Obedience to Authority

by Stanley Milgram

  The advantages of social organization reach not only outward, toward external goals, but inward as well, giving stability and harmony to the relations among group members. By clearly defining the status of each member, it reduces friction to a minimum. When a wolf pack brings down its prey, for example, the dominant wolf enjoys first privileges, followed by the next dominant one, and so on down the line. Each member’s acknowledgment of his place in the hierarchy stabilizes the pack. The same is true of human groups: internal harmony is ensured when all members accept the status assigned to them. Challenges to the hierarchy, on the other hand, often provoke violence. Thus, a stable social organization both enhances the group’s ability to deal with its environment and by regulating group relationships reduces internal violence.

  A potential for obedience is the prerequisite of such social organization, and because organization has enormous survival value for any species, such a capacity was bred into the organism through the extended operation of evolutionary processes. I do not intend this as the end point of my argument, but only the beginning, for we will have gotten nowhere if all we can say is that men obey because they have an instinct for it.

  Indeed, the idea of a simple instinct for obedience is not what is now proposed. Rather, we are born with a potential for obedience, which then interacts with the influence of society to produce the obedient man. In this sense, the capacity for obedience is like the capacity for language: certain highly specific mental structures must be present if the organism is to have potential for language, but exposure to a social milieu is needed to create a speaking man. In explaining the causes of obedience, we need to look both at the inborn structures and at the social influences impinging after birth. The proportion of influence exerted by each is a moot point. From the standpoint of evolutionary survival, all that matters is that we end up with organisms that can function in hierarchies.10

  The Cybernetic Viewpoint

  A clearer understanding will be found, I believe, by considering the problem from a slightly different point of view—namely, that of cybernetics. A jump from evolution to cybernetics may appear at first arbitrary, but those abreast of current scientific developments know that the interpretation of evolutionary processes from a cybernetic viewpoint has been advanced quite brilliantly in recent years (Ashby, 1956; Wiener, 1950). Cybernetics is the science of regulation or control, and the relevant question is, What changes must occur in the design of an evolving organism as it moves from a capacity for autonomous functioning to a capacity for functioning within an organization? Upon analysis certain minimum requirements necessary to this shift become apparent. While these somewhat general principles may seem far removed from the behavior of participants in the experiment, I am convinced that they are very much at the root of the behavior in question. For the main question in any scientific theory of obedience is, What changes occur when the autonomously acting individual is embedded in a social structure where he functions as a component of a system rather than on his own? Cybernetic theory, by providing us with a model, can alert us to the changes that logically must occur when independent entities are brought into hierarchical functioning. Insofar as human beings participate in such systems, they must be subject to these general laws.

  We begin by specifying a design for a simplified creature, or automaton. We will ask, What modifications in its design are required if it is to move from self-regulation to hierarchical functioning? And we will treat the problem not in a historical manner but purely formally.

  Fig. 19. Simple homeostatic model.

  Consider a set of automata, a, b, c, and so on, each designed to function in isolation. Each automaton is characterized as an open system, requiring inputs from the environment to maintain its internal states. The need for environmental inputs (e.g., nourishment) requires apparatus for searching out, ingesting, and converting parts of the environment to usable nutritive forms. Action is initiated via effectors triggered when inner conditions signal a deficiency threatening the automaton’s vital states. The signal activates search procedures for nutritive inputs that restore the system to a state of viable functioning. Cannon’s homeostatic model (1932) points to the ubiquitousness of such state-restoring systems in living organisms.

  The automata now dwell apart as self-regulating omnivores. To bring them together, even in the most primitive and undifferentiated form of social organization, something must be added to the model we have designed. A curb must be placed on the unregulated expression of individual appetites, for unless this is done, mutual destruction of the automata will result. That is, other automata will simply be treated as parts of the environment and destroyed or acted upon for their nutritive value. Therefore a critical new feature must be added to the design: an inhibitor that prevents automata from acting against each other. With the addition of this general inhibitor these automata will be able to occupy the same geographic area without danger of mutual destruction. The greater the degree of mutual dependency among the automata, the more widely ranged and effective these inhibitory mechanisms need to be.

  More generally, when action is initiated by tensions originating within the individual, some mechanisms internal to the individual must inhibit that expression, if only to prevent its being directed against kindred members of the species in question. If such an inhibitory mechanism does not evolve, the species perishes, and evolutionary processes must come up with a new design compatible with survival. As Ashby (1956) reminds us:

  The organisms we see today are deeply marked by the selective action of two thousand million years attrition. Any form in any way defective in its power of survival has been eliminated; and today the features of almost every form bear the marks of being adapted to ensure survival rather than any other possible outcome. Eyes, roots, cilia, shells and claws are so fashioned as to maximize the chance of survival. And when we study the brain we are again studying a means to survival. (p. 196)

  Is there anything in human beings that corresponds to the inhibitory mechanisms this analysis requires? The question is rhetorical, for we know that the impulse to gratify instincts destructive to others is checked by a part of our nature. Conscience or superego are the terms used to refer to this inhibitory system, and its function is to check the unregulated expression of impulses having their origin in the tensional system of the person. If our automata are beginning to take on some of the properties and structures present in human beings, it is not because human beings provided the model, but rather because parallel design problems arise in constructing any system in which the member organisms sustain themselves through environmental inputs but do not destroy their own kind.

