constant eye onthe speedometer. When she notices the needle deviating from a given mark onthe dial, she either presses down on the accelerator or lets it up to adjustand control the car's speed (or, rather, the car's speed as she perceivesit). The perceptual signal for the driver's control loop comes from aseries of nerve impulses transforming the light rays from the speedometer whichstrike the retinas of her eyes. A comparator mechanism in her brain works byregistering as an error signal any image of a gap between the speedometerneedle and her chosen mark on the dial. Her central nervous system transmitsthese error signals as neural messages to the muscles in her leg and foot tomake the output movements which either depress or release the accelerator andso control the speed of the car. Moreover, the mechanical version of the cruise control system pays no attention(because it has no perceptual means for doing so) to any obstacles in the pathof the car; thus, it will inflexibly maintain a speed of sixty-five miles anhour, regardless of any slow-moving vehicles blocking the lane ahead. Whensuch obstacles arise, the human driver must take over control from themechanical system by stepping on the brake pedal or pushing a button todisengage the mechanism. If in an accident the car's wheels were to losecontact with the road (say, the car overturned in a ditch), the controlmechanism would keep the wheels turning at a rate corresponding to the desiredspeed until the engine stopped or the cruise control disengaged, even if thecar were going nowhere. As this example makes clear, a closed-loop controlsystem keeps certain perceptual inputs under control, but the actual output ofthe system is not controlled. Often, though not always, the process of controlling a perception has theby-product of achieving environmental control: the combined action of thecontrol system and whatever environmental disturbances are at work brings somephysical aspect of the environment to a constant or nearly constant level. Forexample, as long as the mechanical or human cruise control system worksproperly and nothing gets in the way, all observers will agree that the car ismoving down the highway at a more or less constant speed. Thus, perceptualcontrol may also have the effect of creating order or regularity in the controlsystem's environment (see Powers 1989a:268-69; Bourbon 1990:95-96). Although some controlling system, either human or artificial, must be inconstant operation in order to maintain such constancies in the environment, anoutside observer might well overlook the system's behavior, simply becausenothing seems to change (except the speed of the engine in our example). Underproper conditions, control systems like those described here can achieve avirtually instantaneous correction of errors to maintain a desired state ofaffairs. Occasionally an extreme environmental disturbance occurs, say a steephill or a sudden gust of wind, which requires an output beyond the rangeavailable to the system, and then the car slows down even though the throttleis fully open. Most of the time, however, the environmental disturbances wouldbarely be noticed by an onlooker; whenever a small discrepancy between theinput signal and the reference value occurs, the system very quickly increasesor decreases the output by an amount sufficient to correct the error. Thus, anactive control system may not draw the observer's attention either to thecontroller's behavior or to the environmental disturbances which the system isresisting. Nevertheless, we can tell control is occurring when something inthe environment which is susceptible to change does not change. (SeePowers 1979e.) Finally, note that servomechanisms employing negative feedback loops have noneed for sophisticated mathematical computations. Cruise-control devices, forinstance, predate the on-board computers so common in today's cars, and,working by analog computing principles rather than the digital logic ofcomputer chips, they are far simpler in their mode of operation. Cruisecontrol keeps a car's speed constant without receiving any information aboutthe velocity and direction of the wind, the degree of slope of the road, thefriction being generated by the tires on the road's surface, or any otherdisturbing factor (see Cziko 1992). Likewise, human drivers can keep a car'sspeed constant without reference to any such considerations. Merely bycorrecting errors in perceived speed, a negative feedback system can achieve alevel of performance and stability that would require complicated mathematicalcalculations and an enormous computing capacity if attempted in other ways. Obviously, the actions involved in performing the function of cruise controlare far more complicated for the human version than for the mechanical versionof the system. In order to control the perceived position of the speedometerneedle, a driver must at any given moment be able to accomplish very finelycalibrated perceptual and muscular control in many parts of her body. She mustbe able to aim her eyes at the speedometer and fix her eyes at the appropriatefocal length to see the dial. She must be able to feel the pressure of theaccelerator pedal on the bottom of her foot and must able to make minusculeadjustments in the angle of her foot in order to regulate the position of thepedal. She must be able to hold her leg in the proper position to reach thepedal and at the same time must be able to maintain the posture of her body andthe position of her head so as to see both the road ahead and the speedometer. On a more abstract level, the driver must somehow maintain her intention ofkeeping the car traveling at a given speed, but be prepared at any time toabandon that intention if an obstacle appears. While all this is happening,she must go through another equally complicated set of maneuvers in order tosteer the car. Simultaneously, she is quite capable of carrying on aconversation with her traveling companions, daydreaming about what she will dowhen she gets home that night, or switching the station on her car radio. Toaccomplish all these things at once, the driver's nervous system must encompassnot one but a great many control systems, all operating simultaneously. ThePCT model (Powers 1973, Ch. 6) proposes that human nervous systems do indeedcomprise a very large number of control systems organized and coordinatedhierarchically. The key to producing this remarkable coordination is that one control loop canprovide the reference signals for another loop. In the case of the mechanicalcruise control system, the reference signal of desired speed is simply imposedby a control system embodied in the driver, who may then reset the speed at anytime to stay in control of her own perceptions. Other common servomechanisms,like the thermostats of furnaces and air conditioners, have reference signalswhich can be reset by human users, while still other servomechanisms havereference signals built in by the engineers who designed them. But whoprovides the reference levels for the human's own control systems? To answerthat question we must look more closely at the central nervous system of thehuman body. According to the PCT model, the human central nervous system comprises at leasteleven hierarchical levels (Robertson and Powers 1989, Ch. 5). Concurrent with this flow of perceptual information upward through thecontrol-system hierarchy, output signals are moving downward in a similarfashion. At each level, comparators of the control loops produce error signalswhich are sent to the next lower level and recombined into reference signalsfor the control loops at that level, and so on down to the lowest level, whichactivates muscles and thus causes the physical movements of the organism in itsenvironment. In humans and other organisms, then, the source of a reference signal for acontrol loop is always another control loop at the next higher level in thehierarchy. Where, finally, do the control loops at the highest level of thehierarchy get their reference signals? Powers suggests at least twopossibilities (1973:173): first, that some reference signals may have beengenetically fixed in the process of evolution; and, second, that the referencesignals for many of the higher-level loops may be socially acquired, apossibility that sociologists should find intriguing. This lowest, or first-order level, registers only degrees of intensity (Powers1973, 1979a; Robertson and Powers 1989, Ch. 5). The perceptualsignals sent by systems at this level to higher levels are not just singleimpulses but are changes in the rates of firing of individual nerve cells orgroups of cells, with firing rates increasing or decreasing as intensitychanges. The driver controlling the speed of a car in our example getsinformation at intensity level from light rays striking the retinas of hereyes, from sound waves entering the chambers of her ears, from physicalpressures on the sole of her foot caused directly by her shoe and indirectly bythe accelerator pedal, and also from nerves which register the tensions on themuscles of her foot and other parts of her body. At the second layer of the perceptual hierarchy, neural circuits gatherimpulses from groups of first-level receptors and combine them in various waysinto the patterns we recognize as sensations, such as images of light and darkemanating from the speedometer of the car, or the "feeling" of exertion andpressure on the foot which comes from pressing on the accelerator. A thirdlevel of perception involves combining and