Control Theory for Humans

Quantitative Approaches To Modeling Performance

Richard J. Jagacinski, John M. Flach

• Publisher: CRC Press
• Year: 2018
• ISBN:
  9781498720496
• Edition: 1
• Topic: Technology & Engineering
• Subtopic: Industrial Health & Safety
• Index: Technology & Engineering

This textbook provides a tutorial introduction to behavioral applications of control theory. Control theory describes the information one should be sensitive to and the pattern of influence that one should exert on a dynamic system in order to achieve a goal. As such, it is applicable to various forms of dynamic behavior. The book primarily deals with manual control (e.g., moving the cursor on a computer screen, lifting an object, hitting a ball, driving a car), both as a substantive area of study and as a useful perspective for approaching control theory. It is the experience of the authors that by imagining themselves as part of a manual control system, students are better able to learn numerous concepts in this field. Topics include varieties of control theory, such as classical, optimal, fuzzy, adaptive, and learning control, as well as perception and decision making in dynamic contexts. The authors also discuss implications of control theory for how experiments can be conducted in the behavioral sciences. In each of these areas they have provided brief essays intended to convey key concepts that enable the reader to more easily pursue additional readings. Behavioral scientists teaching control courses will be very interested in this book.

1

Perception/Action: A Systems Approach

What is a system? As any poet knows, a system is a way of looking at the world.

–Weinberg (1975)

Today we preach that science is not science unless it is quantitative. We substitute correlation for causal studies, and physical equations for organic reasoning. Measurements and equations are supposed to sharpen thinking but…they more often tend to make the thinking non-causal and fuzzy. They tend to become the object of scientific manipulation instead of auxiliary tests of crucial inferences.

Many – perhaps most – of the great issues in science are qualitative, not quantitative, even in physics and chemistry. Equations and measurements are useful when and only when they are related to proof; but proof or disproof comes first and is in fact strongest when it is absolutely convincing without any quantitative measurement.

Or to say it another way, you can catch phenomena in a logical box or in a mathematical box. The logical box is coarse but strong. The mathematical box is fine grained but flimsy. The mathematical box is a beautiful way of wrapping up a problem, but it will not hold the phenomena unless they have been caught in a logical box to begin with.

—Platt (1964; cited in Weinberg, 1975)

As Weinberg noted, a system is not a physical thing that exists in the world independent of an observer. Rather, a system is a scientific construct to help people to understand the world. In particular, the “systems” approach was developed to help address the complex, multivariate problems characteristic of the behavioral and biological sciences. These problems were largely ignored by conventional physical sciences. As Bertalanffy (1968) noted, “Concepts like wholeness, organization, teleology, and directiveness appeared in mechanistic science to be unscientific or metaphysical” (p. 14). Yet, these constructs seem to be characteristic of the behavior of living things. General systems theory was developed in an attempt to frame a quantitative scientific theory addressing these attributes of nature. This quantitative theory is intended as a means (not an end) to a qualitative understanding of nature. This book intends to introduce some quantitative tools of control theory that might be used to build a stronger logical box for capturing the phenomena of human performance.

The term system is sometimes used in contrast to environment. The system typically refers to the phenomenon of interest (e.g., the banking system or the circulatory system), and the environment typically refers to everything else (e.g., other political and social systems, other aspects of the physiology). Again, as the opening quote from Weinberg indicated, the observer determines the dividing line between system and environment. For example, in studying chess, the system might include both players and the board. Here the phenomenon of interest might be the dynamic coupling of the two players through the board and the rules of the game. In this case, the environment would include all the social, psychological, and physical processes assumed to be peripheral to the game. Alternatively, the system might be a particular player (and the board). From this perspective, the opponent is part of the environment. Thus, the observer is not primarily interested in the motives and strategies of the opponent. The primary goal is to determine how the player of interest responds to different configurations of the board. Finally, the system might be the game of chess. That is, neither opponent is considered in the system description, which might be a listing of the rules of the game or an enumeration of all possible moves. These are not the only choices that might be made.