  The presence of conscience in men, therefore, can be seen as a special case of the more general principle that any self-regulating automaton must have an inhibitor to check its actions against its own kind, for without such inhibition, several automata cannot occupy a common territory. The inhibitor filters or checks actions that have their origin in internal imbalances of the automaton. In the case of the human organism—if we may employ psychoanalytic terminology—instinctual urges having their origin in the id are not immediately channeled into action but are subjected to the inhibitory checks of the superego. We note that most men, as civilians, will not hurt, maim, or kill others in the normal course of the day.

  Hierarchical Structuring

  The automata now act individually, limited only by the inhibition against hurting their own kind. What will happen when we try to organize several automata so they function together? The joining of elements to act in a concerted fashion may best be achieved by creating an external source of coordination for two or more elements. Control proceeds from the emitting point to each of the automata.

  Still more powerful social mechanisms can be achieved by having each subordinate element serve as a superordinate to elements in a level below.

  The diagram comes to assume the typical pyramidal form for hierarchical organization. Yet this organization cannot be achieved with the auto
mata as we have described them. The internal design of each element must be altered. Control at the level of each local element must be given up in favor of control from a superordinate point. The inhibitory mechanisms which are vital when the individual element functions by itself become secondary to the need to cede control to the coordinating component.

  More generally, whenever elements that function autonomously are brought into a system of hierarchical coordination, changes are required in the internal structure of the elements. These changes constitute the system requirements, and they invariably entail some suppression of local control in the interest of system coherence. System coherence is attained when all parts of the system are functioning in harmony and not at cross-purposes.

  From an evolutionary standpoint each autonomously functioning element must be regulated against the unrestrained pursuit of appetities, of which the individual element is the chief beneficiary. The superego, conscience, or some similar mechanism that pits moral ideals against the uncontrolled expression of impulses fulfills this function. However, in the organizational mode, it is crucial for the operation of the system that these inhibitory mechanisms do not significantly conflict with directions from higher-level components. Therefore when the individual is working on his own, conscience is brought into play. But when he functions in an organizational mode, directions that come from the higher-level component are not assessed against the internal standards of moral judgment. Only impulses generated within the individual, in the autonomous mode, are so checked and regulated.

  The hierarchy is constructed of modules, each consisting of one boss with followers (e.g., A: B, C). Each follower, in turn, may be superior to others below him (e.g., B: D, E), the entire structure being built up of such interlocking units. The psychology of obedience does not depend on the placement of the module within the larger hierarchy: the psychological adjustments of an obedient Wehrmacht General to Adolf Hitler parallel those of the lowest infantryman to his superior, and so forth, throughout the system. Only the psychology of the ultimate leader demands a different set of explanatory principles.

  Variability

  We now need to make clear a point that has been implicit in the argument—namely, the relationship of variability to the need for systemic modification. Where variability is present, efficient structuring into larger systems can only occur by ceding local control to a coordinating component. If not, the larger system will be less efficient than an average individual unit.

  Consider a set of identical entities that can function on their own, say a set of five electric trams that possess governing mechanisms that brake each tram precisely at 50 miles an hour. As long as there is no variability among the individual units, when they are linked together in a five-car train, the train can move along at 50 miles an hour. Consider now that variability is introduced, and the automatic speed governors brake the five cars at 10, 20, 30, 40, and 50 miles an hour respectively. If the cars are formed into a supraordinate system, the train as a whole cannot move faster than the slowest unit.

  If a social organization consists of individuals whose judgments on a course of action vary, coherence can only be secured by relying on the least common denominator. This is the least efficient system possible and hardly likely to benefit its members. Thus suppression of control at the level of the local unit and ceding to higher-level components become ever more important as variability increases. Variability, as evolutionary theorists have long told us, is of enormous biological value. And it is conspicuously a feature of the human species. Because people are not all alike, in order to derive the benefit of hierarchical structuring, readily effected suppression of local control is needed at the point of entering the hierarchy, so that the least efficient unit does not determine the operation of the system as a whole.

  It is instructive to list a few of the systems that function by suppression of local control: individual pilots cede control to the controller in the tower as they approach an airport so that the units can be brought into a coordinated landing system; military units cede control to higher-level authority to ensure unity of action. When individuals enter a condition of hierarchical control, the mechanism which ordinarily regulates individual impulses is suppressed and ceded to the higher-level component. Freud (1921), without referring to the general systems implications of his assertion, spelled out this mechanism clearly: “. . . the individual gives up his ego ideal and substitutes for it the group ideal embodied in the leader” (page 78, Group Psychology). The basic reason why this occurs is rooted not in individual needs but in organizational needs. Hierarchical structures can function only if they possess the quality of coherence, and coherence can be attained only by the suppression of control at the local level.

  Let me summarize the argument so far: (1) organized social life provides survival benefits to the individuals who are part of it, and to the group; (2) whatever behavioral and psychological features have been necessary to produce the capacity for organized social life have been shaped by evolutionary forces; (3) from the standpoint of cybernetics, the most general need in bringing self-regulating automata into a coordinated hierarchy is to suppress individual direction and control in favor of control from higher-level components; (4) more generally, hierarchies can function only when internal modification occurs in the elements of which they are composed; (5) functional hierarchies in social life are characterized by each of these features, and (6) the individuals who enter into such hierarchies are, of necessity, modified in their functioning.

  This analysis is of importance for one reason alone: it alerts us to the changes that must occur when an independently functioning unit becomes part of a system. This transformation corresponds precisely to the central dilemma of our experiment: how is it that a person who is usually decent and courteous acts with severity against another person within the experiment? He does so because conscience, which regulates impulsive aggressive action, is per force diminished at the point of entering a hierarchical structure.

  The Agentic Shift

  We have concluded that internal modification is required in the operation of any element that can successfully function in a hierarchy, and that in the case of self-directed automata this entails suppression of local control in favor of regulation by a higher-level component. The design of such an automaton, if it is to parallel human function, must be sufficiently flexible to allow for two modes of operation: the self-directed (or autonomous mode), when it is functioning on its own, and for the satisfaction of its own internal needs, and the systemic mode, when the automaton is integrated into a larger organizational structure. Its behavior will depend on which of the two states it is in.

  Social organizations, and the individuals who participate in them, are not exempt from the requirements of system integration. What in human experience corresponds to the transition from the autonomous to the systemic mode, and what are its consequences in specifically human terms? To answer the question we must move from a general level of discourse to the close examination of a person as he shifts into a functional position in a social hierarchy.

  Where in a human being shall we find the switch that controls the transition from an autonomous to a systemic mode? No less than in the case of automata, there is certainly an alteration in the internal operations of the person, and these, no doubt, reduce to shifts in patterns of neural functioning. Chemical inhibitors and disinhibitors alter the probability of certain neural pathways and sequences being used. But it is totally beyond our technical skill to specify this event at the chemoneurological level. However, there is a phenomenological expression of this shift to which we do have access. The critical shift in functioning is reflected in an alteration of attitude. Specifically, the person entering an authority system no longer views himself as acting out of his own purposes but rather comes to see himself as an agent for executing the wishes of another person. Once an individual conceives his action in this light, profound alterations occur in his behavior and his internal functioning.
These are so pronounced that one may say that this altered attitude places the individual in a different state from the one he was in prior to integration into the hierarchy. I shall term this the agentic state, by which I mean the condition a person is in when he sees himself as an agent for carrying out another person’s wishes. This term will be used in opposition to that of autonomy—that is, when a person sees himself as acting on his own.

  The agentic state is the master attitude from which the observed behavior flows. The state of agency is more than a terminological burden imposed on the reader; it is the keystone of our analysis. If it is useful, we shall find that the laboratory observations will hang together when linked by it. If it is superfluous we shall find that it adds nothing to the coherence of our findings. For clarity, let me again define what is meant by the state of agency. It may be defined both from a cybernetic and a phenomenological standpoint.

  From the standpoint of cybernetic analysis, the agentic state occurs when a self-regulating entity is internally modified so as to allow its functioning within a system of hierarchical control.

  From a subjective standpoint, a person is in a state of agency when he defines himself in a social situation in a manner that renders him open to regulation by a person of higher status. In this condition the individual no longer views himself as responsible for his own actions but defines himself as an instrument for carrying out the wishes of others.

  An element of free choice determines whether the person defines himself in this way or not, but given the presence of certain critical releasers, the propensity to do so is exceedingly strong, and the shift is not freely reversible.

  Since the agentic state is largely a state of mind, some will say that this shift in attitude is not a real alteration in the state of the person. I would argue, however, that these shifts in individuals are precisely equivalent to those major alterations in the logic system of the automata considered earlier. Of course, we do not have toggle switches emerging from our bodies, and the shifts are synaptically effected, but this makes them no less real.