controlling sensory patterns inorder to recognize specific configurations, such as the image of thespeedometer needle, the shapes of the numbers on the dial, or the angle atwhich one is holding one's foot. At the fourth perceptual level, humansrecognize and control transitions, such as the movement of the speedometerneedle or changes in the angle of one's foot. At the fifth level, humansperceive and control events, for example, the action of flooring theaccelerator pedal or the action of hitting the brake. An event may involvetransitions but also has a definite beginning and ending. The sixth levelinvolves control of relationships, such as coincidence or lack of it betweenthe speedometer needle and the mark on the dial which indicates the driver'sdesired speed. Perceived relationships may imply causation as well asjuxtaposition. For instance, relationships which might be perceived betweenthe driver and the car include not only the location of the driver in the frontseat but also the fact that the driver is driving the car. Higher levels of control have as yet been less well defined by PCT theoristsbut are generally of more interest to sociologists. Powers (1979a;Robertson and Powers 1989, Ch. 5) sees a seventh perceptual level arising whenwe use verbal symbols to name events and relationships and thereby place themin categories. An eighth perceptual level (Robertson and Powers1989:76) involves combining these categories of events and relationships intosequences, for instance, setting the speed for the cruise controlmechanism by first accelerating to the desired speed and then pushing a buttonto engage the mechanism. In sequences, the order of events is crucial. At theninth level, events, relationships, categories and sequences are arranged intoprograms of action for reaching goals (see Fararo and Skvoretz 1984). In thecontext of our driving example, a simple program might include something likethis: IF a stopped vehicle blocks the lane ahead, THEN hit the brakes.Programs, however, can contain much more complicated patterns of contingenciesand instructions and may indeed create many additional levels of perceptualcontrol all running in parallel (Powers 1979a:50, Robertson and Powers1989:78). Above the level of programs is a tenth control level which specifies theprinciples by which programs are chosen. An intention to be a "good driver,"for instance, might cause a driver to choose a program of action in order toregulate her car's speed, while a definition of the situation as an emergencymight prompt her to cast caution to the wind. Here, Goffman's (1974)conception of frames for action, which mediate between the program of actionand the identity of the actor, may be relevant. The highest (eleventh)perceptual level which Powers has described involves what he terms systemconcepts, which provide the contexts for defining the principlesused in selecting programs of action. According to Powers, system conceptsinclude our images of our selves: the woman in our example may see being agood driver as an important component of her self image (see Burke1991a, 1991b; Burke and Reitzes 1991). More broadly, systemconcepts could include our mental models of such social entities as bodies ofknowledge, or organizations, or even society as a whole (Robertson and Powers1989:78ff.). To make complex coordination possible, systems at different levels of theperceptual hierarchy must operate at different speeds. In general, the lowerthe level, the higher the speed of processing, because perceptions controlledby higher levels of the hierarchy are built up out of lower-order perceptionsand thus take longer to occur. For instance, the driver must integrate severalimages of the gap between the speedometer needle and the reference mark on thedial (the configuration level) in order to perceive that the needle has begunto move (a transition). At a higher perceptual level this movement eventuallyregisters as a change in the relationship between the speedometer needle andthe mark, and, an instant later the driver may reach the verbally formulatedconclusion that the car is slowing down and remind herself to initiate theprogram of bringing the car back up to speed (if she hasn't alreadyunconsciously started to press the accelerator). Thus, at any given instant, lower-order control systems are more likely to beseen by an outside observer as "in control" of their perception thanhigher-order systems because they respond faster to environmental changes.These differences in processing speed have been demonstrated in laboratoryexperiments (Marken and Powers 1989a). By allowing for the slower speedof processing of the higher-order systems and viewing each system in itsappropriate time frame, one can see that perceptual systems at all levelsconstantly