Another aspect of the “way of looking at the world” is how clean the break is between the system and the environment. For example, Allport (1968), in discussing the systems approach to personality, wrote:

Why Western thought makes such a razor-sharp distinction between the person and all else is an interesting problem….Shinto philosophy by contrast, regards the individual, society, and nature as forming the tripod of human existence. The individual as such does not stick out like a raw digit. He blends with nature and he blends with society. It is only the merger that can be profitably studied. (p. 347)

Western science, in general, tends to have a preference for a clean, “razor-sharp” break between the system and the environment. So, for example, physicists typically go to great lengths to isolate the systems of interest from environmental influences (e.g., vacuums and huge concrete containments). In psychology, Ebbinghaus’ choice of the nonsense syllable was an attempt to isolate human memory from the influence of the environment. As a result of his success, generations of psychologists studied human memory as a system sharply isolated from environmental factors, such as meaning and context. Descriptions of systems where the break between system and environment is sharp (i.e., where the environment is considered irrelevant) are generally referred to as closed systems.

The information-processing approach to cognition tends to break up the cognitive system into isolated components that can be studied independently from each other (e.g., sensation, perception, memory, decision making, and motor control). Each process is treated as a distinct “box” that is only loosely coupled to the other components. Laboratory tasks are designed to isolate component processes, and researchers tend to identify their research with one box or another (e.g., “I study perception” or “I study decision making”). Although the links between components are explicitly represented in information-processing models, research programs tend to focus on one component or another, and the other components tend to be treated as part of the environment. Thus, research programs have traditionally been formulated as if the component of interest (e.g., perception or decision making) is an effectively closed system. For example, each chapter in a standard cognitive psychology text is relatively independent from the other chapters. In other words, there are relatively few cross-references from one chapter to another.

However, appreciation for the coupling between system and environment is growing in all fields of science. Psychology is beginning to appreciate the importance of interactions between components within distributed cognitive systems. Also, there is a growing appreciation that the boundaries between human and environment are not so sharp. For example, researchers are beginning to realize that the sharp boundaries between human memory and environment created by the exclusive reliance on nonsense syllables was a very narrow perspective on the phenomenon of remembering. For example, Kintsch (1985) explained:

What a terrible struggle our field has had to overcome the nonsense syllable. Decades to discover the “meaningfulness” of nonsense syllables, and decades more to finally turn away from the seductions of this chimera. Instead of the simplification that Ebbinghaus had hoped for, the nonsense syllable, for generations of researchers, merely screened the central problems of memory from inspection with the methods that Ebbinghaus had bequeathed us. (p. 461)

In some respects, the growth of general systems theory reflects recognition of the rich coupling that generally exists among natural phenomena. Thus, no matter how a system is defined, there will almost never be a razor-sharp break between system and environment. The flow of matter, energy, and information from environment into systems is considered to be fundamental to the self-organization exhibited by living systems (e.g., biological, cognitive, and social systems). Descriptions of systems where the flow between system and environment are recognized as fundamental to the phenomenon are referred to as open systems.

Some open system approaches to human performance are referred to as “ecological.” These approaches tend to emphasize the dynamic interaction between humans and environments (e.g., Brunswik, 1955; Gibson, 1966, 1979). The term ecology is often used in place of environment to emphasize the relational properties that are most important to the coupling of human and environment. Von Uexküll (1957) used the term umwelt to refer to the world with respect to the functional abilities of an animal (i.e., the ecology). For example, a person confined to a wheelchair may live in the same environment as a person with normal locomotor abilities, but the functional significance of objects in that environment (e.g., stairs) will be very different. Thus, these people live in different “ecologies” or different “umwelten.” The stairway is a passage for one, but an obstacle to the other. Shelves that can be easily reached by one are impossible for the other to reach. A hallway that allows one to turn easily will be too narrow for the other. Radical ecological approaches tend to approach the Shinto philosophy in which humans as distinct systems disappear and the ecology (or umwelt) becomes the system of interest.

The study of perception and action, in particular, seems to be well suited to an open systems or ecological perspective. Action reflects a “hard” (i.e., force) coupling and perception reflects a “soft” (i.e., information) coupling between a human (or human–machine) system and its environment (Kugler & Turvey, 1987). The idea of coupling is not particularly radical. Many images of human performance systems include both forward loops (action) and feedback loops (information) to represent the coupling between human and environment. However, the language, the analytic techniques, and the experimental logic employed to study human performance often reflect a simple stimulus–response logic that greatly underestimates the richness of the coupling. In particular, causality is sometimes viewed as unidirectional (behavior is a response to a stimulus) and the creative aspects of behavior that shape the environment and that seek out stimulation are often ignored. This stimulus–response framework leads to a reductionistic approach to human performance in which perception and action are isolated (both in theories and in laboratories) as distinct stages in a linear sequence. Perception and action tend to be treated as distinct systems (and only marginally open systems).

