Phenomena Phirst Redux

Fundamentals
1

Warren Mansell suggested that I post this offline Q & A between Bruce Nevin and myself so here it is. Warren suggested the topic title since that seems to be the main theme of our interaction. This topic of Phenomena Phirst has come up years ago on CSGNet but Warren (and I) think it’s important so here we go again. This comes from an email exchange and I’ve reformatted it slightly for Discourse. ========================================
Hi Bruce

Great questions!

On Sat, Nov 23, 2024 at 6:36 AM Bruce Nevin <bnhpct@gmail.com> wrote:

BN: Hm. Where did Martin say that information is causative?

RM: I don’t know that he ever said that explicitly but it was certainly implied. Martin thought that the disturbance had a detectable effect on the controlled variable that provided the information that allowed a person to act so as to compensate nearly perfectly for the disturbance. So, in the compensatory tracking task, the nearly perfect mirroring of the apparently invisible disturbance occurred, according to Martin, because the effect of the disturbance was visible in the derivative of the movement of the controlled variable. That’s the information that lets the person produce output (move the mouse) in a way that compensates for the effect of the disturbance.

RM: Of course, it doesn’t work that way – the position of the CV is at every instant during a tracking task a simultaneous result of output (O) and disturbance (D). So the derivative of the CV – dCV/dt – is always the derivative of O + D, not just D. The way it works is simply that the closed loop acts to keep the error, r - p = 0 and, as a side effect, O = -D. But Martin was pretty committed to the idea that output was based on (if not caused by) input (information about the disturbance).

BN: Do you have measurements of the perceptual signal represented by the theoretical quantity p to demonstrate an exact equivalence to the environmental measurement (the experimenter’s perception) q.i ?

RM: No, but maybe Henry will soon! The fact is that there has to be a nearly exact equivalence of the actual neural perceptual signal to the aspect of the environment under control – the controlled variable, q.i – or that perceptual signal is not the analog of q.i. For example, in a simple compensatory tracking task, the aspect of the environment that is controlled, CV, is the distance between cursor, C, and target, T; so CV = C - T. In theory this control is achieved via control of a perceptual signal, p, that is an analog of C-T. So I would expect to find an afferent neuron – or neuron bundle? – with a (average?) firing rate that is at least a monotonic function of the CV. Of course, this firing rate will be a bit noisy but the noise should be less than the noisiness of the observed control itself. For example, since we are getting correlations on the order of .99 between actual and model output, O, in a control task, I would expect to find correlations on the order of .99 for the correlation between the C-T and the neural perceptual signal that is thought to correspond to p in the model. This would actually be a possible way to determine whether or not the neural signal being observed does, indeed, correspond to the perceptual signal, p, that is the analog of the CV.

RM: By the way, it’s important to determine the correlation between neural signal and CV when the CV is NOT under control since there is very little variability in the CV when it is under control. So for the tracking task, validating a neural signal as the analog of the CV would involve looking for the correlation between that signal and variations in C - T over a reasonably wide range – like the range of disturbance-produced variation in the CV if it were not under control.

This is from you next email, Bruce:

RM: PCT started with the OBSERVATION of the FACT OF CONTROL AS SEEN IN THE BEHAVIOR OF LIVING THINGS.
BN: This is historically false. Bill learned control theory first, in the Navy and then in subsequent study.

RM: Yes, it’s true that Bill learned control theory before he developed PCT. But it was not Bill’s knowledge of control theory that led him to develop PCT. After all, control theory was being applied in psychology well before Bill developed PCT. Control theory was being applied in human factors psychology since at least 1947 and in cybernetics since at least 1948. But all these applications got it wrong because they based their application of control theory on a behavioral illusion – that inputs cause output. As Powers explains in his 1978 Psych Review paper, earlier applications of control theory failed to see that behavior itself was a control process. Powers’ familiarity with control theory and, especially, with building devices that control, made it possible for him to OBSERVE the fact that BEHAVIOR IS CONTROL. It also didn’t hurt that he was a physicist so he could see that the consistency of behavior was a remarkable achievement.

RM: So it was nice that Bill had learned control theory before developing PCT. But what actually happened is that Bill’s remarkable OBSERVATION of the PHENOMENON OF CONTROL as seen in the BEHAVIOR of living things led to the correct application of a theory with which he was already familiar as an explanation of that phenomenon. Phenomena phirst!

RM: By the way, I’m watching a nice PBS documentary on Leonardo da Vinci. It turns out he was a big “phenomena first” guy, just like my other scientific heroes: Galileo, Faraday, & Powers.

2

The relationship between theory and observation of phenomena is reciprocal, and each informs the other. It is doubtful that Bill would have invented control theory had he not been previously familiar with it. Control theory had been developed first for systems for which reference values are set from outside the system, as in our familiar examples of the thermostat and cruise control, or as in gunnery systems in the navy.

Theory and observation have a reciprocal relation and each informs the other. The phenomenon of control Bill had already observed, and he found the existing control-theoretic explanations of the phenomenon dissatisfying. From what he wrote and said I think an important part was in applying it to his own experience. (He ridiculed conventional psychologists for applying their notions to everyone but themselves.) Certainly the development of the perceptual hierarchy and much of the discussion in B:CP and even in the earlier joint articles has a strong basis in subjective phenomenological observation. Observation led Bill to modify the existing statements of control theory to postulate a hierarchy of cascading control with internally generated reference values, rather than a ‘plant’ with externally set reference values.

In the military, in business, in academia, and in families, people are familiar with the phenomenon of some people telling other people what to do, and in these systems of hierarchical authority the subordinates often do what their superiors tell them to do so as to avoid adverse consequences or receive preferred benefits. Like it or not, this is an observable phenomenon. Behaviorists had operationalized this general phenomenon. It is not surprising that Weiner, Ashby, and others left unquestioned the existing approach to control in which the setpoints are adjusted from outside the system.

I say like it or not, and there are many indications that Bill did not like these hierarchies of authority at all. He referred to managers as ‘bloodsuckers’. He abandoned a psych Ph.D. program rather than torture rats long enough to get the degree and then develop PCT. (I have had my own share of this je m’en fou-tism .) His challenges to efforts by Tom Bourbon and Ed Ford to work PCT into a public school system reduced to a rejection of the very notion of children being required to be obedient to a system of rules. If the child were playing chess and the teacher said “I see you have chosen to take the knight rather than keep your pawn controlling the center” he would not have objected.

In B:CP, Bill nicely describes how a military parade unit carries out synchronized movements according to the leader’s commands, and how members of a chorus produce a pitch designated by the conductor. Why do they set their reference values accordingly? How else might a person’s autonomous reference values be influenced? Not a major concern in PCT. It was important instead to show that the influences of rewards and punishments do not provide a theory of behavior, whence the decades-long battle against linear causation in behaviorism and in its ‘mediated by cognitive processing’ successors. Such influence nonetheless is a phenomenon requiring explanation. Counter-control provides a start at some aspects of collective control. Investigation of collective control phenomena further informs the theory.

3

Yes. “Phenomena Phirst” just means that this reciprocity starts with observation. Based on observation you come up with a theory. Then you test the predictions of the theory against observation, which may require changes to the theory. The changed theory is further tested against observation. So observation is an intrinsic and essential part of this reciprocity. Indeed, I think it’s the biggest part since once you have a good theory, such as Newton’s “laws” of motion, most of the scientific effort is spent on testing and/or applying those theories.

Newton’s theory held up for over 200 years until observations were made (such as the measured velocity of light being the same regardless of the velocity of the measurer) that were a problem for the theory and a clever German fellow came along and showed an elegant theoretical way to account for those puzzling observations as well as all those that were consistent with Newton’s laws.

In the hands of Martin Taylor (and his acolytes), Powers’ theory has been added to or changed over and over again despite the fact that there have been no observations made that required these additions or changes. That is why Martin’s work was not my cup of tea.

4

There are no theory-free observations in science, and in particular Bill did not invent PCT ex nihilo based solely upon the observation of the phenomenon of control.

In the reciprocal relationship between theory and observation, observation comes ‘first’ only in the ultimate sense that nature is the final arbiter. Observation of phenomena is epistemically primary, but not always temporally prior. A confounding of priority and precedence muddles what you intend to be saying.

In the work of developing an accurate and useful account, theory and observation are equally present, but the point of view shifts. To the immediate point, there are no theory-free observations. Bill’s recognition of the phenomenon of control was enabled and informed by his prior knowledge of control theory.

There are at least two ways in which theory comes ‘first’.

  1. Identify predictions of the theory. Then test whether or not observations confirm or refute them. The tests against nature are further observations which are epistemically primary but temporally follow after the theoretical considerations.

  2. Is the theory consistent with what is verified in other sciences? Especially the physical sciences, physics and chemistry. These theories are both temporally and epistemically prior to PCT, unless and until PCT proposes revisions to physics or chemistry. As it applies explicitly to living things, biology, physiology, neurophysiology, biodynamics, and kindred fields must also be taken as both temporally and epistemically prior to PCT. They are more vulnerable to influence from PCT, if workers in those fields are open to entertaining ‘advice’.

Martin was involved in both these areas. In (1) he did more of identifying predictions than of testing them in recent years. I think probably no longer having access to the DCIEM lab was a factor. In (2), PPC places PCT in the broader context of the physical sciences and relevant mathematics. Just as PCT makes predictions that can be tested, these wider theoretical considerations make predictions which are subject to test, and much of PPC is concerned with this.

The scope of PCT extends beyond the walled garden in which only the mathematics of a single (elementary or hierarchical) control loop are relevant.

5

Yes, that’s a good way to put it.

I don’t think it makes much sense to call these examples of “theory first” but they are certainly essential features of hte roles of theory in science.

Predictions from theory are predictions of what will be observed. And these predictions should be very clear about what will be observed, and how these observations map to variables in the theory. I don’t recall Martin making any such predictions.

You bet. It would be great if Martin had done some of that extending!