control their input (Pavloski 1989a). The hierarchical arrangement of control loops means that the internal referencesignals which control perceptions can change and often do. In fact, accordingto the PCT model, many higher-order control loops achieve control ofperceptions only by sending out a constantly changing series of error signalswhich then produce changing reference signals for lower-order loops. Forinstance, suppose a driver wants her car to accelerate. This involves pressingdown on the accelerator pedal, or, in other words, changing the angle of herfoot. The perceptual system controlling this transition in foot angleaccomplishes it by sending a series of reference signals to the nextlower-order system, which controls the specific configuration: the angle ofthe foot (see Powers 1979a). If the foot starts out at, say, at 60deg.,the reference signals for the transition will be 59deg., 58deg., 57deg.,etc., all changing in rapid succession and of course all expressed in theinternal code of the nervous system. Simultaneously, a higher-order controlsystem will send reference signals to determine the timing of thesetransitions, i.e., when to speed up and when to slow down. In the face ofconstant environmental changes, the overall system can hardly be static ifhigher-level perceptions are to be controlled. Powers goes on to propose that memory signals may not only be sent downward asreference signals for the loop below but may also be sent upward as inputsignals (1973:219). When a control loop is short-circuited by substituting amemory signal for a current input signal, the person, in effect, replays theexperience in memory. Powers argues that we consciously remember something byperceiving it again, although in most cases the reliving seems like a mereshadow of the earlier occasion, because not all input channels are fullyengaged in recreating the former event. Powers calls this operation the"memory mode" of the control system. An ability to substitute memory signals for current inputs allows a controlsystem to function in several additional modes, besides action mode and memorymode, according to Powers. By reviewing memory signals, we can imagine whatwould happen if the remembered reference signals were put into effect in thecurrent situation. When sleep switches off our perceptual input, our controlsystems can still operate in imagination mode in our dreams. The memory modeof operation also permits a kind of "automatic pilot," or habitual behavior, inwhich one relies on memory to perform a routine action without paying closeattention to how the action is proceeding (see Powers 1973:220ff.). Finally,Powers describes another possible configuration of a control system: a"passive observation" mode in which the control system monitors the currentinput signal and stores it in memory but takes no overt action to counteracterrors, perhaps because we are actively controlling other perceptions at themoment. These uses of memory and imagination in the PCT model make possible a number ofbehaviors we recognize as typically human. On the one hand, the model explainsthe occurrence of habitual or stereotyped behaviors which are tightlycontrolled in accordance with reference signals based on the memories ofprevious performances. On the other hand, the model allows for episodes ofconscious deliberation, as one pauses to review in imagination previousexperiences and then selects from one's store of memories the course of actionwith the most favorable outcome to serve as a reference signal for currentbehavior. Memory is, of course, also implicated in processes of learning. Many kinds ofbehavior, especially actions requiring close control, depend on possession of awell-articulated memory trace of one's previous performances of the action,which can then be matched as a reference signal against the current perceptualinput. Practicing an action over and over again can refine and strengthen thereference signal in memory, so that the action becomes smoother, moreautomatic, more finely tuned, thus producing the simplest kind of learning:rote, or drill-and-practice. However, according to PCT, a rather more profoundkind of learning occurs when rote learning proves ineffective andreorganization of the system is necessary to overcome a chronic failure tomaintain good control. The first step in eliminating a prolonged mismatch in a control loop is tochange the reference signal. In the hierarchical PCT model, this adjustmenthappens as a matter of course when the next higher-order loop (which alwaysoperates more slowly than the lower-order loop) registers the error resultingfrom the failure of the lower-order loop to send up the proper perceptualsignal. The error signal produced by the higher-order loop then specifies asan