Figure 1.1 illustrates multiple ways to look at a cognitive system. At the top, the system is treated as a “black box.” From this behaviorist perspective, the focus is on relations between stimuli and responses. Alternatively, the black box can be pictured as a sequence of information-processing stages. From this perspective, researchers attempt to describe the transfer functions for each stage. An implicit assumption of this information-processing perspective is that behavior can be understood as a concatenation of the transformations at each stage of processing. That is, the output from one stage is thought to be the input to the next stage of processing. The presence of feedback is often acknowledged as a component of the information-processing system. However, this feedback loop is typically treated as a peripheral aspect of the process and the implications of closing-the-loop are generally not reflected in the experimental logic or the computational models of information processing. One implication of closing-the-loop is that the cause–effect relation that is typically assumed between stimulus (cause) and response (effect) breaks down. In a closed-loop system, the stimulus and response are locked in a circular dynamic in which neither is clearly cause or effect. The stimuli are as much determined by the actions of the observers as the actions are determined by the stimuli. The intimacy of stimuli and response is better illustrated when the boxes are shifted so that action is to the left of perception in the diagram. Note that this change in how the processing loop is pictured does not alter the sequential relations among the stages. With respect to the logic of block diagrams, the third and fourth images in Fig. 1.1 are isomorphic. Within this circular dynamic, the boundaries among the component information-processing stages become blurred and emergent properties of the global dynamic become the focus of interest. This ecological perspective tends to focus on higher order properties of the perception–action dynamic, rather than on the local transfer functions of component stages.

This book introduces some of the quantitative and analytical techniques that have been developed to describe the coupling of perception and action. The language of control theory will help researchers to move past simple stimulus–response descriptions of behavior without falling into the trap of mentalism. Many mental constructs (e.g., percept, memory, schema, etc.) that pepper the field of cognitive psychology might be better understood as emergent properties of complex dynamic systems. These are real phenomena (e.g., perceiving, remembering, knowing) that reflect dynamic interactions between humans and ecologies. The language of control theory will play an important role in the discovery of the psychophysical basis for these emergent properties of the cognitive system. This language will provide a perspective for looking at the world in a way that will enhance our appreciation and understanding of the dynamic coupling of perception and action within natural ecologies.

FIG. 1.1. The cognitive system can be modeled as a “black box”; or it can be partitioned into information-processing stages; or it can be viewed as an abstract dynamical system in which the distinctions among stages and between actor and environment become submerged within the overall dynamic.

It is important to keep Weinberg’s view of a system as a “way of looking at the world” in mind while reading this book. Control theory can be both a metaphor for human information processing (e.g., the cybernetic hypothesis) and an analytic tool for partitioning and modeling human performance data (e.g., Bode analysis). In this respect, it is similar to other tools that are familiar to behavioral researchers (e.g., analysis of variance and the theory of signal detectability). The utility of the analytic tools does not necessarily depend on embracing a particular associated metaphor or a particular theory of behavior. For example, analysis of variance provides a useful analytic tool, independent of the value of additive factors logic for identifying stages of information processing from reaction time data. Examining performance in terms of relative operating characteristics can be useful, independent of the value of the ideal observer metaphor for signal detection. Also, it is important to appreciate that the analytic tools often provide the means for testing the limits of the metaphors. This book focuses on control theory as a tool for evaluating human performance in the context of dynamic tasks. The analytic tools of control theory provide a valuable perspective on perception and action. Further, the value of these tools greatly exceeds the value of any particular metaphor. That is, the language of control theory spans the multiple perspectives shown in Fig. 1.1.

One area of research that has benefited from the use of a control theoretic perspective is the study of “manual control,” which is typically associated with engineering psychology. This research focuses on the human’s ability to close-the-loop as the “driver,” or “pilot,” in control of a vehicle, or as an “operator” managing an industrial process. For example, there has been a great investment to develop models of the human pilot so that designers can better anticipate the stability limits of high performance aircraft. There are several excellent reviews of this area of research (e.g., Frost, 1972; Hess, 1997; Sheridan & Ferrell, 1974; Wickens, 1986). Examples from this literature are used throughout this book. A goal of this book is to provide a tutorial introduction to make the manual control literature more accessible. Many lessons learned in the study of human–machine systems can be generalized to help inform the basic understanding of human performance.

The ultimate goal of this book is not to advocate for a particular theoretical perspective, although the theoretical biases are surely apparent. It does not pus...