In his book Behavior: The Control of Perception (1973), William T. Powers offered a radically new approach to understanding human behavior. Now called "Perceptual Control Theory" or "PCT," this theory proposes a specific mechanism whose output behaviors serve to keep an individual's myriad of controlled perceptions near internally determined reference levels. In this paper I offer a synopsis of perceptual control theory.
At the heart of perceptual control theory is the idea that human beings are essentially intricate control mechanisms that function to keep certain intrinsic (or essential, see Ashby, 1952) variables within survivable limits. Intrinsic variables (variables intrinsic to the organism) include basic physiological variables such as blood glucose levels or body temperature, as well certain high-level variables whose maintenance in certain states are crucial to the individual's well-being; the references of these intrinsic variables are genetically specified. With respect to physiological quantities, the body is known to house numerous control mechanisms that help to maintain them within the narrow limits required for efficient operation and survival. These mechanisms are capable of sensing the current levels of these controlled quantities and automatically initiating physiological changes as necessary to correct deviations of these levels from reference values, a process that the early 20th century physiologist Walter Cannon (1932) termed homeostasis. Such systems actively defend against any disturbances to their controlled quantities.
Although at any given moment a tremendous number of physiological quantities is being automatically regulated through nonbehavioral (purely physiological) means, the regulatory mechanisms by themselves are not capable of countering all the sources of potential disturbance to the intrinsic variables. To take one example, because humans are not rooted in the soil like plants, we must seek out and consume food and water. Automatic physiological mechanisms do act against disturbances to internal levels of water and nutrition, but these only can take the form of actions to reduce the rate of depletion of these quantities. To replenish them, we must behave. That is, we must move our muscles in a way that ultimately leads to locating, obtaining, and consuming food and water. Behavior, then, is a means by which humans (and other animals) defend their intrinsic variables against disturbance.
Although, according to PCT, behavior is a means of defending intrinsic variables against disturbance, that control is indirect. Dehydration of the body, for example, does not lead directly to behavior. Instead, control of the state of hydration is achieved indirectly, by controlling one's perceptions. For example, in the usual course of events, one would rehydrate the body by drinking water. However, drinking water is a complex act that may involve obtaining a glass, filling it with water, raising the glass to the lips, pouring the water into the mouth, and swallowing. Each of these acts is itself a complex act that requires the monitoring and control of many variables.
Monitoring a variable requires some kind of sensory apparatus, which senses the variable and produces an output whose values correspond in some systematic way to those of the sensed variable. In the nervous system, sensory receptors output a time-varying rate of neural impulses as functions of the physical quantities to which they are sensitive. In PCT these sense-related neural outputs are referred to as perceptual signals or "perceptions."
It is important to note that this use of the term "perception" conveys a somewhat different meaning in PCT than is familiar to sense physiologists and psychologists. As usually defined, a perception is a cognitive process of interpretation of sensory input; in PCT a perception is simply a signal conveying sense-data from sensory receptors or a signal derived from other such signals. A perception so defined need have no conscious representation.
Not all perceptual signals are necessarily controlled, but when they are then according to PCT, behavior is the means of control; hence the title of Powers' book, Behavior: the Control of Perception. The central assertion of PCT is that behavior exists solely for the purpose of controlling one's perceptions. Those perceptions, in turn, are controlled as a means of keeping intrinsic variables within their critical limits.
The basic PCT thesis simply asserts that behavior is the control of perception, and leaves open the exact nature of the control systems that accomplish the control (to be determined in future experimental work). However, Powers has provided a more detailed if speculative proposal outlining a possible organization of the human system. To distinguish this more elaborate proposal from the basic thesis, Powers refers to it as Hierarchical Perceptual Control Theory or HPCT. HPCT includes a hierarchy of perceptions and a parallel hierarchy of control.
HPCT specifies a hierarchy of perceptions, beginning at the bottom with simple "intensity" signals and running through a number of hierarchically organized levels (currently 11). Each succeeding level builds new perceptual signals by combining in various ways the perceptual signals from the level immediately below (except for the first, which derives its signals from sensory mechanisms).
The lowest-order perceptions in the hierarchy derive from sensory endings and, according to Powers, convey only the intensity of stimulation. Thus, signals at this level are termed "intensity" signals and when consciously perceived convey only the impression of intensity, devoid of any qualitative labels.
Intensity signals combine at the next level to produce vector quantities that, according to Powers, we experience as sensations (when we do experience them). Higher levels of perception are derived in like manner from the perceptual signals arising at lower levels.
