Revised draft of a paper presented to the 1996 Annual Meeting of the ControlSystems Group, Flagstaff, Arizona.

• INTRODUCTION
• FUNDAMENTALS OF PERCEPTUAL CONTROL THEORY
  • The PCT Model and Its Empirical
    Support
  • Characteristics of the PCT Model
• SIMULATIONS OF SOCIAL
  INTERACTION
  • Simulating Cooperation
  • Simulating Conflict
  • Lessons from Simulations of Conflict and Cooperation
• SOCIAL STRUCTURE AND CULTURE AS COLLECTIVE CONTROL
  • Social Activities
  • Social Designations
  • Cultural Contingencies
  • Cultural Discourses
  • Connections Among Domains
• CONCLUSION
• APPENDIX
  • Modeling Control-System Interactions
  • Limitations of This Investigation
• REFERENCES

This article presents a new way of thinking about sociological theory, based onpsychological insights from Perceptual Control Theory (PCT). After reviewingthe PCT model of goal-directed behavior and describing its empirical base inexperimental studies of tracking behavior, I report the results of computersimulations which apply the PCT model to elementary social interactions. Mykey finding is that the collective control of perceptions can stabilizevariables in a shared environment, even when interactants conflict. Suchconflictive control, I argue, occurs frequently in both macro and microsituations, leading observers to mistakenly infer that some unknown collectiveactor controls the outcome. After discussing the characteristics of "virtualsocial actors" which emerge from collective control of perceptions, I outline aframework for classifying social interactions from a PCT perspective, anddescribe the three main hierarchical layers of human perception. Finally, bypresenting four broad domains of collective control, I offer innovativedefinitions of social structure and culture as collectivelycontrolled perceptions. My conclusion discusses the implications of this newperspective for a person-centered sociology.

[T]he imaginations which people have of each other are the solid factsof society, and to observe and interpret these must be the chief aim ofsociology. (Charles Horton Cooley 1902, p. 87, emphasis in original)

INTRODUCTION

Sociologists who read philosopher Stephen Turner's book, The Social Theoryof Practices (1994), are likely to find it depressing. A formersociologist and friendly critic of the field (1980; Turner and Turner 1990;Wardell and Turner 1986), Turner here carries his unflinchingly logicalarguments to a disturbing conclusion--that social theories based on any sort of"hidden collective objects" (p. 102), whether Durkheim's "social facts," orParson's "internalized norms and values," or even contemporary theories of"culturally shared practices," have no value as causal explanations for socialbehavior. These theoretical conceptions of collective entities, according toTurner, serve merely as "analogical objects" which may promote a vague sense ofunderstanding but refer to nothing demonstrably real (1994, p. 60).

Although sociologists can easily observe patterns of similarity in thebehaviors of individuals in any society, Turner argues, no one has yet observedany of the internalized collective entities which allegedly cause thosesimilarities (1994, pp. 18-19). Even when two individuals' overt acts areindistinguishable, we cannot safely infer that the "presuppositions" of theactions, or "techniques" of producing the actions--or whatever the unseenentities that individuals theoretically share--are exactly the same (1994, pp.28-43). The crux of the problem, says Turner, is transmission. How do thesehidden collective entities come to be located in individuals (1994, p. 60)? Ifpractices have no tangible substance, the usual ways of acquiring knowledgearen't sufficient to ensure that the entities internalized by differentindividuals are identical, as they must be, Turner maintains, for the practicesto be identical (1994, pp. 61-63, 76-77). And if practices can somehow bereproduced identically in different individuals, then how can those practicesever change (1994, p. 83)?

Having reached the comfortless conclusion that all such theories of hiddencollective objects must fail because of "epistemic problems" (1994, p. 103),Turner retreats to a common-sensical but sociologically unsatisfying positionthat similarities in behavior arise simply out of the habits that individualsadopt as they observe and emulate others. In his words,

. . . what we have are private habits, with a variegated causal structure, thatarise in response to public things, [and] there is nothing to be learned aboutthem beyond the summary description of practice in terms of its overtmanifestations, because there is nothing there--nothing non-individual, thatis. (1994, p. 105)

Turner can see how his argument devastates most sociological theories and putssociologists in the embarrassing position of having nothing interesting tostudy. Sociologists, he suggests, could sidestep the problem as Weber did byavoiding any reference to unconscious processes and by focusing entirely onovert actions and on motivations that can be explicitly verbalized. But anyanalysis which tries to go deeper by invoking unseen collective entities willinevitably, he warns, run "into the sand" as its initial successes fail tocumulate beyond "a certain level of fuzziness" (1994, pp. 114-116).

If Turner were correct, sociological theorists might as well give up and startlooking for a real job. Fortunately for the field, however, Turner's analysisis incomplete. This paper presents a new theory of collective socialentities, a theory that reveals both their illusory essence and their realconsequences. The theory agrees with Turner that collective entities, likesocial structure, culture, tradition, group identities, and a host of other"collective nouns" (Alford 1995), have no existence independent of individualperceptions. And the theory affirms Turner's argument that individuals differin their ways of developing the perceptions and in the perceptions themselves.The theory recognizes, however, that individuals do share such perceptions atleast approximately, and it shows how the regularities of social structure andculture can emerge from such sharing, even when the sharing is imperfect.

To be precise, this paper shows how collective phenomena can be understood andanalyzed as collectively controlled perceptions. Working from aninnovative view of individual psychology (Powers 1973, 1989c, 1992)called Perceptual Control Theory (PCT), I argue that sociologists can constructon this new theoretical base a scientifically testable model of the emergenceof collective entities out of social interaction, a model invulnerable toTurner's criticisms. Moreover, the model can provide new and more satisfactoryanswers to fundamental questions which have troubled sociologists for years:for instance, how can we bring into a coherent theoretical whole suchoppositions as order and conflict, structure and process, and stability andchange?

In an earlier paper (McClelland 1994), I presented a comprehensive introduction toperceptual control theory and its implications for the sociologicalunderstanding of power. Since PCT is still relatively unknown, I begin thispaper with a review of the fundamentals of PCT, focusing on the theory'sempirical base in experimental and computer-modeling studies. I next describesome basic findings from my own computer simulations of PCT models in "social"interaction, results which show how the illusion of a "social actor" can emergefrom the interactions of individuals independently controlling their ownperceptions in a common environment. I go on to discuss the characteristics ofthe virtual collective entities produced by collective control of perceptions.Finally, I present a new map of the social landscape by offering a preliminarycategorization of the kinds of perceptions which can be collectivelycontrolled. My paper closes with a discussion of the implications of this newviewpoint for progress in sociological theory and empirical research.[2]

FUNDAMENTALS OF PERCEPTUAL CONTROL THEORY

Most sociologists give minimal attention to the psychological assumptionsunderlying their theories; if they worry about it at all, sociologistsgenerally start from the premise that people produce actions in order to reachgoals. In other words, social behavior is goal directed, and, when proceedingrationally, actions are under control. PCT makes only one small but highlyconsequential change in this line of thinking. PCT holds that human behaviorresults from controlling perceptions, not actions. In other words, people actin order to stabilize their own perceptions or to match their perceptions tosome desired or imagined state of the world.

What does it mean to control perceptions and not actions? First one mustunderstand the concept of control as it is used in PCT. The phenomenon ofcontrol, per se, is easily visible. No observer could fail to see how oftenthe outcomes of human action are highly controlled. Social life is repletewith astounding regularities, in which people manage to reach the same goalover and over again in spite of changes in the environment, unpredictableobstacles, and even active opposition from other people. Millions ofcommuters, for instance, brave the rigors of the highways or the subways tomake their daily trek to work. Their predictably repeated arrival at work isevidence of control.

Control in the PCT sense does not mean triggering changes in things, nor makingother people do what you want (see McClelland 1994). Control, as used here, meansto stabilize a pattern, to keep it constant. Control occurs when thingsdon't happen that might otherwise take place, because someone acts toprevent change by counteracting the forces that cause change. For example, acity abandoned, bereft of human control, inevitably begins to decay as forcesof nature erode it. By actively controlling their own perceptions of theirimmediate environment, the inhabitants of a city keep it stable enough to beviable. In this sense, every structure built by humans (and indeed everysocial structure) is a monument to human control. Thus, if we want tounderstand how social structures are achieved, we must understand how controlis achieved.

A basic principle of PCT is this: only what can be perceived can becontrolled.[3] If perceptions are cut off,control is lost--when the lights go out abruptly, we stumble and fumble.Moreover, PCT implies that our actions, as such, are seldom under consciouscontrol, because in most circumstances we have no reason to pay attention toour own movements. The consequences are what we care about, not the actionsper se. We act pragmatically, doing whatever is necessary to make ourperceived world conform to our goals. Part of the reason for inattention toour own actions is that our perceptual apparatus is better designed formonitoring other people than ourselves. We cannot see ourselves as others seeus. Of course we do get some perceptual clues about our own actions--we cansee movements of hands and feet, can feel efforts of our muscles, can hear ourwords through our headbones, can feel expressions flitting across our face--butsuch perceptual feedback is severely limited, as one quickly perceives from anaudio or video tape of one's own behavior. Even when directing attention mostclosely to our actions and doing our best to act in a "controlled" manner (atdinner with prospective in-laws, for instance), we end up controllingperceptions of our actions, not the actions themselves, because our perceptionsare our only means for controlling anything.

The PCT Model and Its Empirical
Support

But how, according to PCT, can control of perceptions be achieved? PCT differssignificantly from previous psychological theories because it explains humanbehavior by a generative model of neural processing--a model both simple enoughto be expressed in a few concise mathematical equations (see Appendix), andpowerful enough to reproduce observed behavior accurately. Unlike behavioristpsychologists, PCT theorists assume that the interior organization of theorganism has as much impact on its behavior as does the environment, and thusthey agree with cognitive psychologists that one cannot understand behavior bysimply focusing on input and output and treating the organism as a black box.Unlike most cognitivist approaches, however, PCT does not rely on vaguelyworded suggestions about the way the neural system is organized but insteadprovides a mathematically defined and testable model. Unlike the approach ofcognitive psychologists working on artificial intelligence, PCT starts from acore model of neural organization simple enough to be simulated in a few linesof computer code, one with which researchers can duplicate ordinary behavior inreal time without recourse to the computations of a super computer. And incontrast to all of these other approaches, the PCT model has been shownempirically to predict certain samples of individual human behavior withnear-perfect exactitude.

In their empirical tests of the PCT model, Powers (1973, 1978, 1989d;Marken and Powers 1989) and other PCT researchers (Bourbon 1989, 1990, 1996;Bourbon et al. 1990; Marken 1980, 1985, 1986, 1988, 1991; Pavloski, Barron, andHogue 1990) have focused on a type of goal-directed behavior called tracking.A person performing a tracking experiment sits in front of a computer screenwith two cursors, one of which moves up and down or side to side in a patternrandomly generated by the computer, while the other stays at a fixed point.The subject's task is to line up the two cursors. By manipulating a mouse orjoy stick which affects the position of the moving cursor, the experimentalsubject can resist the cursor's random movements and thus control its placementon the screen. If the cursor starts to move left of the target, the subjectmoves the mouse right, and vice versa. Experimental subjects never quitesucceed in keeping the cursor perfectly steady, but most achieve a level ofcontrol in which the cursor stays near the target with relatively littlemotion. A properly calibrated PCT model almost precisely matches the mouse orhandle movements of a well-practiced subject completing this task, includingthe characteristic errors that human subjects make. For streams of continuousbehavior lasting up to several minutes, the PCT model has been repeatedly shownto predict an individual's behavior with impressive accuracy, the correlationsoften exceeding 0.98. Bourbon and his co-workers (1989, 1990; Bourbon et al.,1990) have extended this research to elementary social behaviors bydemonstrating that certain cooperative tracking tasks, which involve closecoordination of behavior, can be performed by two humans, or by a human and acomputer-simulated control system working together, or by two simulated controlsystems, with nearly indistinguishable results. In sum, PCT researchers havedeveloped a model which in sheer predictive power easily surpasses otherpsychological research--reason enough for sociologists and others in the humanand life sciences to take PCT seriously.