alternative reference signal another memory trace. If the ensuing change inreference signal fails to reestablish control, the higher-order system willcontinue to cycle through the available repertoire of memory traces until oneof them works as a reference signal for the lower-order loop. If none of theavailable reference signals serves to reestablish control, the prolongeddiscrepancy for the higher-order loop will begin to register as an error in thenext level up and thus move the whole process of seeking appropriate referencesignals up another level in the hierarchy. If nothing works, the person mayeventually be compelled to give up and begin doing something entirelydifferent, thus changing the reference signal at a relatively high "program"level. Another way to think about such chronic errors is to consider the concept ofsystem gain. Engineers define gain as the amplification of a signal as it goesaround a control loop (see Powers 1979d:144). If small errors in inputare met with immediate and vigorous output, so that the perceptual signal neverhas a chance to wander far from the reference standard, the control system issaid to have high gain. Systems with high gain can control their inputs veryclosely, while systems with low gain must tolerate large errors. In general,Powers argues, human control systems tend to have rather high (negative) gains,with the gain increasing (and the operating speed slowing down) as one goes upthe perceptual hierarchy (see Marken 1990; Pavloski etal 1990). Although gainis largely a matter of the output function of a loop--the amount of output inproportion to the input--the feedback function in the environment and the inputfunction can also affect gain (see Figure 1). For instance, one might increasethe gain of a control process by applying more effort to the output, by usingof a power tool to affect the feedback function, or by employing more than onesensory modality (touching an object as well as looking at it) to enhance theinput function. Chronic perceptual errors occur not only when one's control systems have lowgain but also when one is caught between two incompatible goals. Powers (1973,Ch. 17) describes the result as system conflict taking place between twoequally matched control systems on the same hierarchical level. As he puts it,"Conflict arises when the systems attempt to use the same external variable toachieve goals that imply different states of that same variable"(1989a:292). Such conflict tends to immobilize both systems in anuncontrolled but high-output state where neither system can bring itsperceptions into line with its reference signals. If some environmentaldisturbance impinges on the variable in the environment that both systems aretrying to control, and the contested quantity is deflected toward the referencevalue of one of the systems, that system will experience less error and willrelax its output, while the other system will increase its output to cope withits perceived increase in error. As a result, whatever is being manipulated inthe environment is pushed back toward some point in the middle that satisfiesneither system. As fatigue builds up in the organism due to the high levels of fruitless effortrequired simply to maintain an unsatisfactory status quo, such conflicts candegenerate into wild oscillations and total loss of control. This is whatmight happen if slamming on the brakes of one's car did not automaticallydisengage the cruise control mechanism. The outcome is unpredictable butpotentially grisly. Within an organism, a typical example of low-orderconflict is a muscle spasm, or charley horse, which occurs when one set ofmuscles contracts but the opposing set fails to relax. A higher level conflictis illustrated by the indecision and vacillation which may result from havingsuch an overload of important commitments that one doesn't know what to startfirst. At any level, conflicts interfere with effective control. When something does work, reorganization stops and the organism thus haslearned a new way of perceiving the world, and in the same process has learnedthe behavioral outputs to accompany these new perceptions. A failure to findany adequate solution, however, may eventually be fatal if the "essentialvariables" out of balance pertain to the organism's supply of food, air, orwater. Nevertheless, when reorganization is successful, it generates animportant and useful kind of learning, as the organism's supply of functionalcontrol systems expands and adapts to the environment. In one of the few places in which Powers treats the subject of emotion(1992:31-40), he suggests that intrinsic error signals which promptreorganization are themselves perceived as the "feeling component" of emotions,although he observes that emotions generally