Ramifications of collective control

Fundamentals System Constraints
1

Not everyone is interested in exploring the limitations and possibilities of control theory. This is a basis, I think, of the long-running disagreements between Rick and Martin, exemplified by a failure to come to mutual terms beginning with post #4 of the Autonomy and evolution topic. As I understand them,

  • Rick’s interest is in experimental work, especially that form of experiment done by building and testing generative PCT models of observed behavior, building on work that he did with Bill.
  • Martin’s interest is in the mathematical and biological foundations of the theory, PCT, and the consequent constraints and possibilities of what control systems are capable of, individually and collectively—“more about PCT than Bill put in his writings”—and “to suggest a variety of lines of research in different domains in which PCT could be a fundamental premise” (PPC I:19).

This is not to say that Rick is disinterested in theory, but for his practice testing the canonical formulation is quite enough. And of course Martin has deep experience in designing and conducting PCT psychological experiments, but he has his hands full with the work that engages him now. We each devote the time available to us preferentially.

In the cited discussion, Martin is saying that the only way to understand collective control is to play with its possibilities and find out that there are some unexpected properties of sets of autonomous control systems mutually influenced through a shared environment. For example:

The premise of the second paragraph may not be correct, but there’s nothing wrong with doing experimental and modeling work within the capacious scope laid out in Bill’s writings.

In that discussion, Rick is saying he won’t ‘waste his time’ building any models unless Martin specifies some particular behavior to be modeled. He levels the same criticism against existing models of collective control, for example:

Both preferences are essential to a science. They rely on each other, or should.

I have opened this topic for two reasons; first, because the extension of PCT into collective control requires the experimental side to understand results in the theoretical side and to test them. Both sides may participate in devising ways to test, but historically that has fallen almost always to the experimentalists.

Secondly, investigations into the theory as a mathematical and conceptual object has been limited, as evidenced perhaps by there being no posts in this category about system constraints (and possibilities) until now.

1 Like
2

This is a very useful post. Thank you.

If I might add, observation and theory without experiment can also advance a science. Cosmology is an extreme example, starting with precise observations of the apparent movements of the planets in the sky, leading to the work of Ptolemy and then Kepler, backed by the theoretical work of Newton, all the way to Einstein a century ago, and continuing today with such observations as the imaging of a black hole.

Experiment is better, more reliable and precise, if the subject matter is open to action. In PPC, it is relatively easy to experiment on ideas relating to tracking, but not on the movements of crowds of real people, in which we have only observation matching (or not) theoretically anticipated results from simulations.

Collective control is of this nature. Simulation of simpler collective control effects is relatively easy, but to simulate what seem to me to be at least a two-level structure of collective control in the Jan 6 attack on the US Capitol would, I think, be rather difficult.

Overall, tracking the maturing organism in the development of control structures, some of which involve interactions with other living control systems of all kinds, is something on which many experimental possibilities are at least borderline unethical (such as depriving kittens of vision for a period of their growth).

Observation that does not influence the observed is possible, though often very hard to arrange, which leaves us with the kind of theoretical science used by astronomers. Astronomers can extrapolate from the experimental work of nuclear and other physicists on the assumption that what works on one (human-observable) scale of space and time will work on a scale many orders of magnitude greater. And then they can see how precisely their extrapolations work on observable quantities.

Most experimental work in PCT requires extrapolation to be useful, and to test the validity of extrapolation requires observation of situations in which experiments are either limited in scope or unethical. Kent, in particular, has done this kind of work in social applications of PCT (I have consulted him on much of my social extrapolation in PPC, but never should he be charged with my mistakes in this area).

Anyway, that’s all a long-winded way of saying that I see experiment, observation, and analysis as an entire self-supporting ecology of interactions, stable until a real disruptive event such as an an unexpected observational corroboration of an unpopular theory. Different people with different kinds of expertise test different parts of this sprawling network, and that is (in my mind) as it should be.