How exactly does the PCT model work? Basically, the model provides a proposedorganization for the central nervous system. Although the human brain containsbillions of neurons connected in circuits of formidably complexity, the PCTmodel proposes that a simple circuit repeated and interconnected thousands ormillions of times serves as the building block of neural organization. The basic PCT circuit, a negative feedback loop, is shown in Figure 1.Arrows in the top half of the model (above the line separating systemfrom environment) stand for nerve signals, or rates of firing ofindividual neurons.[4] What happens in the tophalf of the loop is that a nerve signal generated in response to contact withthe environment (a sensor signal in PCT terms) is compared to anothernerve signal constructed from memories of earlier perceptions (shown as thearrow coming in at the top of the figure), called a reference signal.Comparison of the two signals consists of a continuous process ofsubtracting one from the other, producing an error signal. The errorsignal, transmitted to the appropriate muscles, activates the movements we callbehavior.

We can see how this model applies to a tracking experiment by beginning withthe sensor signal constructed by the brain from light striking receptors in theeyes.[5] This sensor signal, of course, changeswith the instantaneous position of the moving cursor on the screen.Concurrently, the subject constructs, by selecting an image from memory, areference signal for the relationship between the two arrows, a signal whichrepresents the desired perception of the two cursors perfectly aligned. Thesensor signal and the reference signal meet at a comparator circuit within thebrain which subtracts the sensor signal from the reference signal in order toproduce an error signal. When the two cursors line up on the screen, the errorsignal is zero, and the error signal constantly varies with the size of theperceived discrepancy, getting larger any time the distance between the arrowsincreases. This error signal travels back down the spinal cord to activatemuscles in the arm and hand grasping the mouse. The larger the error, the morethe subject moves the mouse.

So far I have described only the top half of the circuit shown in Figure 1.Note the important similarities between this part of the PCT model and modelsproposed by cognitive psychologists. Like cognitivist models, the PCT modelassumes that input in the form of stimulation from the environment gives riseto perceptual data in the nervous system, which is then processed internallyand emitted as output in the form of observable movements.[6] Several considerations, however, make the PCT proposalradically different from conventional accounts of behavior. First, the PCTmodel depicts continuous streams of behavior, not discrete stimuli andresponses. Second, PCT is explicitly a model of intentional behavior, and themodel shows how the physical structures and processes of the brain can generatepurposes or goals. Applied to the tracking experiment, the PCT model describesan intentional actor--the subject--engaged in purposefulbehavior--tracking--with the reference signal serving as the physicalembodiment of the subject's goal. Although many cognitive psychologists talkin terms of an intentional actor (for example, Pylyshyn 1984, pp. 9-10;Barsalou 1992, pp. 64-67), their models still attribute the ultimate causes ofbehavior to environmental stimuli, a contradiction between theory and rhetoricmost psychologists have failed to recognize. Finally, the PCT model (in thebottom half of Figure 1) draws attention to an important phenomenon whichcognitive theorists have largely neglected: the instantaneous environmentalfeedback of a person's own actions.

Putting it bluntly, all previous psychological analyses of behavior have beenfaulty, because none except PCT has taken into account the fact that virtuallyevery stimulus an organism receives is simultaneously contaminated by theeffects of the organism's own actions.[7] Ourbodies are part of the environment. Thus, the slightest movement we makeimmediately changes the environment from which we receive all further stimuli.Our mere presence and the way we direct our attention have unavoidable impactson the environment we experience. In short, environment and behavioral outputare inextricably intertwined in our perceptions, and any theory which givescausal weight to the effects of the environment on an organism's perceptionsbut ignores the simultaneous effects of the person's own actions on thoseperceptions is incomplete.

How does feedback from actions affect a perception, according to PCT? Thelower half of Figure 1 indicates the essential steps. Arrows below the linewhich separates system from environment stand for physical forcesor processes outside the person's skin. As Figure 1 indicates, output causesphysical changes in the environment which become feedback, as theycombine with other environmental forces to create the input variable.(Note that a person's behavioral output may also have some physicalside-effects having no impact on the perception controlled). The person's ownactions can thus cancel some of the effects of other environmental forces whichtend to produce changes in the environmental stimulus and so might change ordisturb the sensor signal. In the model, all such environmental forces aredisturbances. These environmental forces combine with feedback from theperson's own actions to determine the physical state of the environmentalvariable being perceived.

In a tracking experiment, for example, if a subject were to stop moving themouse, the cursor affected by the mouse would no longer stay in one place.Instead, it is programmed to wander randomly across the screen, which of coursethen disturbs the subject's perception of alignment between cursors. PCTtheorists refer to this pattern of environmental change, absent any action bythe person, as "the disturbance." The magnitude of the disturbance isproportional to the amount of movement that occurs whenever the subject stopsmoving the mouse. Subjects can control the position of the cursor and thustheir own perception of alignment by moving the mouse just enough to counteractthe disturbance, assuming of course that the disturbance stays withinreasonable bounds and does not occur too rapidly or too strongly for the personto overcome. Subjects who control successfully are able do it without directlyperceiving the magnitude of a disturbance or knowing anything ahead of timeabout its pattern. All they need to do is to correct perceived errorsimmediately as errors begin to occur. Specifically, if the cursor is perceivedas off-target in one direction, the error signal in the control system'snegative-feedback loop results in muscle movements which push the mouse in theother direction. The more the disturbance tends to separate the cursor and thetarget, the more the mouse must be moved to bring the perception intocontrol.[8]

Characteristics of the PCT Model

In the PCT model, unlike conventional psychological models, the input does notcause the output any more than the output solely determines the input. Themodel does not work by chains of cause and effect (as in conventional models ofcognitive psychologists), because the variables in the loop are all affectingeach other simultaneously. In fact, to find causal paths in this model, onemust trace the loop backwards. The reference level, the actor's purpose,determines perceptual input: when the model is working correctly theperception is stabilized at precisely the reference level, that is, withcursors in alignment. Similarly, the random disturbance causes the pattern ofphysical output, the movement of the hand holding the mouse, as long as thereference signal remains unchanged, because the output must be a negativefunction of the disturbance in order to counteract it (see Marken 1980, 1990a).For example, a typical graph from a tracking experiment of the output anddisturbance over time (Figure 2) shows the output (or mouse position) as nearlya mirror image of the disturbance.[9] Thevertical axis in Figure 2 represents distance measured by deviation from anarbitrary zero position on the computer screen, while the horizontal axisrepresents time measured in sixteenths of a second. Note also in Figure 2 thatthe position of the cursor jiggles a bit but stays almost constant, because thecontrol system has acted to stabilize its perceptual input.

An important thing to remember about the PCT model, in comparison toconventional cognitive models, is that everything happens at once. Although totalk about it we must trace the events around the feedback loop, what reallyhappens is continuous input and continuous output and continuously changingenvironmental disturbances. Consider a goal-directed task similar to trackingbut more often encountered in everyday life. You are trying to follow anotherindividual through a throng of people. The person you're following doesn'tstop for you to catch up, nor do other people stop moving to let you pass; andyou yourself must move constantly to keep track of the person you're following.Only a model of continuous action can possibly describe this behavior. Thestandard cognitivist model of interpretation of stimulus, followed by cognitiveplanning of action, followed by execution of the planned response, has nochance of capturing the dynamics of the situation.[10]

In sum, the PCT model shows how the human nervous system can be organized sothat people can produce a continuous stream of goal-directed behavior withoutnecessarily planning actions in advance or doing any complicated computations,and so that people as control systems can perform successfully in environmentsfull of unpredictable disturbances. The key, as we have seen, is theerror-correcting capacity of negative-feedback loops to control perceptualinput. Of course, a fully worked out model of the human nervous system wouldbe immeasurably more complex than the single control loop illustrated in Figure1.[11] But tracking experiments havedemonstrated how this core version of the PCT model can accurately simulate theway in which people achieve simple goals. In the next section I will turn toquestions of how perceptual control systems interact, when two or more arefocused on the same environmental variable. I will argue that simulationsbased on simple PCT models can provide new insights about patterns of humaninteraction.

SIMULATIONS OF SOCIAL
INTERACTION

In recent years, several sociologists (Burke 1991a,1991b; Burke and Reitzes 1991; Fararo 1989; Fararo andSkvoretz 1984, 1986; Heise 1977, 1979, 1987; McClelland 1994; McPhail1989a, 1989b, 1991, 1994; McPhail and Tucker 1990;McPhail, Powers and Tucker 1992; Riley and Burke 1995; Stets 1993,1995a, 1995b; Tucker and Stewart 1989) have begun toreconceptualize the study of social interaction on the basis of the PCT model.For individuals, this process of perceptual control is well understood, atleast in principle, but its implications for the behavior of groups of controlsystems sharing a common environment have been much less studied.[12]

The Appendix describes a series of computer simulations using a simple PCTmodel to explore patterns of interaction among independent control systemslinked only by their common surveillance of a single environmental variable.My intent in presenting these simulations is to map out the interactionalpossibilities available to small numbers of generic control systems sharing thesimplest of environments--a one-dimensional variable. While such simulationsgreatly oversimplify the complexities of human interaction, the assumption thathumans behave control as control systems in pursuing their goals implies thatthe basic principles discovered in these simulations should also apply to humaninteractions.

In Figure 3, the behavior of a control system simulated by a computer model(see Appendix) has been superimposed on the data displayed in Figure 2, whichcome from an actual tracking experiment. When the simulated control systemencounters the same random disturbance as was used in the tracking experiment,the curve of its output closely matches the mouse positions produced by thehuman subject (r = 0.998 for 500 data points collected in a 30-second run).Similarly, the modeled cursor position reproduces the main lines of movement ofthe cursor during the experiment.[13]Clearly, this simulated control system can reproduce human tracking behavior.

Simulating Cooperation

My first step in modeling the interaction of control systems will be to showhow simulated systems can work together on a tracking task. Devising a modelfor cooperative behavior is a good starting point for applying PCT to socialinteractions. Cooperation, of course, is a basic feature of social life, andwithout cooperation no social structure could be maintained. One can modelcooperative action with PCT by setting up two or more control systems toperform a control task together. In the first simulation reported here, twocontrol systems get their perceptual input from the same environmental variable(cursor position), and their outputs, added together, determine the simulatedmouse position. The two control systems are programmed to operateindependently, connected only by their joint monitoring of the singleenvironmental variable. Although in real life two independent control systemsmight often have different reference standards for a given perceptual variable,the simulation makes the simplifying assumption that their reference signalsare exactly equal to each other and to zero, as they would be if two peoplewere cooperating to keep the cursors in a tracking experiment preciselyaligned. In this simulation, the only way in which the two independent systemsdiffer is in their gain, that is, the rapidity with which they move tocorrect their perceived errors. In general, systems with higher gain controltheir perceptions more tightly, tolerating less variation from the referencestandard (see McClelland 1994, p. 474). For this simulation, I gave one system again of 300 and the other, 200 (figures arbitrarily chosen to add to 500, thegain which I previously used in constructing a single control system tosimulate the experimental data, as shown in Figure 3).

Figure 4 shows the joint action of these two cooperating systems. Becausetheir gains are different, their output curves are easily distinguishable, withthe higher-gain system having the greater output in both directions.[14] The results show that their outputstogether add up to exactly the same output as was produced by the singlesimulated control system shown in Figure 3. Furthermore, the impact of theirjoint action (together with the random disturbance) on the controlled variable(the movement of the cursor) is also equal to that of the single simulatedsystem. The dark line showing simulated cursor position in Figure 4 coincidesprecisely with the simulated cursor position in Figure 3.