involve a "thought-like" componentas well, and that some intrinsic error signals may not be sensed at all(1992:31-33). Drawing the same connection between intrinsic error and emotion,but from the other direction, Pavloski (1989b) argues that lack ofperceptual control is physiologically stressful, and he shows in the laboratorythat loss of control is correlated with the physiological changes, such as anelevated heart rate, which accompany emotional reactions. This equation of physiological error signals and negative emotional reactionscomplements certain sociological perspectives on emotion, particularlyHochschild's observation that "every emotion has a signal function," in thatfeelings provide "clues" to the "self-relevance" of perceptions (1983:28-29). Moreover, Hochschild's discussion of "feeling management" (1983:35ff.; seealso Thoits 1985) suggests the possibility of control systems in the human bodyfor perceptions of feelings, systems which work to reduce discrepancies betweenexperienced emotions and the emotions recognized to be situationallyappropriate according to "feeling rules" (Hochschild 1979, 1983:56ff.). The more positive varieties of emotion may be related in the PCT model toincreases in system gain. Just as loss of control and the consequent loweringof system gain invites negative emotions, which can be intensified by memoriesof previous losses of control, a rapid increase of gain often seems to bringpositive emotions.[6] Consider a driverencountering a traffic jam. The driver no longer perceives herself to bemoving toward her destination. While the sudden loss of gain may cause angeror frustration, recollection of a handy shortcut for getting around the jamwill increase the system gain once again and with it the driver's spirits. Integration of emotions into the PCT model helps account for yet another kindof human dynamism. Not only do emotions accompany learning by reorganization,they also appear to be involved in the more routine perceptual adjustments wedescribe as decision-making. From the PCT point of view, making a decision canbe defined as selection of an appropriate reference signal from those encodedin memory for some high-level control loop, such as a sequence or a program.Ordinarily, the decision-maker begins by trying out in imagination variousreference signals, to see which is likely to work best: Shall I do A,B, or C? For most people, the determination of what might work bestinvolves not only some tracing through of the likely consequences of each lineof action but also attending to the feelings associated in memory with eachreference signal. One decides on A, instead of B or C,because it might work and it "feels right." B might potentially beworkable but no fun. C is too frightening to contemplate. Thisperceptible "ping" of emotional memory which accompanies replay in imaginationof a reference signal thus guides our impulsive decisions. Deliberatedecisions are more likely to occur when perceptual conflicts force the decision"up a level" (see Powers 1992:41-53)to the level, say, of principles or perhaps system concepts. Such higher-levelsystems are not only slower in their operation, thus taking longer-range goalsinto account, but they also are more likely to invoke socially deriveddefinitions and identities, rather than idiosyncratic perceptions. Although the PCT model of the human is based on an intricate hierarchy ofclosed-loop cybernetic systems, the system as a whole is by no means closed toits environment. On the contrary, every control loop at every level passesthrough the environment, and energy and information flow copiously across theboundary between environment and organism. Thus, PCT humans are neithercontrolled by their environment nor by their own internal structure. Rather,they achieve control of their own perceptions, and as side-effects ofthat process cause observable regularities to appear in their own behavior andin the environment. The simple feedback principle which humans share withservomechanisms, can, when expressed in an appropriately articulated model,produce behavior of human complexity. Here, finally, is a conceptual model ofa human that acts like a human. NOTE: A reference list may be foundat the end of Part 2.WHATIS PERCEPTUAL CONTROL THEORY?
TheClosed Feedback Loop and How It Works
Overviewof PCT system model
Powers (1973) bases his psychological theories on a model of a closed-loopfeedback control system which differs in important respects from previousdepictions of systems (e.g., Bates and Harvey 1975; Miller 1978)familiar to most sociologists. Although the PCT model incorporates the usualarray of input and output mechanisms, interconnected components, and feedbackloops, Powers organizes these elements in an unconventional way (Figure 1). 
Exampleof a Feedback Loop--Cruise Control
An everyday example of a negative feedback loop, or servomechanism, whichconforms to this PCT model is the cruise control system of a car. A driversets the cruise control mechanism by pushing a button when the car hasaccelerated to some desired cruising speed. The control mechanism then takesover and begins constantly comparing the car's current speed with the desiredcruising speed. If the car slows down on an upgrade, the cruise controlmechanism automatically depresses the accelerator and thus bring the car backup to cruising speed. Similarly, if the car begins to exceed the cruisingspeed on a downgrade, the mechanism releases the accelerator to slow the cardown. Whenever the car is cruising at the desired speed, the mechanism makesno changes. The feedback loop is negative in the sense that any perceivederror, such as a drop in speed, produces action by the system to counteract theerror by changing the output of the engine. Howa Feedback Loop Works
Whether we are talking about the strictly mechanical or the human version ofcruise control, either system conforms to the PCT model by controlling onlyinput, not output. Both versions of cruise control keep the car's speed, asrevealed by the perceptual tool of the car's speedometer, at a predictable,constant level. However, the system's output--engine speed--will varycontinually and somewhat unpredictably as the car goes up and down hills orencounters other disturbances, such as headwinds or rough roads. Noteespecially that the system, whether mechanical or human, controls only itsperception of the car's speed. This perceived speed may or may notcorrespond to some outside observer's measurement of the speed of the car, asthe driver may find to her sorrow if her speedometer is incorrectly calibratedand she gets arrested for speeding. TheHuman-Machine Analogy
Our comparison between a human driver and her machine counterpart rests on morethan merely a behavioral analogy. Experimental results suggest that a deeperlevel of functional congruence must also be involved. Powers and a number ofother PCT researchers (Bourbon 1989, 1990; Bourbon, Copeland, Dyer, Harman andMosley 1990; Marken 1980, 1985, 1986, 1988, 1991; Marken and Powers1989a, 1989b; Pavloski, Barron, and Hogue 1990; Powers, 1973,1978, 1989b) have conducted computer-tracking experiments in which humansubjects use a computer joystick or mouse to keep a cursor pointed at a targeton a computer screen, while the position of either the cursor or target is alsobeing affected by a series of randomly generated disturbances. This taskduplicates the challenge faced by a driver who seeks to keep her car'sspeedometer at a constant speed. Such experiments have shown repeatedly that aPCT model can reproduce the handle or mouse movements of the human subject withconsiderable accuracy. Correlations between the recorded movements of thehuman and the predictions of the model range from about 0.95 up to severalfindings in excess of 0.99. Thus, the tracking behavior of the human subjectand the computer model are virtually indistinguishable, implying a strongsimilarity in functional organization between the human nervous system and thePCT model. Of course, humans exhibit many other types of behavior besidestracking, but the PCT model can be extended to account for a wide variety ofthese behaviors.TheHierarchical Organization of Control Loops in Humans
Although the successful operation of control systems can result in stability incertain aspects of the environment, a successful control system must beanything but static in its own operation. Furthermore, to have a repertoire ofbehavior flexible enough to cope with a wide variety of environmentaldisturbances, complex control systems like human beings must be capable ofmaking rapid changes in their own reference signals. While humansevidently operate using the same basic principles as servomechanisms, theirbehavior necessarily involves much greater dynamism and complexity. Howthe Human Perceptual Hierarchy Operates
Human brains are enormously complex, with tens of billions of interconnectedneurons. Such complexity allows for a very intricate organization of controlsystems within the brain and central nervous system. While much work remainsto be done before the number and description of hierarchical levels can befully mapped out (see Powers 1973, Ch. 7-14; Powers 1979a, Robertson andPowers 1989, Ch. 5), the proposition that human perceptual and behavioralsystems must be hierarchically organized appears to be well established (Markenand Powers 1989a; see also Simon 1973), and several of levels ofperceptual control can be described with some confidence. WhatLevels Make Up the Human Perceptual Hierarchy?
Within the human organism, the highest levels of this hierarchy are mostcentral, and lower levels are increasingly peripheral, so that, according toPowers (1973), higher level functions take place in the cerebral cortex of thebrain, while the neurons that incorporate lower levels are distributed in otherparts of the brain, the spinal cord, and throughout the body. Plooij (1984:13)describes the perceptual control hierarchy as a series of concentric spheres,like layers of an onion. The outermost sphere, which is the lowest level ofthe system, involves all of the various sense receptors in the body, thoselocated in the eyes, the ears, the nose, the mouth, and the skin, as well asthe proprioceptive receptors embedded in muscles and tendons that indicate thedegree of muscle contraction. HowDo Hierarchical Levels Communicate?