3

The basis of my disagreements with Martin, and with everyone else with whom I disagree about PCT and, of course, that includes you, is that I think he is importantly wrong about whatever it is we are disagreeing about.

Best, Rick

4

Many years (decades, actually) ago, you expressed on CSGnet the opinion that if I said anything substantive about PCT, it was ipso facto wrong. Nothing much seems to have changed in the intervening years. If I make a statement relating to PCT it is “importantly wrong”.

So before we begin a discussion, the end is already being controlled for with the reference value that you are right and I am wrong, so “forge ahead and damn the torpedoes” put in your way by mathematics or other sciences such as physics. You won the argument before it was begun, but in doing so, how often has that actually advanced the science of PCT as understood by interested third parties?

Your experiments are where I think you shine in advancing the understanding of PCT as a science rather than as a literary gospel. That’s why in some of our arguments, such as recently on collective control, I suggest experiments to illustrate a point.

5

Given your record of putting words in my mouth I doubt that I said anything like that. But if I did, I surely said it out of exasperation. I don’t think everything substantive you say about PCT is necessarily wrong. I just disagree with you when you seem to be wrong, which, I agree, is fairly often. But I don’t think you are wrong about PCT much more often than anyone else here.

I describe several studies of collective control – the study of the controlling done by a collection of control systems – in my book The Study of Living Control Systems. I suggest that you read the chapter on Social Control (Chapter 7) to see that kind of research (some experimental, some observational, and all involving the fitting of models to data) that I think advances the science of PCT with respect to collective control.

6

What, in terms of collective control, do you make of the 2021.01.06 attack on the US Capitol Building?

7

How about providing a synopsis here. I do believe I have the book in question, but I can’t find it, so I cannot continue the discussion without the cost either of an extended search on my bookshelves, or in dollars to buy it again.

8

I think it was a crowd of pawns who were conned into controlling for storming the Capitol by a sleazebag who was clearly controlling for getting that to happen because he wanted to remain President. The con was the lie that the election was stolen and that pawns were convinced that the certification of the results had to be halted by entering the Capitol and killing the vice president, if necessary. Individuals in the crown were probably controlling for different many different things when they were incited (control by deceit) to go to the Capitol; some wanted to enter the capitol, some probably just wanted to “peacefully” protest, etc. I think the a large proportion of those involved in the storming didn’t come there to storm but ended up doing so because the were imitating others – I think the human inclination to control for imitating others, which is adaptive in infants, is the basis for what is seen as “mob behavior”.

Of course, there was also conflict involved because there were police there trying to prevent people from getting into the House chamber. So I suppose there was a virtual controlled variable involved – something like “distance of stormers from from the House chamber”. But this variable wasn’t very stable during the storming and, thankfully, ended up in the state “far away from the Capitol”.

9

I found “The Study of Living Control Systems” and read Chapter 7. From my current viewpoint, I think it covers one form of collective control very nicely — perhaps two forms. But I think where it is lacking is in filing to note that the virtual reference and virtual perceptions allow the collective control loop to be treated as though it was a “real” control loop.

In fact, any control loop that uses the Powers concept of a neural current is, by definition, a collective control loop, and the perceptual values created by the analyst from the firing rates of multiple neurons are virtual values. They work similarly to a stochastic collective controller (discussed a while back on CSGnet and extended in PPC III.1.2 and III.1.3.). each nerve firing instance being the equivalent of a pebble-throw in the gedanken experiment.

10

In everyday terms I have to agree with this, but I was actually looking for a PCT analysis of the varieties of collective control involved. For example, I see one main overall collectively controlled variable, the state of the certification of the electoral vote, plus some secondary controlled variables, such as the physical health of Mike Pence and Nancy Pelosi. Only the first had a reference value provided by Trump.

11

In section 7.1.7 of Chapter 7, which deals with conflictual control, I discuss the concept of a virtual reference state and give reasons why a variable held in such a reference state – a virtual controlled variable – shouldn’t be treated as though it were a real controlled variable – a variable controlled in a real control loop. One reason is because there is no control in the dead zone of the virtually controlled variable – the zone where the that variable is far from each system’s reference (Powers, 1973, p. 255). Disturbances to the virtually controlled variable are completely effective in this zone, showing that the variable is not really under control.

Another reason is that the parties to the conflict that results in the appearance of a virtually controlled variable are not themselves in control; for all parties to the conflict, the “virtually controlled variable” is, on average, quite far from their reference specification for it.

As Dylan said to the guy who yelled “Judas” when he went electric, “I don’t belieeeve you”.

Best, Rick

12

Actually, Kent’s 1993 demo showed that disturbances to the virtually controlled variable were resisted as though by a virtual controller with a reference value a weighted average of the two “real” controllers and a loop gain the sum of the two “real” loop gains. You are correct that neither of the two “real” controllers can influence the value of the variable, but the effect of an added disturbance is treated exactly as it would be if the virtual controller was the only controller of that variable. No test of which I am aware could distinguish a “real” from a virtual controller in this or analogous situations.

Exactly. That’s why one has to accept the virtual controller just as though it were a real one.

Your last line simply says you don’t believe my analytic statement, with no reason given. Not very helpful.

13

Readers of B:CP in its most readily available form will be puzzled that turning to p. 255 lands them in the middle of the chapter on emotion, with no mention of anything like this. For readers who do not have ready access to the first edition of B:CP, it’s page 267 in the 2005 revised edition.

The subject at that place is conflict. Conflict is the limiting case in the spectrum of forms of collective control. You persist in identifying collective control with its limiting case and generalizing the limitations of outright conflict as though they are characteristics of all forms of collective control.

This ‘dead zone’ phenomenon is predicted to manifest when two control systems in conflict are each producing their maximum output. This is a limiting case of that limiting case.

Examples in nature are scarce and peculiar. Equally matched arm wrestlers (or teams at tug of war) do not provide evidence of a dead zone. They would hardly be insensible of a third influence on the CV.

Higher-level systems intervene. As Bill writes on the preceding page,

When two people arrive simultaneously at opposite sides of a swinging door they do not simply stop as two rocks rolling against each other would stop. The efforts they employ suddenly rise to ten, twenty, fifty times the amount normally required to open the door. Muscles can be pulled loose from their attachments by such an encounter before higher- order systems can alter lower-order reference levels for the position of the door.

In the past, some of us have discussed this as resolution of conflict. Perhaps problem-solving routines at the higher level. These can act before any actual conflict commences, as may be seen by making it a glass door or an open doorway. But rather than problem-solving, those higher-level systems are engaged in collective control of a well-recognized variable. Whether or not a swinging door blocks the view, “doorway” is a collectively controlled variable, similar to “hallway”, “road”, and “lane of traffic on the road.”

This is in general not true of collectively controlled variables.

I think that you will believe that I am wrong. Maybe I am not importantly wrong. Is it important enough that you will make the effort to become familiar with the varieties of collective control beyond the limiting case of simple conflict?

14

To err is human, and I believe I am human. But when I am wrong, I prefer to have my error explained, rather than simply being told that I am wrong without being told in what way or how the error night be corrected. I try to give others the same courtesy of explaining in what way they are wrong and, if I can, how to fix the problem. (Shouldn’t this observation be in the “Standards of conduct” category rather than here?)

15

I just set up a little simulation of two control systems in conflict. It was set up in a spreadsheet so there might have been odd computational issues so it would be nice if you could send me a pointer to Kent’s 1993 paper that shows that you can’t distinguish a real from a virtual controller. I would like to look at it because I believe I have found a way to distinguish a virtually from an actually controlled variable.