I have performed many variations of this simulation (not reported here) usingdifferent numbers of cooperating control systems, different combinations ofgains, and different disturbance patterns. All these simulations support thesame conclusion: when two or more independent systems use identical referencevalues to control their own perceptions of a single environmental variable,their joint effect on that environmental variable exactly duplicates the effectof a single control system acting alone. Furthermore, their system gains areadditive. The cooperating systems produce the same environmental effect as asingle system with a gain equal to the sum of all their gains. Thus, each ofthe cooperating systems can be relatively weak, but the environmental result oftheir collective control of perceptions is indistinguishable from the resultproduced by the more powerful single system. Evidently, the common-sense loreabout "strength in numbers" applies directly to cooperating control systems.

These simulations, then, together with the coordinated tracking experiments ofBourbon and his co-workers (Bourbon, 1989, 1990; Bourbon et al., 1990),demonstrate that the PCT model can reproduce and thus account for thefundamental social fact of cooperation. Control systems are capable ofcooperating to control a perception jointly, and under optimum conditions theircooperative behavior is additive in terms of system gain, with an impact on theenvironment equivalent to that of a single super-control system.

Of course, human groups often fall below optimum in the efficiency of theircooperative efforts. Though one hundred people working together may be able toaccomplish tasks that an individual could not hope to tackle, one hundredworking on the same task are rarely a hundred times more effective than asingle individual. The main problem is that people physically get in eachother's way and block each other's view. Five or ten people might be able tolift a heavy object much more easily than a single person, but, unless theobject is very large, one hundred people cannot grasp it at once. Thesimulations assume that each cooperating system has an identically perfectperception of the environmental variable and that each system's output impingeswith no loss of energy on the environmental variable. However, when people'sperceptual channels are blocked, or when their output channels are interferedwith, they can no longer contribute efficiently to the joint effort of control.Human groups can get around this difficulty by dividing tasks into specializedsubtasks assigned to individuals and by using remote techniques ofcommunication such as computers, phones, or radios to allow many people toperceive a given environmental variable simultaneously. Still, the physicalenvironment puts limits on how efficient human cooperation can be, and anyfuture simulations of cooperative behavior will need to allow for less thanoptimal perception and for inefficient output (which can be done by changingthe mathematical form of the input and output functions in the PCT model).

These simulations of cooperative interaction have relevance for sociologicaldiscussions of power. In an earlier article (McClelland 1994), I argued thatsocial power occurs when people align their reference standards for controllinga perception. Simulations of cooperating control systems show how such socialpower can be achieved. The joint actions of a large number of people who agreeabout the "right way" to see something or to do something can have the force ofa superman in their effect upon the environment, and any individual who getsthis group to change their minds in concert about what's right will seempowerful indeed. Moreover, the deviant individual who sees or does thingsdifferently from the group must confront powerful opposition in thatenvironment.

Must we infer, however, that social power in everyday situations can only bebased on complete consensus among the cooperating people about the referencestandard? In this first simulation, I arbitrarily set the reference values tobe equal for the cooperating systems, as if people could find perfect agreementabout their common projects. Obviously, what one person thinks is happening insome joint project might often be different from another person's perceptions.Yet they can still manage to cooperate, even without perfect agreement. How isit possible? Turner (1994) attacks precisely the same point in his criticism(discussed above) of social theories which hold that the same "practice" or"norm" is somehow located in every individual in the group. Habits differ, heargues, just as individuals differ.[15]Interestingly enough, PCT perspectives on human learning suggest a similarconclusion. PCT theorists argue that humans develop their perceptual controlsystems through a process of random reorganization of neural connections, whichoccurs most rapidly when the person experiences chronic error from inability tocontrol an important perception, and which eventually ceases when the processyields neural circuitry capable of adequate control (see Powers 1973, ch. 14;Cziko 1995, ch. 10). This random learning process makes it possible that twoindividuals can end up with different reference values or even differentcontrol systems for perceiving key variables in their environment. In otherwords, people often learn to see and do things in their own idiosyncratic ways.Realistic models for cooperative behavior must cover situations in which theparties involved cooperate but use different reference values for theperception they collectively seek to control.

There is a catch, however. Cooperation by parties using different referencevalues is something less than pure cooperation. In fact, PCT defines theinteraction of two control systems seeking to control the same variable atdifferent reference values as conflict (Powers 1973, ch. 17). Powershas argued that this sort of conflict between the control systems within anorganism can have "extraordinarily bad consequences" (1973, p. 253) and thatour attempts to control other people's behavior inevitably lead to conflictbetween people, sometimes with dire consequences like violent crime or war.How can cooperation also be conflict? Before going any further to simulatecooperation, we must understand what happens when control systems get involvedin conflict.

Simulating Conflict

To simulate conflict in a control-system world, one only need give interactingsystems two different reference values for controlling the same perception.Figure 5 shows the results of one such simulation. The disturbance curve isthe same as in all the previous simulations, and the control systems areidentical to cooperating systems shown in Figure 4, except that the system withhigher gain has been given a reference value of 1.0, and the other system areference value of -1.5. The gap in reference values is so small that linesfor the two are indistinguishable from zero if shown on a graph like that inFigure 5. Nevertheless, this tiny difference in reference values generates thecharacteristic pattern of control-system conflict: a rapidly increasingdivergence in the outputs of the conflicting control systems. Note that thevertical scale of this graph has had to be extended to accommodate the sharplyseparating output curves. Figure 5 demonstrates that control systems, likepeople, start pulling in opposite directions when they disagree.

Although the sharp divergence of output curves in Figure 5 is the most dramaticfeature of the graph, the least interesting-looking curve in the graph alsomerits attention. Note how the curve for cursor position, the controlledvariable in the previous simulations, appears to be about as flat in thissimulation as in the previous ones. It turns out that the cursor-positioncurve in Figure 5 is precisely identical to the curve for the cooperatingsystems in Figure 4, and also for the single system in Figure 3. The twocontrol systems in Figure 5, although engaged in conflict, have jointlycontrolled the environmental variable. Thus, the paradoxical conclusion to bedrawn from Figure 5 is that control systems are capable of simultaneousconflict and cooperation. Despite the divergence in their outputs, theenvironmental variable they perceive is stabilized just as effectively by theirconflictive interaction as by their actions when perfectly aligned or even bythe actions of a single stronger system.

This central finding (confirmed in many other simulations not shown here)--thattwo or more control systems can conflict and cooperate at the same time--mayseem counter-intuitive. Common sense defines cooperation and conflict asopposites, assuming that one precludes the other. And the battles in sociologybetween "order" theories and "conflict" theories have done little to shake ourcommon-sense notion that cooperation and conflict are incompatible. Of course,Simmel argued nearly one hundred years ago that "conflict contains somethingpositive"(1955, p. 14), that it acts as an "integrative force in the group"(p.17), and must be included among the "elementary forms of sociation [sic]"(p. 20). But neither Simmel nor the Functionalists who later stressed "thefunctions of social conflict" (Coser 1964) made clear that conflict andcooperation can both flow from the same interactional process.

Once the connection between conflict and cooperation in a control-system worldis pointed out, one can easily find instances in human situations. Peopleoften engage in conflict and cooperation simultaneously. Any number of groups,such as political parties, "dysfunctional" families, or academic departments,are fraught with internal conflicts yet carry on their activities from year toyear with little change. Such high-tension arrangements may satisfy none ofthe participants but still can provide enough stability for everyone to carryon. In short, conflictive cooperation, even shading into cooperative conflict,seems nearly ubiquitous in human social life.[16]

Of course, conflicts cannot escalate unchecked if social stability is to bemaintained, and there are obvious limits to how much effort any individual candevote to any particular conflict. Thus, to get more realistic simulations ofconflictive cooperation, one needs to set limits on the amount of output eachsystem can produce. As it turns out, this more realistic design for asimulation also leads to some interesting results. Figure 6 shows a simulationusing the same two control systems as in Figures 4 and 5 and the samedisturbance pattern as in all the other simulations.[17] The key difference in this graph is that each system'soutput has been limited to a maximum of 100 units. Moreover, the referencevalues have been set farther apart, at +10.0 for the system with gain 300, and-15.0 for the system with gain 200. Because the gap in reference values iswide enough to show up on the graph, the reference lines have been plotted.

The wider gap between reference values leads to a much more rapid escalation ofthe conflict, and both systems soon hit their output limits and becomedeadlocked at 100 points of pull in either direction. This simulation of adeadlocked conflict provides another illustration of the subtleties of thecontrol-system model. As long as the two outputs are equally balanced againsteach other, the only force leading to any change in the environmental variableis the disturbance, and the environmental variable begins dutifully followingthe disturbance, until the disturbance pulls the variable outside of thedisputed region between the reference lines. Whenever that happens, the systemwhose reference line has been crossed can relax enough to move away from itsoutput limit and thus begin again to control. So the variable stays near thereference line for the system in control. That control lasts, however, onlyuntil the disturbance begins pulling the variable back toward the othersystem's reference line, at which point the first system once again runs intoits output limit and loses control.

Looking carefully at the Figure 6, one can see that the system pulling in theupward direction, aided by the disturbance which is pulling the same way, keepsthe cursor near its own reference line from about 50 to 175 on the horizontal(time) scale of the graph. From then until about 250, the two systems tradecontrol momentarily while the disturbance fluctuates near zero, until thedisturbance (by veering sharply in the negative direction) hands over controlof the variable to the system pulling downward. Finally, from about 400 untilalmost the end of the run, the disturbance moves back inside the "dead zone"(Powers 1973, p. 255) between the reference lines, and the variable is nolonger controlled by either system but simply follows the path of thedisturbance.

The dynamics of loss of control in deadlocked conflict, as revealed in Figure6, are more familiar than might be evident from this abstract presentation.They can be easily seen in a person's internal conflicts, where the discordantcontrol systems take the form of contradictory impulses. Wanting twoincompatible things at once and not being able to choose between them is acommon experience for most people, whether the things involved are as trivialas chocolate or vanilla, or as serious as a wrenching choice between family andwork. Vacillation and vulnerability to shifting influences usually accompanythe experience of a deadlocked internal conflict. Sometimes, too, whencircumstances seemingly conspire to foreclose the decision, one's own impulsesreassert the conflict: If there's nothing but vanilla left, what I reallywanted was chocolate.

Similar dynamics apply to all-out conflicts between people. When both partiesare stretched to the limit and neither can gain a decisive advantage, thesituation often drifts out of everyone's control, as temporary disturbancesbring the disputed variable frustratingly closer first to one party's goal andthen the other's, with neither party able to control it. The simulation inFigure 6 shows vividly how third parties can play a key role in this sort ofdeadlocked conflict (compare Bercovitch 1984). As long as the efforts of thetwo main combatants roughly balance, a third party need only expend a littleeffort to disturb the situation enough to give control to one or the other orelse deprive both of control by keeping the disputed variable somewhere in themiddle. However, when a third party's intervention is extreme enough, the twoerstwhile combatants may temporarily find themselves pulling in the samedirection, at least until the third party stops interfering, as when warringspouses close ranks against an outsider's gratuitous advice.

The simulation also shows that, from the point of view of a combatant, someonewith a reference point more extreme than one's own might prove an ideal ally ina conflict. A knee-jerk liberal or conservative, for example, who can becounted upon to react consistently and violently in his preferred direction nomatter what happens, may give the more moderate proponent of the same positionroom to maneuver and thus to keep the situation in relative control, becausethe extremist has immobilized the opposition by forcing him to his limits. Anadded bonus from the moderate's point of view occurs when the extremist allyhas also hit his own output limit and thus removes himself from any directcontrol of the outcome, allowing the moderate to play the strategic role of thethird party in a deadlocked conflict.