The orders of perception included in the perceptual hierarchy were developed so as to be consistent with available physiological evidence (lowest levels) and experience (conscious perception). The nine levels proposed in Behavior: The Control of Perception were as follows:
Associated with each level of perception is a level of perceptual control. Each level except for the top one receives its reference signal from the outputs of the control systems in the next higher level. The outputs of the lowest level go to the muscles (and glands). The reference values of the highest-level control systems are assumed to be fixed. Thus there is a perceptual control hierarchy that exactly parallels the perceptual hierarchy. Because each level of control acts in the service of the level immediately above it (except for the top level, whose reference levels are assumed to be fixed), this is a purely top-down model of control.
Perceptual signals at levels above the bottom one are controlled by manipulating the reference signals of control systems at the next lower level. Those lower-level control systems then act to bring their controlled perceptions into line with the new reference values. These changes in turn alter the values of the next-level-up perceptions that are synthesized from the lower-level perceptions. In this way the higher-level systems use the next-lower-level ones as the means whereby the higher-level systems control their own perceptions.
An interesting possibility within this hierarchical arrangement is that changes ordered in lower-level perceptions by higher-level systems may act as disturbances to other controlled perceptions. The systems controlling these signals will then act automatically to correct these disturbances. However, the control hierarchy must be constructed so as to allow all systems to maintain reasonable control over their own perceptual signals. If not, conflict between control systems occurs that may have devastating effects on the individual's ability to maintain control over certain perceptions. In the worst case this could lead to a failure to maintain intrinsic variables within their necessary limits. This consideration takes us to the next element in the HPCT proposal: reorganization.
HPCT proposes a second control mechanism lying outside of the perceptual control hierarchy. Called the reorganizing system, this mechanism functions to change the organization of the perceptual control hierarchy so as to remove damaging conflicts and improve control of perceptual signals at all levels. Reorganization can involve merely changing the parameters of control (e.g., the loop gain) within a given system so as to increase its efficiency or sensitivity or eliminate problems such as a tendency to oscillate. Alternatively, reorganization can involve establishing new connections between levels of the control hierarchy, destroying existing connections, or even creating new perceptual signals (by combining next-down perceptual signals in new ways) and new control systems to control them.
If a reorganizing system is to be effective, it must have a mechanism to determine when changes are needed, and some means of implementing the changes. Powers proposes (after Ashby, 1952) that what drives reorganization is persistent error in intrinsic variables. This presupposes some mechanism capable of sensing the states of these intrinsic variables, perhaps through chemical means if not via neural signals. According to Powers, these states are compared to genetically given reference values to compute the level of error. In Powers' proposal, reorganization occurs continually at a rate proportional to the level of persistent error. Thus reorganization will proceed more swiftly when error is large then when error is small, and may nearly cease entirely when persistent error is kept very low by the perceptual control hierarchy.
As a starting point, Powers proposes that reorganization is essentially a random process. The reorganizing system will keep making changes to the control hierarchy until persistent error declines to low levels. This method constitutes a Darwinian process of variation and selective retention and produces a biased random walk toward more effective control configurations. An important feature is that "the process of reorganization is independent of the kind of behavior being reorganized" (Powers, 1973, p. 187).
Powers recognizes that more efficient, "targeted" methods of reorganization may be required if the system is to make changes only to those portions of the control hierarchy that are not functioning well. He suggests, for example, that a more elaborate mechanism might be able to sense the total level of error in a given hierarchy and act to reorganize only those perceptual or control hierarchies for which total error tends to remain high (Powers, 1973, p. 195).
The reorganizing system is posited to come into being early in fetal development, along with the necessary components for the lowest level in the control hierarchy. The reorganizing system then creates each succeeding level as solutions to the problem of reducing error in the intrinsic variables. This proposal amounts to a theory of physical development of the nervous system which says that the control hierarchy is self-organizing through a Darwinian process of variation and selective retention, as opposed to being genetically specified in detail. Because random reorganization will "discover" different solutions across individuals, this view implies that nervous system organization that emerges will be very different from person to person.
Powers defines memory as "the storage and retrieval of information carried by neural signals" (Powers, 1973, p. 205). Memories are accessed via "associative addressing," a process in which a part of the stored information itself acts as a key that selectively enables access to other stored information associatively linked to it. The fact that human beings can act to reproduce perceptions (as when someone mimics the behavior of another) led Powers to modify one important element of the control hierarchy as originally proposed. The original conception had the outputs from higher-level control systems serving as the reference signals for the lower-level systems. In the modified model, outputs from higher-level systems now function as address signals to retrieve stored perceptual signals. These retrieved signals become the reference signals of the lower-order systems. Powers notes (p. 217) rather cryptically that this new arrangement "solves the problem of translating an error in a higher-order variable into a specific value of a lower-order perception . . ." The retrieved perceptual signal may be routed to the lower-order system's reference-signal input as described and/or it may replace the perceptual signal arising from that system's input function that ordinarily would be routed to higher levels in the brain.
To prevent the two signals from being hopelessly mixed, Powers introduces a memory switch and a perceptual switch. Various combinations of switch position yield different modes of operation, as follows: