Comment on Tom Bourbon’s “Invitation to the Dance”
Just 30 years ago, Tom Bourbon published a paper called “Invitation to the Dance” [Bourbon, W. T. (1990). Invitation to the dance: Explaining the variance when control systems interact. American Behavioral Scientist , 34 (1), 95-105.]. This paper presented the first experimental evidence of collective control, as I would describe it, although Tom offered a much different interpretation of his results.
The paper begins with a concise description, well worth quoting, of how to tell from empirical evidence that control has occurred.
- The most certain evidence of control is the elimination of variability in part of the world at the expense of increased variability in the actions of the person, or any other organism, that is in control. (p. 95)
Tom’s warning, that control efforts to create order in one place in our environment will inevitably produce side effects of disorder someplace else, sounds a bit like the Second Law of Thermodynamics applied to control theory. In the PCT world we may easily lose sight of that immutable fact.
Tom’s paper describes an elegant series of experiments designed to demonstrate how we can “coordinate the movements of parts of our own bodies, or our own movements and those of another person.” The paper thus offers a PCT-based theory of how people interact with each other, and it represents the earliest attempt I know of to apply PCT to social psychology. Unfortunately, in my view, his theoretical interpretation of his results was incomplete and thus misleading, and his arguments may actually have set back the acceptance of PCT among social scientists.
In the setup for his experiment, Tom attached two joystick handles to a personal computer (this was before the days of mice). He then programed the computer display to show a cursor in the middle of the computer screen that served as a fixed target for two cursors programed to wander vertically up and down the screen on either side of the target cursor in response to random disturbances.
He connected the handle of the joystick on the right to the cursor on the right, and the handle of the joystick on the left to the cursor on the left, which created an experimental setup that was a kind of double-barreled version of the classic tracking experiments done by Bill Powers. But there was a twist.
The movements of the handle on the right were programmed to affect not only the position of the cursor on the right, but also the position of the left-hand cursor, only just half as much. Similarly, the left-hand handle affected both cursors, but its effect on the right-hand cursor was only half as big as on the left.
In the first run of the experiment, one subject was given left-hand joystick and told to keep the left-hand cursor even with the target in the middle of the screen. The subject was easily able to perform this classic tracking task. By moving the left-hand handle to control the position of that cursor, the subject, of course, disturbed the position of the right-hand cursor, but it didn’t really matter, since the subject was only concerned with controlling (his perception of) the position of the left-hand cursor, and not the right.
In the second run of the experiment (with the same random disturbances), the subject was given both handles and asked to keep both cursors even with the target. The subject performed this more complex task successfully, in spite of the fact that the motions of the cursor on the left added to the disturbances on the cursor on the right, and vice versa.
In a third run with same set of random disturbances, two subjects participated, one using the handle on the left to control the cursor on the left, and the other using the handle on the right to control the cursor on the right. Again, the subjects had no difficulty performing these tasks.
Tom then presents graphs showing the handle positions, cursor positions, and disturbance vectors for these runs, which of course resemble the classic graphs for PCT tracking experiments. Tom points out in his writeup that the graphs for runs number two and three—where the task of keeping both cursors even with the target was performed first by a single person and then by two people— look almost identical. Of course, this result was not really surprising, since the handle motions that served to keep the cursors even with the target were the same in both cases.
Tom goes on in his paper to present a computational simulation of the results of third run of the experiment, where two subjects performed the task. Using control-system models (PCT bots, you might call them) with parameters tuned to simulate the performance of the two subjects on their previous tracking runs, Tom shows that the PCT models can do an excellent job of simulating the actual experimental results.
So far, so good. Tom’s evidence shows convincingly that, whether it’s the two hands of a single individual controlling the cursor positions, or two individuals, or two PCT bots, the control agents must resist disturbances in the exactly same way in order to keep the left and right cursors aligned with the target. Furthermore, he has shown that all of those combinations of control agents are equally able to accomplish the task.
In a concluding section of the paper, where Tom offers a “general discussion” of the results of the experiments, his interpretation goes off track. Tom begins in a reasonable way by discounting the prevalent psychological theory that the brain uses perceptual inputs to calculate in advance the movements necessary to keep the cursors even with the target, and then commands the hands and arms to move accordingly. As he argues, “commands computed in advance cannot anticipate the effects of unpredictable continuous disturbances like those used here” (p.103).