A final simulation can give us some insight into what happens whencontrol-system conflicts are resolved. The simulation in Figure 7 has the sameconditions as Figure 6, except that after 50 time units (out of 500) the twoconflicting reference values are brought quickly back together into precisealignment at a reference value of zero, as if the parties to an intense disputewere suddenly to patch up their differences and agree on a compromise position.Notice that the conflicting outputs do not come back together quickly.Instead, the output curves from the two systems start to converge quite slowlyafter the agreement on a common reference value, so that by the end of another450 time units the two curves are still about 65 distance units apart.[18] The sharp divergence produced by theconflict, which initially pushed both systems to their limits of output in lessthan 50 time units, is still not fully erased after almost ten times as long.Thus, in a control-system world a complete resolution of conflicts takes muchlonger than an initial divergence.

Experts in conflict resolution have frequently observed how human conflictsalso intensify more quickly than they relax. As Fisher (1990, p. 26) notes, insummarizing Sherif's classic Robbers Cave Experiment (Sherif et al. 1961),"[C]onflict escalation appears much easier to induce than is conflictresolution." With control systems, the reason for the slower pace ofconvergence in output compared to divergence is because control systems onlychange their output when they experience perceptual errors. In Figure 7, whenthe compromise in references values is reached, it reduces the perceptual errorfor both systems to nearly zero, and with perceptual error at zero the logic ofcontrol means no change in output. If a control system suddenlystopped pulling when its perceptual error was zero, its error would probablyincrease abruptly because its output would no longer balance the disturbances(including the other party's opposite pull). A careful look at Figure 7 showsus that most of the convergence in output in this simulation takes place attimes when a large value of the disturbance pushes one system or the other toits output limit (as happens between times 250 and 300 on the graph). Theother system then takes over as the only effective controller and moves itsoutput nearer to the center, thus erasing part of the gap in outputs.[19]

Because control systems can be producing widely different outputs even whenagreeing on a reference standard, an outside observer may easily overestimatethe extent of underlying conflict among groups of people who appear to bepulling in contradictory directions. The outside observer can see only theactions of the individuals involved, not their reference values, and mysimulations show that small disagreements in reference value typically lead towide disparities in output. Furthermore, once such disparities areestablished, even perfect harmonization of reference values does not produceany immediate reconciliation in output, and relaxation of an overt conflict cantake far longer than the initial divergence. Not infrequently, group memberswho seem to be completely at odds with each other will unite against anoutsider who threatens the group (see, for instance, Anderson 1978, pp.163-170). Their basic similarity in reference values is revealed when thedisturbance presented by an outsider with a sharply different reference valuegets them pulling together in the same direction again.

Lessons from Simulations of Conflict and Cooperation

As my examples in preceding paragraphs have suggested, conclusions drawn fromsimple control-system simulations about social conflict and cooperation canapply to interactions involving not just two or three people, but thousands ormillions, all seeking to control their imperfectly shared perceptions of, forinstance, the outcome of a political controversy. The most important findingfrom these simulations has been that interacting control systems need not holdthe same reference standards in order to achieve cooperative outcomes.Turner's (1994) assumption, that hidden collective entities must be identicalacross individuals in order for coordination to occur among them, is evidentlyincorrect for interacting control systems. As long as systems with conflictingreference values can stay inside their maximum limits of output, theirinteraction results in joint control of the contested environmental variable.As Powers puts it, conflicting control systems will stabilize the environmentalvariable around the "virtual reference level" (1973, p. 255) of the group as awhole, even when all the individuals within the group are experiencinguncorrected errors.

Consider what happens when large numbers of people all become involved incontrolling a single perception, say the outcome of a national politicaldecision. The participants in such a widespread cooperative-conflictiveinteraction can represent all shades of opinion, or in other words might holdalmost as many different reference values as there are participants. Peoplewill then pull in many different directions to adjust their perceived realityto fit their preferences. Because everyone is experiencing perceptual errors,except those few whose preferences approximately match the virtual referencevalue of the group as a whole, nearly everyone may end up more or lessfrustrated with the situation. As PCT points out, prolonged errors generatenegative emotions. Some participants may throw maximum effort into theinteraction, pulling as hard as they can, while others act as free riders,observing but adding nothing to the collective output, and still others becometotally apathetic, turning their attention elsewhere. Others may cope withperceptual error simply by adjusting their reference values to match the statusquo as they perceive it. The choice of action depends on participants'reference values, their energy and endurance, other disturbances affectingthem, and other perceptions they seek to control. If some participants withextreme views "max out" in their output, it will have little effect on thecollective outcome as long as those with more moderate views can maintain theircontrol by keeping their own output within reasonable bounds.

My simulations imply that out of this seeming chaos of collective control ofperception on a grand scale will emerge stability on the same scale. A virtualreference value will emerge from the interaction and decide the outcome. To anindividual participant, it might well seem like some invisible super-humancontrol system is imposing stability, since no action by that single individualwill have noticeable impact. The individual can pull, give up, or startpulling just as hard in the opposition direction without perceptibly budgingthe virtual reference point. But even if no particular individual'scontribution counts for much, the collective outcome is wholly determined bythe joint efforts of the participants.[20]

While the average participant in a large-scale collective-control episode likethe one above may be powerless to affect the outcome, some few may be able toexercise significant leverage in the conflict. If an individual or a smallcoordinated group of individuals possesses sufficient influence--by virtue ofcharisma, organizational authority, celebrity, or the like--over the positionstaken by an appreciable faction of other participating individuals, so that theinfluential person or persons by announcing a change in position speak not onlyfor themselves but for a large group of others who reset their own referencevalues accordingly (see McClelland 1994), the center of gravity of the overallconflict can shift substantially. Mouzelis refers to such influentialindividuals as "macro-actors" (1991, p.107), which he further designates as"collective actors" if their influence derives from representation of others,and "mega-actors" if their influence is based on ownership of resources.[21] The factions of participants for which suchleaders speak need not be perfectly united among themselves for the leaders to"exercise power," as long as enough people within the faction are willing tofollow the lead, so that the virtual reference point for that subgroup shiftsperceptibly. Of course, a leader who cannot influence followers to changepositions has stopped being a leader.

In general, social structures of all kinds are built and maintained by thissame process of collective control. People come together seeking to controltheir individual perceptions of a common variable. Their conflictivecooperation results in environmental control of the variable at a virtualreference point which corresponds approximately to the centroid of theirpreferences.[22] To an individual it may feelas if some external force has determined the outcome, and it may even seem asif the control which emerges is the responsibility of some virtual collectiveactor, personified, perhaps, as "the group," "the family," "the corporation,""society," the "great leader," or even "God." Although illusory, this virtualcollective actor may appear to have human and even super-human qualities, inthat the outcome of the collective control is indistinguishable from thestability that could have been produced by a single, powerful control system.No wonder, then, that sociologists in their explanations of social structurehave invoked hidden collective entities! Sociologists, too, have experiencedthe same illusion.

Characteristics of Virtual Social Actors.--Sociologists who hope todisabuse themselves of the illusion of hidden collective entities must seek tounderstand the virtual control systems which emerge from collective control ofperceptions. How do these virtual actors compare to the humans whosecollective actions produce the illusion? First, as I have just noted, thevirtual actors, like super-high-gain control systems, produce an environmentalstability far greater than any single individual could achieve. Even when theindividuals involved participate inefficiently, if enough people are involvedthe outcome will be invulnerable to almost any disturbance the averageindividual could produce, so that the greater the number of individualsinvolved in the action of collective control, the less likely it is that asingle individual can make a dent in the collective outcome.

The stability produced by any widespread collective control is also likely tolast longer than the results of a single individual's efforts. Thus, thevirtual actor will seem not only to be stronger than the individual, butusually slower as well. If only a few individuals are involved, an episode ofcollective control may be over in minutes, but collective control involvinglarge numbers of people generally takes longer to get organized and longer todie out. Individuals participating in the effort of collective control maycome and go, but the attention span of the virtual actor will appear to lastfrom the time the first two participants get together till the time the lasttwo quit. Some efforts of collective control--for example, the maintenance ofa building or public monument--may even continue for centuries, as generationsof caretakers succeed each other.

Although collective control of a perceived variable can outlast the lifetimesof the currently controlling participants, the outcome of the control need notbe static. Thus, virtual social actors can appear to change their minds,though typically more slowly than individuals do. Consider the slow drift inthe currently fashionable vocabulary of a language--another example, as I willargue below, of a collectively controlled perception. The center of gravity ofa language gradually shifts as old generations of language users die off andnew ones are born, and also as new words are coined and become more widelyaccepted. Thus, the virtual reference value of collective control will changeslowly as participants with their own reference values exit or enter thecontrol effort, and change rapidly on the less frequent occasions whensubstantial numbers of participants change their own reference values in thesame direction at the same time.

If enough participants in an effort of collective control can coordinate theirchanges in reference values, thus generating the social power of a newalignment (see McClelland 1994), the resulting lurch in the collectivelycontrolled variable can be revolutionary. A dramatic example occurred acrossEastern Europe in 1989-92, as Communist governments toppled almost overnight,not because of any armed conflict, but because millions of inhabitants of thecountries agreed with new leaders that a change had taken place. Not everyoneneeded to come to that agreement, but when enough did, the outcome was nolonger in doubt.

So far, I've argued that the virtual actors emerging from the collectivecontrol of perceptions tend to produce more environmental stability than anindividual human actor could, both in their greater resistance to disturbancesand in their generally slower pace of change. In effect, virtual actors tendto be conservative. A final important comparison between virtual social actorsand individual human concerns their degree of intelligence. Virtual actors maybe stronger and more durable than humans, but they are generally not as smart.Individual humans can rapidly shift from controlling one perception tocontrolling another, as circumstances around them change, and theirresponsiveness to changing opportunities shows their intelligence. Virtualsocial actors tend to be stuck singing in one key, and in a monotone at that.Their very resistance to change means they can't easily be swayed by changingcircumstances, and the more participants involved in the collective controleffort, the harder it becomes to coordinate any general shift of referencevalues. Moreover, the kinds of perceptions that people can collectivelycontrol are limited. To be shared among a large throng of disparate humans,the perceptions must either be relatively crude or so highly abstract as toleave much room for individual interpretation. Precision and nuance can bestbe achieved by individuals, not groups (symphony orchestras or precision drillteams to the contrary, and even there, the actions performed in unison tend tobe simplified and predictable in comparison to an individual virtuoso'sperformance). Macro-actors, influential individuals who succeed in gettingothers to conform to their reference standards, can speed up the changes incollectively controlled variables, but even those leaders often seem frustratedat the unwieldiness of the human machine they attempt to control.[23]

In sum, my argument here is that collective control of perceptions can fosteran illusion that an imagined hidden actor is responsible for the resultingcontrol. The apparent behavior of such virtual social actors makes them seemdifferent from ordinary humans--both more powerful and less intelligent thanthe individual. As my examples have indicated, collective control ofperceptions occurs in a wide variety of forms throughout the social world.Before providing an inventory of those forms, however, I need to say somethingabout human interactions that do not exhibit collective control, as wellas something more about PCT and orders of perception.

Classifying Interactions.--A central finding from my simulations ofcontrol-system interactions is that conflict and cooperation can both occur inthe same interactional process. Although "pure" cooperation is theoreticallypossible, most episodes of collective control of perceptions will involve amixture of cooperation and conflict. The greater the difference between thereference values of the interactants, the greater the divergence in output, andthus the greater the conflict and the greater the likelihood that the conflictwill drive the interactants to the limits of their output, where cooperativecontrol is lost. One can classify interactions, then, by locating any giveninstance of collective control on a continuum from pure cooperation, wherereference values are identical, to maximal conflict, where reference valuesdiverge as much as possible.

