Control Theory for Whom?

Control theory for humans: Quantitative approaches to modeling performance

By R. Jagacinski and J. Flach (NJ: Erlbaum, 2002)

Reviewed by Richard S. Marken, Senior Behavioral Scientist, The RAND Corporation

Control theory is a set of mathematical equations that describe the mechanisms that make it possible for living and artificial systems to control variables in their environment. Engineers use the theory as a basis for building artificial control systems (such as thermostats) and behavioral scientists use it as a basis for understanding the behavior of living control systems (such as humans). While the control theory used by engineers is mathematically identical to that used by behavioral scientists, the way the theory is used is quite different in the two cases. The difference turns on the fact that engineers know what variables they want their artificial systems to control while behavioral scientists want to find out what variables living systems actually do control. Engineers use the equations of control theory to help them build systems that will effectively control the variables they (the engineers) want controlled. Behavioral scientists use the equations of control theory to help them understand how living systems effectively control the variables that they (the living systems) want controlled.

While there are many texts that describe control theory for the benefit of engineers, there are few that describe it for the benefit of behavioral scientists. Control theory for humans represents an attempt to remedy this imbalance by providing a detailed description of control theory for students in the behavioral sciences. The book is successful inasmuch as it provides a very complete and lucid description of the mathematical details of control theory itself. It is less successful, however, in presenting the theory in a way that is useful to the behavioral scientist. This is because the book approaches control theory more from the point of view of the engineer than from that of the behavioral scientist. In particular, the book assumes that the behavioral scientist, like the engineer, knows what variables are being controlled by the system under study. Control theory is then presented as a way of evaluating how well the system (a human, in this case) is controlling these variables rather than as a way of determining what variables the system is actually controlling.

Control theory for humans introduces the tools of control theory in the context of a control task called manual tracking, where a person acts to keep a cursor aligned with a target. Manual tracking is an excellent example of human controlling because all the variables involved in control are clearly identifiable. But, for the same reason, it is a poor example of human controlling as well. In manual tracking a person is instructed to control a particular variable, such as the relationship between cursor and target position, and to keep that variable in a particular state, such as "cursor aligned with target". If the person follows instructions (and most do) then we know that the relationship between target and cursor is under control and we can use the tools of control theory (linear systems analysis, Bode plots and so on, which are described so well in Control theory for humans) to evaluate how well this control is being carried out. But manual tracking is a poor example of human controlling because it can give one the misleading impression that the variables involved in everyday examples of human control — particularly the variables that people are controlling — can be easily identified. In fact, they cannot. Indeed, the main problem involved in understanding everyday control behaviors (such as walking, talking, and catching fly balls) is identifying the variables people are controlling when they carry out these behaviors.

The variables people control are called controlled variables. A controlled variable is a perceived aspect of a person’s environment that is kept in a predetermined state and protected from disturbances by the person’s actions. The relationship between cursor and target in a tracking task is a controlled variable that is kept in a predetermined state (cursor aligned with the target) and protected from disturbance (such as movements of the target) by the person’s actions (such as movements of a "control" stick). Controlled variables can be simple (such as the relationship between cursor and target) or complex (such as the relationship between a husband and wife). They are perceptual variables because they are often complex functions of many physical variables. For example, "sweetness" is a variable that people control yet this variable corresponds to no single physical variable in the environment. Human behavior is organized around the control of such perceptual variables (Powers, 1973) so knowing what perceptual variables people control is essential to the analysis of human behavior from a control theory perspective.

The fact that controlled variables are perceived aspects of a person’s environment is one reason why it is difficult to notice these variables when we observe everyday examples of human controlling. It is difficult to notice, for example, what a baseball outfielder is controlling because the controlled variable is a perception in the fielder, not in the person observing the fielder’s behavior. Although controlled variables are difficult to identify, they must be identified before the tools of control theory can be used to model and evaluate human controlling. But controlled variables and the methods for detecting them are never mentioned in Control theory for humans. This is unfortunate because it limits the application of control theory to situations where the variable being controlled by the system under study is already known with considerable confidence. That is, it limits behavioral science applications to engineering-type applications of control theory.