The problems with his explanation begin In the next two paragraphs, where he offers a critique of the theories of social scientists:
To explain coordination between the actions of two people, many social theorists invoke hypothetical goals that are allegedly “shared” by the people. Others hypothesize norms, standards, or guidelines that exist independently of people and subsume them in larger systems (families, armies, and the like), where each person functions as a component. But people are complete control systems, not isolated components such as input functions or comparators. Typically, theories that describe people as “components” in larger control systems cannot predict continuous coordinated actions.
In the simulations here, the only connections between the models were in the environment. When, as was the case here, two systems affect the same variables but to differing degrees, each system reduces the interaction to a net disturbance and then cancels the effect of that disturbance. The result? In a complex world, simple systems achieve coordinated control. (p. 103)
Tom’s theorizing is off base, not because his PCT-based arguments are incorrect, but because his PCT explanation is incomplete. He hasn’t thought through the experimental situation carefully enough to see all of the ways PCT might apply to it. Tom’s problem here is that his theorizing is strictly one-dimensional, but the experimental situation he set up involves two different levels of perception at the same time, and the social situations to which he wants to apply his results always involve perceptions at more than one level.
In the discussions 25 years ago on CSGnet about whether to use the acronym PCT for the theory that Bill Powers had presented in B:CP, Bill said his preferred label was HPCT, or hierarchical perceptual control theory, to emphasize the hierarchical nature of human perception and action, the fact that that humans control perceptions on many different perceptual levels simultaneously. In the interpretations he offers in this paper, however, Tom has completely ignored the hierarchical aspect of PCT.
It isn’t as if Tom is unaware of HPCT. In the concluding section of his paper, he makes this argument:
- With its elegant simplicity and effectiveness, the control theory model stands in sharp contrast to thecomplexity and nonspecificity of most theories of coordination. Control theory offers the possibility of using the same few principles to explain coordination at every level, from movements of parts of our own bodies to interations [sic] like those in simple tracking tasks, in infant-parent dyads, in social gatherings, in marriages, and on the job. (p. 104)
Tom is arguing here that interactions occurring at any perceptual level can be adequately explained by the one-level PCT model that he offers in this paper. However, if we look carefully at his model, we can see how his model is even too simplistic for the relatively simple experimental situation he set up.
A one-level model of a control agent controlling its perception of a single environmental variable works fine for describing behavior in classic tracking experiments. But in Tom’s experimental setup there are three perceptual variables: the relationship of the left-hand cursor to the target, the relationship of the right-hand cursor to the target, and the simultaneous relationship of both cursors to the target.
This third perception is of higher order than the other two, because this perception is constructed hierarchically by combining the other two into a single more complex configuration. Tom completely ignores this third, higher-level variable in his diagrams of the experimental setup and in his simulation modelling. All the same, this phantom variable is what gives his paper its rhetorical punch.
The subject who uses two hands to do the experiment must be able to control this more complex variable, the perception of keeping both cursors even with the target simultaneously. Control of this variable is implicit in the task that Tom has presented to this subject, and the subject’s successful completion of the task is proof that the subject was able to perceive and control this higher-level perception, as well as the lower-level components of the perception, that each cursor separately must stay on target.
In the third run of the experiment, with two different subjects at the computer, both subjects presumably looked at the same computer screen, but as long as they focused their attention on the cursor that corresponded to their own handle, they could complete the assigned task without any problem. They didn’t have to pay attention to the third, more complex variable, and it might have confused them if they had.
In fact, another way for Tom to have set up this third phase of the experiment would have been to give each subject a separate computer with a monitor that showed only the cursor corresponding to their own handle and the target, as long as there was a connection between the computers that then transferred from one computer to the other the disturbances created by the movements of the other handles. The experimental results would have been exactly the same. It this case it was Tom who actually controlled the third variable, the one that he described as coordination of their movements, by the way he set up the experiment and his instructions to the subjects to do their tasks simultaneously.