What about, however, interactions in which the participants are not only usingdifferent reference values but seeking to control entirely different perceptualvariables? The simulations we have looked at assume the simplest possible kindof environment for interaction, one in which only a single perceptual variableis available for control, an environment which PCT theorists, following Powers(1989b), describe as having only one degree of freedom.[24] Real environments, however, nearly always offer manydegrees of freedom, and when two control systems are controlling completelydifferent variables in such a shared environment, they may act alongside eachother without conflict. I will refer to such interactions as mutualaccommodation.[25]

Consider, for an example of mutual accommodation, two people living together ina house with a lawn and garden. Suppose both want the exterior of the house topresent an attractive appearance, but one defines attractiveness entirely interms of fresh paint and clean windows, while the other sees attractiveness ascarefully trimmed lawns and weed-free flower beds. With enough money andeffort, both occupants can have a house which conforms to their own separatestandards of attractiveness. Thus, mutual accommodation can occur, providedneither attempts to control the variables controlled by the other.[26] If they disagree, however, on the color ofthe paint or the length of the grass, or how much of the available money andtime should be given to each type of maintenance, conflict will ensue.

The pure form of mutual accommodation, which occurs when interactants controldifferent perceptual variables in environments with two or more degrees offreedom, does not involve any collective control of perceptions: mutuallyaccommodating control systems neither cooperate nor conflict. Thus, if we wantto classify interactions, we can take mutual accommodation as the polar type ona dimension orthogonal to the conflict-control continuum, defined by the degreeof similarity in the perceptual variables controlled by interactants. In thisdimension, interactions may range from control of identical variables, whichresults in some kind of conflictive cooperation; through control of correlatedvariables, which produces low-level conflict, which I will callinterference; to control of completely dissimilar variables, whichallows accommodation. Figure 8 shows a theoretical triangle for classifyinginteractions in these two dimensions--similarity in reference values andsimilarity in perceptions controlled--resulting in three polar types--purecooperation, maximal conflict, and mutual accommodation. Interactions fallinganywhere in the middle of this space exhibit some combination of conflictivecooperation and mutual interference. The dimension of similarity in referencevalues is irrelevant at the accommodation pole because the perceptions beingcontrolled by accommodating systems are completely different, and thus it makesno sense to compare their reference values.

Accommodation is easiest to achieve in a rich environment, with many degrees offreedom, so that control by one interactant is unlikely to interfere withcontrol of another's perceptual variables. In a restricted environment, onewith few degrees of freedom, interference is harder to avoid, and the actionsof one control system may even make control completely impossible for another.When two prisoners are locked in a tiny prison cell, even the minor tics of onecan drive the other crazy. Moreover, certain kinds of actions, especiallyforceful or rapid ones, tend to use up environmental degrees of freedom,leaving fewer for anyone else sharing the environment. For example, a bridgeof a given width can accommodate many more walkers than bicyclists in a giventime, even though the bicyclists move across it more quickly, and likewise manymore bicyclists than motorists. The greater the speed and the size of theconveyance, the more the available degrees of freedom are exhausted. Somekinds of control, then, are quite expensive in terms of degrees of freedom.

A time-honored strategy for individuals caught in a conflict is to take one'sball and go home, and thus to deprive the other interactant of degrees offreedom for action. I will call it obstruction, when one interactant'sbehavior usurps all of the degrees of freedom available to another. Obstruction can be deliberate, with extremes of obstruction occurring inphysical violence, terrorism, and war. Obstruction may also be unintentional.One can unconsciously take up more than one's share of physical space, asfeminists have complained that men often do in public spaces. Our rules ofpolite interaction help cue us on how to avoid obstructing the perceptionscontrolled by other people: Keep your elbows in and your knees together andsmile.

Obstruction, like accommodation, cooperation, and conflict, defines a polarpossibility for control-system interactions. Figure 9 diagrams thistheoretical relation in a manner similar to Figure 8, except that thehorizontal dimension is different--number of degrees of freedom available forcontrolling perceptions, rather than similarity in reference values. Thediagram is again a triangle, because the degrees-of-freedom dimension isirrelevant to interactions of cooperation and conflict. It only takes onedegree of freedom, and thus a single perceptual variable, to set the stage forconflict or cooperation or both.

Putting Figures 8 and 9 together, I offer in Figure 10 a tetrahedron ofinteractional possibilities. Most interactions will result in someintermediate combination of conflictive-cooperative-obstructive accommodation,rather than any of the pure types at the vertices of the tetrahedron.Encounters between people may also involve more than one type of interaction atonce, as when a party to a conflict seeks to gain advantage by obstructing theopponent's control of another perceptual variable, perhaps by physical violenceor threat of violence --"Give me that or I'll hit you."[27] And further elaborations may be introduced bydistinguishing between intentional and unintentional interactions of eachtype.

Obviously, the tetrahedron of interactional possibilities in Figure 10 opens upnumerous opportunities for our analysis of social encounters. Rather thandescribing the ramifications of the typology in any detail, let me give asingle illustration of its potential usefulness. Relations of status anddeference fit into the tetrahedron near the "accommodation" vertex, though theaccommodation in this case is not mutual but one-sided. When one person hashigher status than another, the lower-status person has an obligation to deferto the other. In PCT terms, this means that whenever the two people's controlcircuits intersect in their common environment, the lower-status individual hasto give up his or her perceptual control at least temporarily to allow thehigher-status individual to continue without interference. In other words, thelower-status individual must try not to use up any degrees of freedom in theenvironment which the higher-status individual might possibly need. Symbolicgestures of deference, then, are to cower in a corner, to fall prostrate on thefloor, to bow one's head and avert one's eyes, or to raise one's hands in the"don't shoot" sign. By such gestures, the lower-status person publiclydisables his or her own means of perceptual control, so as to accommodate theother person by leaving the field of action as wide open as possible. Bycontrast, a handshake symbolizes equality in status, because it symmetricallydisables both parties from their other activities and so signifies mutualdeference.[28]

Layers of Perception.--In developing a typology of fundamental kinds ofinteraction, including cooperation, conflict, accommodation, and obstruction, Ihave sought to distinguish between interactions that involve collective controlof perception and those that don't. I would not want to suggest that everyinteraction is an instance of collective control. My central purpose in thisarticle, however, is to demonstrate the large extent to which our social lifedoes depend on collective control of perception and to document the widerange of perceptions we collectively control. Before I can present mypreliminary inventory of collectively controlled perceptions, one lastelaboration of the theory is necessary. Readers need to understand the PCTmodel in a little more detail, particularly with regard to "orders ofperception" (Powers 1990), in order to appreciate the enormous variety ofperceptions that humans are capable of controlling.

As I suggested above, in the PCT model negative feedback loops (Figure 1) serveas the main building blocks in the organization of the central nervous system.Powers (1973) offers a model of neural organization in which thesenegative-feedback circuits are arranged hierarchically, so that our morecomplex perceptions and actions are constructed from combinations of simplerperceptions and actions. Powers further proposes (1990) that this hierarchymust contain at least eleven distinct "orders" of perception (see alsoMcClelland 1994) from the most basic and concrete order, which he designates as"intensity" of sensory stimulation, to the most complex and abstract order,"system concepts," including our perceptions of individual identity and theidentity of groups, as well as our most general perceptions of widespread humanendeavors like science or art. Powers further proposes that the controlledperceptual signals from circuits of any given order become the input signalsfor circuits at the next more complex order of perception, while the outputsignals become reference signals for circuits at the next simpler order ofperception. All complex perceptual circuits, then, are connected to theenvironment through one or more layers of simpler control circuits, ultimatelyimpinging on the environment through the actions of the most basic circuitsinvolved in the control of muscle tensions, with the feedback loop then passingthrough the environment by means of the organism's effects on physicalvariables and returning via the operations of basic control systems in thesensory organs. Thus, every control loop, whatever the perceptual order,passes through the environment, and causal forces in the environment form anintegral part of every control loop.

Plooij (1984, p. 13) has suggested that the organization of this perceptualhierarchy within the person can be thought of as a series of concentricspheres, like the layers of an onion.[29] Thesimplest orders of perception occur in the outer layers, by means of neuralcircuits located in parts of the body which communicate directly with thephysical environment--the sensory organs and muscles--as well as in closelyconnected circuits in the spinal cord and brain. More complex perceptionsoccur in inner perceptual layers, presumably within the cortex of the brain.For my purposes here, we need only distinguish three broad perceptual layers,each combining two or more of the eleven orders of perception. Layer Icomprises the outermost or concretely physical orders of perception (thesix simplest perceptual orders proposed by Powers--intensities, sensations,configurations, transitions, events, and relationships). Layer I thus providesall the perceptions which connect us to our physical environment, defining itssights, sounds, smells, tastes and tactile qualities, and also the kinestheticperceptions which enable us to move about and function in physical space.Language plays no part in perceptions within this layer. These physicallyconcrete perceptions are wordless, consisting entirely of appearances andimages devoid of any linguistic concepts. Of course, language has a basis inLayer I perceptions, since concretely physical perceptions are a necessary partof the process of speaking or reading or listening to spoken language.However, we make use of a more complex layer of perceptions to turn thephysical perceptions underlying language into language per se.

The next concentric sphere of perceptions, Layer II, comprises abstractlyanalytical perceptions (see Figure 11). Layer II is where language andabstract reasoning appear. This middle layer includes, from Powers'seleven-order scheme, perceptions of the category, sequence, and program orders.People use words to name their category-order perceptions, and producesentences by employing sequence-order perceptions. When they combine verbalpropositions to make rational arguments, they are using program-orderperceptions, which involve logical decision-making and branching from onesequence to another.[30] Our consciousthoughts generally find expression in terms of middle-layer perceptions likethese, as we talk to ourselves about what is happening or what we intend to do.Of course, not all of our perceptions at Layer II involve language; we can forma working category without giving it a name, or follow a sequence of stepswithout verbalizing them. But more often, Layer II perceptions occur aslanguage.

If Layer I perceptions are subverbal, while Layer II are characteristicallyverbal and rational, Layer III perceptions, which I will call abstractlyintegrative perceptions, might be considered super-verbal. Here we reachfor the ineffable, things that words cannot express. Powers distinguishes twoorders of perception in this innermost layer: principles and system concepts.Principles include perceptions of moral qualities such as love, honesty,ambition, or greed, as well as our pragmatic standards for selecting programsof action. Powers's most complex order of perception, system concepts,combines perceptions of principles to form the perceived identities of people,groups, and their common enterprises. Just as Layer III perceptions are toocomplex to express fully in words, neither can we encapsulate them in neatlylogical packages. Perceptions in this core layer transcend the narrowstrictures of rationality, and sociologists have often sought to describe themby talking of group solidarity or sentiments of social cohesion, perceptionsthat form, as Collins (1992) puts it, the "nonrational foundations" of sociallife. Thus, Layer III perceptions might be considered extra-rational as wellas super-verbal.

With the understanding that there are three main layers of perception and thatthe perceptions characteristic of verbal and rational thought occur at LayerII, we can begin to provide a PCT interpretation of two of the essentialconcerns of many sociologists: meaning and symbolism. We ordinarily don'tinquire into the meaning of perceptions occurring in the outermost, concretelyphysical layer. One's perception of a particular cat or a particular chairappears as a concrete reality, not something to be further analyzed. At LayerII of the perceptual hierarchy, however, questions of meaning inevitably arise.When I say the word, "cat," what does it mean to you? The meaning you give tothe word "cat" will most probably take the form a mental image of a cat, apicture in your imagination of a typical cat. The picture, for instance, whichcomes to mind for me is a nondescript gray tabby with mottled black stripes onher back. Maybe yours is similar. Whatever the meaning one gives to a word,and regardless of whether we agree precisely in the meaning we give it, ourindividual construction of that meaning involves an act of imagination based onour own experiences and memories.

In PCT terms, we form mental images by controlling our perceptions in"imagination mode" (Powers 1973, pp. 222-24), in which the feedback loop of thecontrol system is short circuited at some outer perceptual layer. The outputof a middle-layer control system in effect requests a combination ofperceptions from the outer-layer control systems connected to it by specifyingthe reference signals they should use. When our brains operate in imaginationmode, control systems at some peripheral point in the neural hierarchy stopproducing output signals of their own and simply send back as perceptualsignals the reference signals they receive. Thus, the feedback loop for theinner circuit is completed without actually passing through the environment, sothat no physically observable behavior occurs. Controlling a perception inimagination mode is in one sense easier than controlling the same perceptionactively, because the imagined perception is immune to the disturbances alwaysarising in physical environments. But imaginary perceptions tend only to bepale copies of perceptions based on physical reality, because they most oftenoccur as visual images alone or else as disembodied speech, lacking themultiple perceptual dimensions of physical occurrences--smells. tastes,textures, muscular effort, and real-time extensiveness.

Construing the meaning, then, of a word or sentence involves an imaginative actof finding a plausible set of physically concrete perceptions to enable one tocontrol one's middle-layer perception of a word or sentence. In construingmeaning we inevitably act independently, although meaning, of course, issocial. If, for example, nothing in one's own memories is associated with theword "smaragd," the word calls forth no mental image and hence seemsmeaningless. Or else, one might conjure up some fantastic image--maybe it'slike a gigantic dodo bird--but still have no grasp of the word's culturallyaccepted meaning.[31] In trying to establishthe meanings of words or sentences in the context of an ongoing interaction weoften resort to the error-correction process which Ethnomethodologists havelabeled "waiting for clarification" (see Garfinkel 1977, ch. 3). One callsforth a tentative mental image to allow the interaction to proceed, thencorrects or refines the image as more clues to the intended meaning becomeavailable. Thus, one's middle-layer perceptions are brought into focus throughconnections to physically concrete perceptions or perceptual memories.

"Meaning", however, has another more important meaning in socialtheory--symbolic meaning. When sociologists talk of the meaning of an act or acommunication, they are usually referring to its symbolic meaning, rather thanto the denotative meaning which links our Layer II perception of the word withour Layer I perceptions of concretely physical objects and actions. In PCTterms, we construct symbolic meanings for words or objects by imaginingconnections with inner-layer perceptions--ultimately with abstractlyintegrative perceptions like principles or system concepts. In PCT terms, then,a symbol is a perception that points inward in the layers of the perceptualhierarchy, evoking perceptions of greater complexity or scope. For instance,the sounds of a spoken word or the marks of letters on a page are symbolic tous of the word--our Layer I perceptions of physical sounds and shapes symbolizea Layer II perception of the word itself.[32]Similarly, a word can in turn symbolize our Layer III perceptions of a group ofpeople or an abstract virtue. Perceptions of objects--like a flag--or ofrituals--like saluting the flag--serve as symbols when we connect such physicalperceptions with our more abstract perceptions of the social group.

Symbolic meaning, of course, tends to be metaphoric. We use relativelyconcrete perceptions to suggest the essential qualities of more abstract ones.More precisely, symbolic meaning involves metonymy, where the part comes tostand for the whole. In consequence, a symbol inevitably oversimplifies theinner-layer perception it is intended to evoke. The country is far more thanthe flag, for instance, and respect for the flag does not equate to respect forthe country in any but a symbolic sense. We are always condemned someconfusion by our inescapable reliance upon the oversimplifications of symboliclanguage for talking about our most profound perceptions--those that define ourcore identities and our deeply held values. But we have no choice. We havenothing but words to use when we talk to each other. Even supplemented byphysical gestures or objects used symbolically, the words are bound to beinadequate, because the perceptions we seek to express are "more encompassing"(Runkel 1995) than any number of words could possibly capture.

Communication is easier when we stop talking symbolically and stay in the realmof the Layer II perceptions needed for rational exchange, though even there wecan miscommunicate if our experiences of the physical world are so differentthat we attach different denotative meanings to words. The Layer IIIperceptions that make us uniquely human--our perceptions of values andidentities--are the hardest to convey. Perhaps this is why sociologists havefound it such an interesting challenge to decipher the symbolic meanings ofsocial acts.

With this PCT interpretation of the processes behind meaning and symbolism, allthe pieces are now in place for providing an inventory of perceptions wecollectively control and describing how social structures and cultures can beunderstood as collective perceptions. To summarize a complex argument, I haveasserted that the central feature of human goal-directed behavior is thecontrol of perception rather than action, and that the PCT model describeshuman goal-directed behavior with considerable accuracy. I have shown how PCTsimulations of collective action can reproduce familiar patterns of cooperationand conflict, and have argued that these simulations lay bare the importantfact that collective control can emerge from conflictive interactions.

I have argued further that not all interactions need involve collectivecontrol, with its simultaneous cooperation and conflict; instead, some may tendtoward mutual accommodation while others result in obstruction of one party'scontrol. But for actions which do involve collective control, the control hasmuch the same impact on the environment as would the actions of a virtualsocial actor of super-human power and sub-human slowness. This observableimpact of collective control, I argue, can account for the common illusion thatmysteriously hidden collective entities must be involved in collective action.Finally, I have argued that we can collectively control perceptions occuring atseveral different perceptual layers, including concretely physical, abstractlyanalytical, and abstractly integrative perceptions, and that we create our mostsignificant symbols by using perceptions from the outer and middle layers tocommunicate about our core of abstractly integrative perceptions--those whichwe as social beings find most meaningful but most elusive.

SOCIAL STRUCTURE AND CULTURE AS COLLECTIVE CONTROL

In this final section of my argument, I put forward a proposition of greatersweep than could possibly be demonstrated in a single article, even an articleas long as this one! Still, I hope to suggest the proposition's plausibilityand to gain the reader's appreciation of the challenge it offers to sociologyas a discipline. My contention is that all the behavioral and socialregularities of interest to sociologists have emerged from the collectivecontrol of perceptions, and thus that the theory of collective control can, inprinciple, illuminate the subject matter studied by every branch ofsociology.[33] In order to establish theplausibility of this sweeping assertion, I offer here a basic inventory ofcollectively controlled perceptions, dividing them into four broad perceptualdomains, and indicating the many kinds of social phenomena each domainincludes. I will refer to the four perceptual domains as social activities,social designations, cultural contingencies, and culturaldiscourses.

Social Activities

The first and most obvious type of collective control of perceptions occurswhen people coordinate their activities. In fact, a great many socialactivities--for example, engaging in conversation, holding a meeting, playingcompetitive games, performing most rituals, engaging in sexual intercourse, oreven having a fist fight--are simply impossible to accomplish on one's own.McPhail and Tucker (1990; McPhail, 1991, 1994) have described from a PCTperspective how individuals gathered in a common location can come to aligntheir perceptions and thus engage in collective behaviors. They argue that twoor more individuals may independently set similar reference standards inresponse to the situation, or they may interdependently agree on similarreference standards, or they may both adopt reference standards suggested tothem by a third party (often someone functioning as a leader or organizer). Assoon as two or more people have begun controlling together for similarreference standards--and, as I have shown above, the standards they use neednot be identical--collective control of their shared activity occurs. As Ihave also shown, unless the reference standards of the participants arepractically identical, collective control inevitably involves some conflict aswell as cooperation.

Perceptions that people can control together may occur in any layer of theperceptual hierarchy--from the most concretely physical orders of perception tothe most abstractly integrative. In fact, many social activities involvecollective control in all three of the main perceptual layers simultaneously.Consider a pair of individuals jointly controlling a Layer II perception--theperceptual sequence, hug-kiss-caress. While the participants are no doubtfocusing attention on the sequence itself, at any given moment they will alsobe attempting to control many other perceptions that are occurring, one mightsay, in the background of their consciousness. In Layer I, many kinds ofperceptual control are taking place. Is the pressure of our mutual touch justright--not too intense, not too delicate? Are body parts suitably aligned?Has the time come yet to move on to the next event in the sequence? While suchperceptions must be controlled jointly because both partners' movements make adifference, each individual may also be controlling purely personalperceptions: Does this sequence feel right to me in all the various ways thatI have come to expect a hug-kiss-caress to feel?

In Layer II, the layer of rational thought, participants may be controllingtheir perceptions of the hug-kiss-caress sequence in the service of some moreelaborate plan of action, perhaps as the opening move in an evening ofseduction, or as part of a cheer-up campaign to dispel a bad mood. In LayerIII, the more emotionally integrative layer, participants may be regarding theaction sequence as a practical demonstration of some virtue or vice--empathyfor grief, for instance, or family loyalty, or sexual licentiousness. At a yetmore complex perceptual order, the action may be taken to symbolize somelonger-term social project--perhaps an emblem of marital constancy, or, giventhe "wrong" partner, the disruption of a marriage. In the midst of all thisjoint activity, either partner might be devoting private thoughts tofantasizing how it might feel to hug-kiss-caress with a movie star, or toconstructing an imaginary conversation with his or her mother. Participantscan control manifold perceptions in all of these layers at the same time, withthe outer-layer perceptions serving as links in the control loop for the focalactivity, while perception of that same activity becomes part of control loopsfor more complex inner-layer perceptions.[34]

A serious misalignment of the participants' reference standards for any of themany perceptions under joint control will eventually lead to conflict withinthe on-going activity. In general, the more peripheral the perceptual order ofthe misalignment, the more quickly conflicts will become evident, becauseinner-layer perceptions must be constructed from the outer-layer perceptions,and thus take longer to unfold. Conflicts in outer perceptual layers also tendto be easier to fix. "Take it easy, you're squeezing too hard!" has a simplerremedy than, "You don't really love me, do you?"

In principle, one could analyze virtually any kind of structured socialactivity by looking at the participants' collective control of a hierarchy ofperceptions. Future researchers should have no trouble finding collectivelycontrolled activities in every institutional sphere of social life. Politicalcontests, production plans, marketing campaigns, military exercises, religiousrituals, school lessons, and entertainment extravaganzas all quite obviouslydepend on the coordination of collectively controlled perceptions.[35]

Researchers looking for collectively controlled activities need not confinethemselves to micro situations, as my examples of intimate physical contact orfamily routines might have suggested. The same principles apply to macrosituations and the doings of "macro-actors," those people whose actions affectlarge numbers of others (Mouzelis 1991). From a PCT perspective, such peopleembody social power because others look to them for cues in setting their ownreference standards for action (see McClelland 1994). The collectivelycontrolled activities of key groups--like legislators, or members of the board,or teams of negotiators--often determine the activities other people must thenpursue. When the general and his staff say, "March," the regiments march.When the President declares a day of mourning, flags across the country fly athalf staff. Even if an interaction produces great chaos or conflict, like awar or revolution, its outcome will hinge on the extent to which participatinggroups can collectively control the perceptions necessary to produce concertedaction. In sum, every variety of coordinated social action must involve thecollective control of social activities.

Social Designations

In a second domain of collective control, social designations, we shareperceptions about people, places, and things. In order to coordinate theirsocial activities, people must be able to align not just their perceptions ofwhat they are doing but also their perceptions of the things in their sharedenvironment that serve as the objects and tools of their actions. To play aball game together, for example, players need to focus their attention jointlyon a particular ball and a particular playing field. At the same time,co-participants in an activity must coordinate their perceptions of each otherand of the groups or connections they form. "Our team will be Jack, Sandy,Jane, and Peter." All the collectively controlled perceptions which point to aportion of the common environment as an object of shared attention, I will callsocial designations.

Perceptions of groups and group memberships are among the social designationsof greatest interest to sociologists. That such perceptions must be collectiveis obvious. Belonging to a clique, for example, can only be a sharedperception. To join a clique, one must be accepted by the other members as amember; the sponsorship of just one member of the clique will not be sufficientto guarantee admittance a new person if other members fail to agree that thenew person belongs. More formal groups may coordinate their collectiveperceptions of membership and forestall conflicts about acceptance by staginginitiation rituals which symbolize the new member's inclusion in the group. Insuch a case, the group's collectively controlled perceptions of a socialactivity--the initiation ritual--serve as a link in their feedback loops for aninner-layer collectively controlled perception of social designation--groupmembership.

Our perceptions of the social identities of other persons and the intimateconnections among them are also collectively controlled. Very early in life, ababy learns to recognize its own mother or other principle caretaker. When, inthe following year or two, the young child can say, "That's my mommy," and getothers to agree, the child's perception has taken on a social dimension bymerging with the collective control of that mother-child relationship, asrelatives and friends jointly perceive it. The closeness of any relationshipdepends on collective control by the participants of their perceptions ofbelonging together. In a two-person relationship, neither individual alone candetermine the degree of mutual closeness. By doing the social dance of"interaction rituals," to use Collins's (1981) term--actions which also markthe relationship for others to see--both individuals help to control the sharedperceptions which constitute the relationship. Other interested observersoften seek to control these perceptions, as well, and the focal individuals donot necessarily have full control of their own relationship. From Romeo andJuliet onward, romantic couples have tried fruitlessly to prevent relatives andfriends from interfering.

In addition to perceptions of connection with other people, each individualinevitably develops a perception of his or her own physical and social self.Just as we recognize the concrete particularity of other people's bodies,behaviors, and personalities, we come to recognize our own physical being andpsychological qualities. The "self" is a perception we control--the way itfeels to be me. In the course of our everyday activities, we often keep trackof our own behavior and feelings and act to reduce the discrepancies betweenour own perceived behavior and our image of ourselves. For instance, a personmight say, "I'm feeling so depressed, I'm just not myself today," or, "I don'tknow what's got into me--I'm feeling great and can't imagine why." As SymbolicInteractionists beginning with G. H. Mead (1934) have emphasized, suchperceptions of self, of the kind of person one is, are gained first in theprocess of social interaction with parents, teachers, and friends.Furthermore, continued interactions with other people are necessary for us tomaintain a secure sense of self.[36] In mostconversations, for instance, each person monitors the other person's behaviorwith an eye to perceiving the "looking-glass self" (Cooley 1902, p. 152)reflected in the other person's reactions to his or her own words and actions.In short, our perceptions of self, like our perceptions of friends or otherclose associates, are collectively controlled by the intimate social groups towhich we belong.

Perceptions of social designation help us to coordinate our activities withregard to places and physical objects. Just as we recognize certain people ashaving particular relationships to us, we recognize a special meaning incertain objects. That's not just a chair, that's my chair, my ownfavorite rocking chair with the pillow my grandmother embroidered. Anotherchair that looked almost like that one and worked just as well would not be atall the same. Such perceptions, collectively controlled, form the basis of thesocial connections we call "property." As Collins (1988, p. 404) has pointedout, property is a social relationship, not just the relationship between aperson and an object. Something is my property because my feedback loops canpass through that part of the environment without being "tread upon" (Runkel1990, p. 147) by other people's control efforts. Property is perceived asmine-not-yours. Such complex relationships between a person, an object in theenvironment, and other people must become the collective perception of everyonepresent, if what is mine is to remain mine, and yours, yours. Thus, to controlmy perception of my connection to an object as my own property, I must have thecooperation of others in my environment who recognize and honor this perceivedconnection between person and object.

As is true for our collective perceptions of social activities, we collectivelycontrol our perceptions of social designation in all three perceptual layers.In Layer I, for example, we control perceptions of an individual's looks, voicetones, and other identifying characteristics. For the most part, such physicalrecognition involves only individual perceptions rather than collectivecontrol, though while attending a high school reunion, for instance, one mayneed some help from other people in sorting out who is who. In Layer II, weachieve collective control by agreeing upon names for people, places, andthings. Proper names help us to distinguish the individuality andparticularity of the social objects named. "That man making a pass at Linda isLarry Jones, not Harry Jones, though they've both gotten fat and bald." Sharedmemories of people's habits, or stories we tell of the notable things they havedone, may also enter into our social designations of people in the middle layerof perception. In Layer III, we discern their character traits and socialidentities--complex and enduring perceptions that our names for people andstories about them can only serve to symbolize. Perceptions of socialrelationships and groups, as well as of particular places and objects, maylikewise occur in all of these perceptual layers. We control perceptionsranging from the physical closeness among people in a group, to the types ofrelationships among them, to the qualities of loyalty or enmity theirinteractions symbolize.

Although my examples of social designations have been taken for simplicity frommicro situations, people can collectively control such perceptions in the macroarenas of life, as well. We collectively designate organizations and ethnicgroups, highways and monuments, countries and oceans, social movements andcelebrities. Many of the phenomena on which people confer names, qualities, oridentities are too vast or too widely distributed to be entirely under theperceptual control of any small group. However, when social designations arecollectively controlled by large numbers of people, certain individuals orsmall groups within the collectivity may have the effective power of settingand resetting reference standards for others to use. City councils have theprivilege of naming streets, for instance, and residents generally have nochoice but to use the names the city council gives. When collective control ofsocial designations or activities occurs on the macro scale, macro actorsnearly always play consequential roles.

I have presented so far two of four perceptual domains in which collectivecontrol takes place: social activities and social designations. One ofsociology's central theoretical concepts has been "social structure," and I amnow ready to redefine this key concept in PCT terms. Most sociologists seesocial structure as involving the on-going patterns of interaction betweenpeople, and social structure has often been equated with the social positions,networks, or groups emerging from these patterns of interaction. I have arguedalready that through collective control of perceptions of social activity andsocial designation we construct and maintain social patterns of interaction andgroup membership. I want to take the argument one step further by stipulatingthat the social structure of a society consists entirely of its members'collectively controlled perceptions in the domains of social activity andsocial designation. This may sound as if I am saying that social structure isall in people's heads, but the process of perceptual control produces acorresponding stability in the physical environment: the predictably recurringactions and interactions of people and groups, as well as the physicalstructures in the environment that people build and maintain, are all part ofsocial structure by this definition. Collective control makes all these thingshappen, and if the environmental stabilities were not produced, collectivecontrol of the activities would fail. Thus, social structure, by mydefinition, is at the same time a set of perceptions and environmentally"real," because our perceptual control loops pass through the physicalenvironment.

By defining social structure as collective control, I am including in thedefinition some phenomena that standard sociological definitions have left out.Property relations, for instance, as well as the other ways we collectivelyapportion and control objects in the physical environment, fall under theheading of social structure in my definition. This PCT definition thusincludes in the concept of social structure its material base--a clearadvantage, I might add, over the excessively abstracted definitionssociologists usually give.[37] An additionalreason for including the material base of social structure in its definition isto distinguish social structure from "culture"--another idea fundamental tosociology.

Like social structure, culture from this perspective consists entirely ofcollectively controlled perceptions. The defining difference betweenperceptions of social structure and perceptions of culture is in whether theperceptions refer to particular people, objects, and actions in the physicalenvironment. In collectively controlling our perceptions of socialstructure--that is, perceptions in the social-activity and social-designationrealms--we always have in mind particular people or groups of people, theiron-going actions, and the particular physical objects on which their actionsimpinge. In the language of Ethnomethodologists, perceptions of socialdesignation and social activity are always "indexical" (Garfinkel 1967). Weneed some degree of intersubjective agreement about our perceptions of thephysical world, or we simply can't share perceptions within the social-activityor social-designation domains. If you're talking about Fred and I'm talkingabout Freda, we can't gossip till we get things straight. If you're trying towaltz, and I'm trying to two-step, our dance is over before it starts. Socialstructure is rooted in intersubjective agreement about physically concreteperceptions.

Cultural perceptions, by contrast, have no more than a symbolic connection toany particular person or physical object. Our collectively controlledperceptions of culture are not particular but general--applicable to manydifferent people or situations--and we give them specific physical meaning onlywhen we use these perceptions as "tools" (Swidler 1986) to give form to ouron-going activities and relationships. Chief among such cultural perceptionsis language--a versatile symbolic tool, indeed, and applicable to nearly everysituation we can imagine. Culture, as I define it, comprises the last two ofthe four great domains of collective control. Language belongs to the domain Iwill label, cultural discourses, but first I want to describe the domainthat includes norms and rules: cultural contingencies.

Cultural Contingencies

Cultural contingencies are if-then relationships which "create links betweenactions and consequences"[38] in the socialenvironment, in much the same way that natural laws link actions toconsequences in the physical environment. In order to walk through a door onemust open it; to get into an armed camp one must give the password. The formeris an example of a physical contingency, and the latter, a culturalcontingency. In either case, one cannot reach the other side of the entrancewithout having followed the prescribed course of action. A contingency,whether physical or cultural, does not actually control anyone's behavior.Instead, contingencies specify the means by which a given goal can be obtained,or else the necessary outcomes of a particular action. In short, they tell us,"If you do this, that will happen."[39]

Cultural contingencies often take one of two main forms. Some culturalcontingencies govern exchanges of goods or services--if you give me X, I'llgive you Y. Other cultural contingencies define norms, which can beprescriptive--you must do X if you want to accomplish Y--orproscriptive--if you do Z, you'll be sorry. Although many norms grow out oftraditional usage, many other norms are imposed by the collective-controlefforts of members of a formal organization. In particular, governments are inthe business of defining and enforcing cultural contingencies. Similarly, thecultural contingencies of exchange are produced and maintained by economicorganizations like business enterprises or cooperatives (with the help andregulation of the government).

Contingencies of exchange include terms of barter, prices, and even moneyitself. Money, in fact, provides a good example of how cultural contingenciescan work. There are, of course, physical objects we call money: coins, bills,certificates and checks. And there are less tangible indicators of money:IOU's, credit cards, and electronic entries in a bank's computer files. But,as any economist will attest, money per se has only a symbolic connection toany of these tokens of exchange (see Caruthers and Babb 1996). Money is acollectively controlled perception: it is a promise to the holder that thetoken may be exchanged for some unspecified thing at some unspecified futuretime. The promise is good because of the collective perception of lots ofother people that such tokens are "legal tender." If the collective perceptionchanges, the promise evaporates and the token becomes merely a collector'sitem, like a stack of Confederate currency.

Laws, rules, regulations, customs, mores--all of which sociologists generallyregard as culture--are also examples of cultural contingencies by mydefinition. In each case, the contingency apparently exerts control onindividual behavior because individuals, aware of the contingency, use it toset the reference standards for their own behavior. Some of thesecontingencies, of course, are blatantly coercive. People follow the rulesbecause they're afraid to break them. The coercion works because of thecollectively controlled perception of other people (and especially of the macroactors involved) that the rules are correct and the threatened punishments,just. If a rule lacks this collective endorsement, individuals can flout itwith impunity.

Closely connected with rules are social roles, and they too are collectivelycontrolled perceptions that fit in with the cultural-contingency domain. Thecontingencies in this case prescribe the behavior one exhibits if one is toplay the role. To be an X, one must do Y. Again, such prescriptions do notactually control the behavior of the person playing the role, because theperson can always act in-role or out-of-role. But the collectively controlledperception of what a person should do in the role serves as a source ofreference standards for people to control their perceptions of their ownactivities if they so choose.

Technology represents a somewhat different kind of collective perception in thecultural-contingency domain. Technology follows a similar "to achieve Y, do X"pattern, but in this case the rules are about chemical formulae or mathematicalalgorithms or directions for how to make and do things. As is true for all theother kinds of cultural contingencies, the rules and procedures of technologydiffer from cultural group to cultural group. For example, Hutchins (1995, Ch.2) provides a intriguing account which contrasts the technology of navigationused by Micronesian Islanders with that of European sailors.

Like perceptions in other domains, cultural contingencies from more than onelayer can be controlled in macro as well as micro situations. Culturalcontingencies, however, are limited to the two broad inner layers ofperception. As I noted earlier, perceptions in cultural domains are neverlinked uniquely to Layer I perceptions. Rather, as perceptions in Layer II orIII, one can find meaning for them in an indefinite number of concretesituations. A rule, to be a rule at all, must apply to more than one thing.Most cultural contingencies are middle-layer perceptions--sequences orprograms--either propositions about what should happen or arrays of choices onhow to proceed. We can, however, perceive cultural contingencies in Layer III,say, as sets of legal principles, or in the workings of "the market," or in thenebulous but all-encompassing principles of right behavior we call values. Tothe extent that rules or other cultural contingencies become more inclusivethey also become more macro in their scope. The number of people involved intheir collective control becomes larger, and the rules themselves become moreponderous, harder to change. The great advantage of law over custom is thatlaws can be changed by groups of macro actors--parliaments orlegislatures--while widespread customs have all the inertia of thesuper-strong, super-slow, and generally unintelligent "virtual actors" Idescribed above.