The application of control theory need not be limited in this way because simple and intuitively satisfying methods exist that can be used to identify the variables a system is controlling. Controlled variables can be identified without telling the system what to control, as in manual tracking tasks. Methods for identifying controlled variables have been referred to collectively as "the test for the controlled variable" or simply the TCV (Marken, 1997). The basic idea behind all methods that make up the TCV is that a controlled variable is a perceived aspect of a person's environment that will be protected from disturbance. The first (and possibly the most difficult) step in finding out what a person is controlling is to identify variables in the environment that the person might be perceiving and controlling. In the case of a person walking down the street, for example, that might be (among other things) the distance between the person and other people. If this variable is, indeed, under control then the person will protect it from disturbances, such as people who get too close, by moving appropriately. Thus, it is possible to determine whether or not a variable is under control by applying disturbances (such as by moving closer or farther from a person) to the suspected controlled variable and looking to see if that variable is protected from these disturbances. If it is, then the variable is likely to be under control. If not, then another hypothesis about the variable under control is tested.

Once we know what variables people control we can understand a lot about why they act as they do. This is because the actions of a control system (such as a person) are aimed at bringing controlled variables to preferred states and maintaining them in those states, protected from disturbance. Changes in preference for or disturbances to the state of a controlled variable result in observable actions that can be predicted based on an understanding of the environmental constraints under which these control actions occur. In order to make such predictions with accuracy it is necessary to be able to build a model of the human controller. Control theory for humans does an excellent job of introducing modeling as an approach to understanding the behavior of human controllers. However, the success of such models, in terms of how well the behavior of the model matches that of the human being modeled, depends, in large part, on including a correct representation the controlled variable. The models discussed in Control theory for humans assume that the modeler knows what variable a person is controlling when carrying out a particular behavior, such as driving and shows how the equations of control theory can be used to produce a model that is dynamically stable, like the person being modeled. But the book does not show how one determines what variable the model should control in order to act like the human. This is a serious omission in a book aimed at behavioral scientists since it is often the successful selection of the appropriate controlled variable that makes the difference between models that do and don't fit the observed human behavior (Marken, 2001).

Control theory for humans provides an excellent description of the tools that can be used to model and analyze the behavior of living control systems once the variables controlled by these systems have been identified. In this sense, it can be considered the second half of a textbook on control theory for the behavioral sciences, the first half of which has yet to be written. The first half of that textbook will explain how "control theory for humans" (control theory for behavioral scientists) differs from "control theory for thermostats" (control theory for engineers). That is, it will explain how to find out what variables living control systems are controlling and why we have to know what variables they are controlling in order to understand their behavior. Until that first half is written, however, some existing books may fill the void. In particular I am thinking of two books by William T. Powers: Behavior: The control of perception (Powers, 1973) and Making sense of behavior (Powers, 1998). Although these books are not written intentionally as introductory texts, they do cover the introductory topics that are missing from Control theory for humans. These books explain what control is, what it looks like when seen in living systems and how to identify the variables living systems are controlling when they are engaged in everyday behaviors. By reading these books as a preamble to Control theory for human, the student of behavioral science will see that control theory is, indeed, for humans, and not just for thermostats.

References

Marken, R. S. (1997) The dancer and the dance: Methods in the study of living control systems, Psychological Methods, 2 (4), 436-446

Marken, R. S. (2001) Controlled variables: Psychology as the center fielder views it, American Journal of Psychology, 114, 259 -281

Powers, W. T. (1973). Behavior: The control of perception. Hawthorne, NY: Aldine DeGruyter.

Powers, W. T. (1998). Making sense of behavior. New Canaan, CT: Benchmark.