Each hierarchical level operates by matching perceptual input from lower levelsystems to the reference values supplied by higher level systems and by sendingout error signals to produce reference values for lower level systems. Exceptfor the very lowest level of the hierarchy, no level has any direct connectionwith the physical environment of the organism; however, at the lowest level theinformation coded into neural impulses is essentially meaningless untilcompiled and interpreted by higher level systems. In other words, the higherlevel systems must operate through multiple layers of lower level systems tocomplete their feedback loops, and various patterns of connections betweensystems at different levels are possible. Powers (1979b) theorizes thata control system typically connects to many other control systems, both higherand lower in the hierarchy. Figure 2 gives a simplified sketch of some ofthese connections. In Figure 2, dozens of control systems like the oneillustrated in Figure 1 have been arranged into the five lowest hierarchicallevels, and a small fraction of the possible connections have been traced. Asthe diagram suggests, control systems on higher levels of the hierarchycomplete their feedback loops by means of multiple connections through systemsat lower level of perception, till they reach the level that communicatesdirectly with the environment, through which all the control loops must pass.
TheHierarchical System in Action
Memoriesand Control
Another source of the dynamism in human organisms, according to PCT, is memory.Powers suggests (1973, Ch. 15) that memories are not inventoried in somewarehouse of the brain, but are rather stored on-site, throughout the levels ofthe hierarchy. These memories, which are essentially traces of previousperceptual signals, serve as an important link between the error signal at onelevel and the reference signal at the next lower level. In brief, Powershypothesizes that reference signals for lower-order loops are constructed outof memories of the previous input signals from those loops. An error signalproduces a reference signal for the lower level by locating the appropriatememory trace and sending it back down, where it can be compared to the currentinput signal for the lower-level loop. Thus, additional experience, byproviding a larger selection of memories to use in constructing referencesignals, can give a control system more flexibility in coping withdisturbances. PerceptualErrors, System Gain, and System Conflict
To this point my discussion has assumed that control systems are generallysuccessful in meeting their objectives, with perceptual input maintained atreference levels for all the levels of the control hierarchy. Although thismay be true for many kinds of behavior in many circumstances, humans,unfortunately, do not always succeed in controlling their perceptions. Undercertain conditions, no amount of adjustment of output will erase thediscrepancy between an input signal and a reference signal: if the hill issteep enough, the accelerator pedal can be pressed to the floor, but the carwill still lose speed. While temporary discrepancies of this kind may betolerable, a prolonged failure to match perceptions to reference levels demandsmore drastic measures.Emotionsand Learning by Reorganization
Prolonged episodes of system conflict or lack of perceptual control, Powersargues, provide the occasion for reorganization of the control hierarchy, whichresults in a more significant kind of learning than mere drill-and-practice.Reorganization occurs, Powers suggests, when failure to control perceptionsleads to somatic imbalance: errors arise in systems controlling certain"essential variables intimately associated with the physiological state of theorganism," and a "reorganizing system" starts rewiring the neural connectionsbetween levels of the sensory-motor control hierarchy (1973:182ff.). Theprinciple behind reorganization according to D. T. Campbell is that of "blindvariation and selective survival" (quoted in Marken and Powers1989b:379), the same principle as biological evolution. The organismexperiencing prolonged lack of control, having unsuccessfully run through itsrepertoire of reference levels in memory for a given level, will begin randomlytrying out new reference signals, often by making new connections betweenlevels, and its behavior will begin to show unpredictable and perhaps evencreative variations until something works to reduce the intrinsic errors sensedby the reorganizing system.[5] Summary: The PCT Model of the Human Actor
We see from this discussion of hierarchy, memory, conflict, reorganization, andemotion that PCT presents a model of human beings who are intentional actorsand who can relate dynamically to their environments and to other people. PCTactors can decide on a course of action and carry it out in the face ofenvironmental disturbances. They can adjust their intentions to cope withchanging situations and modify their environments to suit their purposes. Theycan display both the rule-regarding predictability (see Petrie 1981) and theimpulsive spontaneity that together seem to epitomize human action.