In the computer simulation of the results, Tom was again the one who controlled the third variable, the one that combined the two lower-level perceptions of each cursor staying on target. The PCT bots in the simulation acted like Tom’s slaves, since he specified the perceptions they needed to control and also specified the references to use in controlling those perceptions. As one-dimensional control systems, the bots obviously couldn’t have comprehended the concept of keeping both cursors on target simultaneously, but it wasn’t any problem. Tom kept this higher-level variable in control, since he was the one who could monitor the coordinated movements of the cursors to make sure that the experiment worked as planned. If anything had gone wrong, Tom could have fixed the problem by reprogramming the computer.
But take note. For Tom’s article to be rhetorically effective—for readers of the article to buy his argument—his readers, too, must be able to control (in their imagination) this higher-level variable based on the combination of the other two variables. They have to be able to imagine the sight of the two cursors staying on target simultaneously. In fact, for his argument to be persuasive, Tom and his readers must be able to share this perception. In my terms, Tom and his readers have to be able to control this higher-level variable collectively.
- In the second run of Tom’s experiment, where a single subject takes a handle in each hand, the subject is actually controlling three perceptions, not two: the perception of the right-hand cursor staying on target, the perception of the left-hand cursor staying on target, and the perception of both cursors staying on target at the same time, a higher-level perception constructed from the other two. Subjects are able to complete this complex task because the hierarchical organization of the perceptual control systems in their brain allows them to control different perceptions simultaneously at more than one level of perception. According to HPCT, the hierarchical organization of the brain implies that control of a perception at a high level depends on the simultaneous control of the cascade of lower perceptions from which it has been constructed.
- In the third run of the experiment, where two subjects were at the computer, each of the two subjects had to control two perceptions (at least): the perception of keeping their own cursor on the target, and the higher-level perception of cooperating with the experimenter by following Tom’s directions to complete the task. This is an example of my second general form of collective control: where two or more people cooperate by collectively controlling a single focal perception at a given perceptual level while independently controlling different perceptions at lower levels.
- For Tom’s argument to his readers to be persuasive, he and his readers must engage in my third general form of collective control: parallel independent control of “the same” perception or set of perceptions by different people who share a set of common references for their own independent control of those perceptions. In this case, Tom and his readers, who may be far separated in location and even across time, must be able to control similar perceptions what it means for two cursors to stay on target at the same time, in order for the readers to accept the argument that Tom’s model explains how “coordination of movements” takes place.
Some additional reflections: Tom argues, in effect, that coordination of movements might somehow spontaneously arise between two control systems controlling only their own independent perceptions without reference to each other. To me, this argument seems as silly as imagining that the molecules of a gas might all spontaneously start to go in the same direction at the same speed.
Social coordination does not, and cannot, happen by accident, and Tom’s argument that it could happen that way would seem absurd to other social scientists, too. No wonder that Tom’s paper was not taken as a great conceptual breakthrough by members of the social science community. Tom may have been arguing against the notion that society is a control system at the macro level and that people are simply components of that overall control system, but that theory was something advanced by some people interested in PCT in the 1990’s, not anything taken seriously by the social scientists of my acquaintance.
From the social scientific perspective, cooperation in social situations happens when people intend to cooperate. In PCT terms, we would say that the parties control a high-level perception of cooperation with the other parties involved, in addition to any lower-level perceptions they control to play their role in the overall cooperation. Which, in my book, represents the second general form of collective control.
Furthermore, our communication between each other hinges on the third main form of collective control. For communication to take place—for you as reader, for instance, to understand the arguments I am rendering symbolically by putting words on a computer screen, or for readers of Tom’s paper to have understood his arguments—the parties to the communication must be able to share similar perceptions, or in other words, to control those perceptions collectively. Now, of course, each person’s perceptions are unique, so the sharing cannot be perfect. Communication is never perfect, but in our semi-successful social life it has ordinarily proved good enough for practical purposes.
Thus, you and I communicate when your perceptions of these words as reader and my perceptions as author are similar enough that we don’t end up taking physical actions in control of our own perceptions that prevent the other person from continuing to control their independent but similar perceptions. As my simulations of collective control have shown, our perceptions need not be exactly the same for our joint actions to stabilize our shared physical environment more effectively than either of us could do on our own, and thus to enhance both party’s independent control of the relevant perceptions.
To sum up, my argument is that one-dimensional analyses of social life, like Tom’s analysis in his paper, are by definition insufficient and that one form of collective control or another is involved in everything that we do as social animals.