My final observation about the domain of cultural contingencies is that it hasan especially close connection with the domain of social activities. Culturalcontingencies are shared perceptions that tell us how our social activitiesshould be carried out. Because cultural contingencies are abstract, however,they cannot be expected to describe in complete detail how a particularactivity should unfold, and every repetition of a given activity will differfrom all its predecessors in its minute physical details, at the very least.Human actors striving to conform to the prescribed contingencies always have tocope--in good control-system fashion--with unpredictable physical disturbancesin their effort to achieve the inner-layer perceptions they intend. Thoughpriests have performed the same ritual every day for thousands of years, theCatholic mass is never quite the same twice in a row. Great musicians can playa piece over and over and perceive something new every time. In short,cultural contingencies allow us to give structure to our social life, withouttotally determining it.

Cultural Discourses

In the same way that cultural contingencies connect closely to thesocial-activity domain, the final perceptual domain I offer, culturaldiscourses, has a close connection to the other social-structuraldomain: social designations. Just as cultural contingencies providedirections for our social activities, cultural discourses provide theconceptual tools we use in our social designations of the people, things, andevents in our social environment. I have said already that language is themost important of the cultural discourses. Without language, socialdesignations would be virtually impossible to make. How could we name thingsif we had no names?

Language is a textbook example of a collectively controlled perception. We alluse it. We each have somewhat different reference standards for meanings ofwords and ways to construct sentences. We control our perceptions of our ownspeech with our own reference standards, and we listen with our own referencestandards, too, which sometimes gets us into trouble when others have standardssufficiently different from ours. Out of all this individual activity emergesa set of virtual reference standards that grammarians seek to describe (thoughmost often they want to enshrine the virtual reference standards of the groupof macro actors who constitute the "highest" strata of society). Virtualreference standards for language change only slowly over time, but change theydo, as words come and go, and grammar shifts.

Language is also abstract. Its relation to any particular perception on theconcretely physical layer is essentially symbolic. Some of these physicalperceptions serve conventionally as the symbols of the abstract concepts oflanguage. We perceive marks scratched on paper and sounds emerging from mouthsas symbols of the perceptions we call words. We all give personal meanings tothose words through our own actions or imaginations, linking perception toperception. Language attains uniformity, to the extent that it does, becausewe imitate others--we imagine the reference standards they must be using andseek to match our own perceptions to them. We conform because we have learnedthat not conforming disrupts communication. Both in its abstraction and in itscollective force, language typifies cultural discourses.

Other cultural discourses, besides language, include bodies of tradition andcommon-sense understandings of the world: our divisions of time into minutesand hours and space into inches and miles; our classifications of people bywhere they live or the clothes they wear; our categories for plants andanimals, stars and seasons; even cuisines, fashions in clothing, and styles ofarchitecture. All these phenomena appear in cultural discourses composed ofcollectively controlled perceptions. Take, for example, the common imagery weuse in talking about gender or race differences. Such perceptions of race andgender provide the social categories into which we sort people, the hierarchieswe perceive among those categories, the myths and stories we tell torationalize such hierarchies, and the innate qualities of virtue and vice weattribute to members of such groups. Cultural discourses of gender and race,together with the perceptions of cultural contingency which specify theexpected behaviors of people in each group, make up the complex of perceptionsoften referred to as institutionalized racism or sexism.

Conventional images of youth and age, family relationships, occupations, andnational character all come from collective perceptions in thecultural-discourse domain, too. We create social roles by combining theseperceptions of the parts people play with our perceptions of the appropriatebehavior for each part--from the helpless newborn to the doddering grandparent.Central to the collective perception of such conventional images, identities,and actions is an emotional evaluation: we admire the handsome hero and loathethe mustachioed villain. Heise (1977, 1979, 1987) and his many coworkers, intheir extensive research on "Affect Control Theory," have shown that affectiveperceptions are widely shared; they have also used the principles of PCT todescribe how people control these sentimentally tinged perceptions of culturaldiscourse.

All types of stock figures, plots, and themes from mythic or literary sourcesalso qualify as collective perceptions in the cultural discourse domain. FromMoses to Mickey Mouse, collectively perceived images, anecdotes, legends, andcharacters provide the common context for the stories we tell each other.Works of art and music, novels, plays, TV shows, movies, and otherentertainments are collective perceptions in their own right, as are commercialslogans, trademarks, and advertisements. Millions of people, for example, viewmass entertainments simultaneously, each through an individual perceptual lens,and from the centroid of these individual reactions comes the virtual actor who"decides" whether the spectacle is a hit or a flop. Works of music or theaterare collective perceptions in another sense, too; many different artists canperform a play or piece of music, just as books can have many readers. Nosingle person can give the "definitive" performance or provide the definitivereading. While performances and productions of art are, of course, socialactivities, and the rules of aesthetics can be classed as culturalcontingencies, art in essence belongs to the domain of cultural discourse, andthus is entirely a matter of controlling perceptions, both in the creating andthe experiencing.

Cultural discourses also include our systematic compilations of knowledge aboutthe world. Scientific theories, legal theories, economic theories, literarytheories, academic fields and disciplines, religions and theologies, bodies ofesoteric lore--and even sociological theories--all fit the definition.Histories of all kinds, from official archives to the daily news and the latestgossip, are cultural discourses, too. For each field of knowledge, a group ofpractitioners exercise collective control of the perceptions which constitutethe field; each individual knows his or her own portion of the whole, but noone knows everything. Yet the field of knowledge does not exist apart from theperceptions of its practitioners, however disparate and conflicting. In maturefields of knowledge, the practitioners' points of view have enough consensusthat the ghost of a virtual actor seems to rise up from the widely acceptedtexts which symbolize their shared knowledge.

Perceptions from the cultural-discourse domain, like perceptions from the threeother domains, occur in macro as well as micro situations. Small groups, bydevising an in-group slang, for example, can establish their own culturaldiscourse and thus create idiocultures (Fine 1979)--which ordinarily will alsoinclude perceptions of cultural contingency, like the rules and expectationspeculiar to the small group. Most other cultural discourses, however, comprisecollective perceptions that are not micro in scale but rather society-wide, oreven world-wide, as the trappings of contemporary capitalist culture, like newsmedia, products, trademarks, and TV shows, have spread to all parts of theglobe. In an "information society," cultural discourses become almostuniversally open to participation. Few pockets of parochial peculiarityremain.

Perceptions from the cultural-discourse domain, like those from the othercultural domain, pertain to either the middle or the innermost perceptuallayer. Cultural-discourse perceptions range from the most basic Layer-IIperception for a category--like "dog" or "cat"--to the most comprehensive ofLayer-III system concepts for a whole field of knowledge--like "science."Commercial products and trademarks neatly illustrate the relationship betweenthese cultural-discourse perceptions and Layer I perceptions of physicalobjects--which are not the cultural-discourse perceptions, but onlysymbols for them. A trademark like "Coca-Cola"--as a Layer IIcultural-discourse perception--can be applied to physical objects, likeparticular cans of "Coke." A can of Coke, however, has no individuality initself, or at least none that we ordinarily care about--a can of Coke is a canof Coke, whether bottled in Atlanta, Seattle, or Kuala Lumpur.[40] Thus, the trademark fits millions of cans of soft drinkequally well, and each logo-emblazoned object symbolizes our abstractperception of the product. Both the logo on the can and the liquid inside itserve as symbols. In sum, the perceptions of cultural discourse never belongto Layer I, but they are closely associated with two kinds of physicalperceptions: a) a substrate of physical objects or actions which give themtheir concrete meaning; b) texts, spoken words, pictures or other"representations" which symbolize the abstract perceptions and allow us tocommunicate with each other about them.

Connections Among Domains

In laying out four domains of collectively controlled perceptions, I have meantto suggest that inner-layer control loops in a given domain are usuallyconnected to control loops at outer layers of the same domain. But I do notmean to imply that the domains are independent of one another. (See Figure12.) For instance one's social identity, one's highest concept of self--whichI have argued is a Layer III collectively controlled perception in thesocial-designation domain--may be constructed in part from social roles, whichare Layer II perceptions from the cultural-contingency and culture-discoursedomains. A map of the connections of control loops in an actual hierarchy,then, might often show connections between outer-layer perceptions in onedomain and inner-layer perceptions in another, as well as connections withinthe domain.

Consider our Layer-III collectively controlled perception for "the state."When we aren't talking about some particular state, like the Federal Governmentof the United States or Her Majesty's Government of the United Kingdom, butabout states in general, our perception of "the state" comes from thecultural-discourse domain. However, we draw upon outer-layer perceptions fromall the other domains to give meaning to this highly abstract system concept.From the cultural-contingency domain, we envision as necessary components of astate a body of laws and legal principles, as well as a set of rules andregulations for enacting, enforcing, and adjudicating those laws. From thesocial-activities domain, we see the processes of political in-fighting,legislative bargaining, bureaucratic oversight, and legal maneuvering throughwhich the rules are implemented. From the social-designations domain weimagine the specific territories and populations the state serves, as well asthe other states with which it must contend. And from the cultural-discoursedomain we draw many additional perceptions--of the histories of various states,of typical divisions and layers of the state machinery, of the usual offices orpositions filled by state functionaries--all of which we connect with ourcomplex, high-layer perception of "the state." The sets of particularperceptions which any two people associate with a complex and generalizedperception like "the state" will inevitably be different. And when pressed,most people will give idiosyncratic meaning to their perceptions by resortingto specific perceptions of the social activities and social designations of thestate or states with which they are most familiar. The great utility of theview I am offering here is that it shows how social order emerges fromcollective control of such perceptions even when strict conformity is lacking,and it opens the way for us to begin disentangling the awesome complexity ofsocial phenomena.

CONCLUSION

Turner (1994), whose sharp critique of sociological theory was reported in myintroduction, is by no means the only observer to bemoan the current state ofsociology. Sociologists from all camps have complained in recent years thatthe field is disintegrating. For example, in calling for a Postmodernistapproach, Seidman charges that sociological theory has become "a dispersed,discursive clamoring that covers a wide assortment of ever-changing issues in adazzling diversity of languages" (1991, p. 133). From the other end of thetheoretical spectrum, Couch begins his Presidential Address to the MidwesternSociological Society by asserting that "sociologists are fragmented" and"[m]any have questioned the viability of sociology" (1994, p. 1). From yetanother ideological camp within the discipline, Horowitiz (1993) offers abook-length account of the historical reasons for sociology's "decomposition."

Observers agree that sociologists have not yet found a comprehensivetheoretical framework to focus and integrate their research programs. Liketheir colleagues in other social sciences--such as anthropology, politicalscience, and psychology--and unlike researchers from the hard sciences,sociologists have made do with an ad-hoc collection of mutually conflictingpartial theories. All attempts so far to concoct grand theories (for example,Parsons 1951, Giddens 1984) have run into insoluble contradictions (seeHolmwood and Stewart 1994), and none now claims the loyalty of a majority ofresearchers. Perhaps the field's predicament stems in part from the complexityof its subject matter--human interaction is more complicated by orders ofmagnitude than, for instance, the inert materials studied by physicists.Without doubt, the theoretical failings itemized by Turner (1994) have madematters worse for sociology. But a general consensus is emerging thatsociologists must find a new way of looking at social life, if sociology is tosurvive as a field.

In this article, I have presented a radically new perspective forsociology--the theory of collective control of perceptions--which, I maintain,provides the needed framework for revitalizing and unifying the field. Thisnew approa