The method is based on the concept of a dead zone, which implies that control of a virtual controlled variable should get better as the amplitude of the disturbance increases. I tested this by measuring the ratio of variance of the controlled variable (CV) to that of the disturbance (D) – Var(CV)/Var(D) – for two different disturbance amplitudes when the CV was controlled by a single controller (it was a “real” controlled variable) versus when it was virtually controlled by two systems in conflict.

All systems had the same gain and slowing; but when in conflict they had different references for the CV. My finding was that Var(CV)/Var(D) was the same for the high and low amplitude disturbance for the “real” CV and much lower (showing better control) for the high compared to the low amplitude disturbance for the virtual CV. So this could be a way to distinguish real from virtual CVs but more careful modeling should be done. Maybe Kent has already done it.

Whether that’s true of not, it sure doesn’t help the people in the collective for whom that variable is “out of control”; they are suffering constant error while you enjoy watching you virtually controlled variable behave like a real controlled variable.

I think the main problem is that each perceptual-signal-carrying neuron in your presumed array of neurons is influenced by the same output. So I don’t see the PCT control loop as being analogous to the situation where a virtual reference state emerges from the actions of a collective of conflicted controllers.

16

That’s because collective control is persistently identified with the existence of variables being kept in virtual reference states. Those are phenomena that emerge from interpersonal conflict and the way they emerge is described in Kent McClelland’s work on modeling conflict between collectives of controllers.

The “dead zone” exists whenever there is a variable being kept in a virtual reference state by a collective of controllers who are in conflict. Since the phenomenon of a virtual reference state (and it is a phenomenon, just a rare and transitory one) seems to be central to what you are talking about when you talk about collective control, I figure that the dead zone, limiting case or not, must also occupy center stage in your notion of collective control.

You betcha. That’s because virtual reference states are scarce and peculiar in nature.

Again, right on.

If what you say is true, then you must be referring to a kind of collective control that doesn’t involve the emergence of virtual reference states. So than I would have to ask why you are always pointing me to Kent’s work as being the source for learning about collective control. Kent’s work is all about virtual reference states, examples of which, as you seem to agree, are scarce and peculiar in nature.

I am familiar with several varieties of collective control, including the rare and peculiar version involving the emergence of virtual reference states, which I describe in Chapter 7 of The Study of Living Control Systems. What am I missing?

17

That is how collective control was demonstrated in 1993, but research in collective control has advanced in the ensuing 29 years.

Collective control explains stable social conventions and socially recognized artifacts representing them, such as crosswalks, doorways, money, and phonemic contrasts. In your view, limited as it is to Kent’s 1993 demonstration of the simplest case of collective control giving rise to the appearance of a virtual controller maintaining a variable in a virtual reference state, every such cultural and social phenomenon is a direct consequence of conflict between those who include them in their control of other variables (i.e. driving to the grocery store without being arrested for a lanes violation—at which point the cop would be controlling the conventional significance of lines of paint on the road as means of controlling her own rather different perceptions).

Either that or PCT cannot explain social and cultural phenomena beyond those simplest which emerge from conflict.

Reductio ad absurdum. Your move.

18

Hmm, just very quick comments to the end of Rick’s message:

Whether that’s true of not, it sure doesn’t help the people in the collective for whom that variable is “out of control”; they are suffering constant error while you enjoy watching you virtually controlled variable behave like a real controlled variable.

Yes it can help them at least to understand why their situation is like it is. I believe it is a human condition to be suffering constant errors because the political decisions, the systems of our work places and other institutions, and the opinions and tastes of our neighbors are seldom just like we would like them to be. I don’t believe science could ever do much more.

I think the main problem is that each perceptual-signal-carrying neuron in your presumed array of neurons is influenced by the same output. So I don’t see the PCT control loop as being analogous to the situation where a virtual reference state emerges from the actions of a collective of conflicted controllers.

Seems like you were captured by the paper and computer models where the input function is neatly in a box and even though there comes many arrows to the box, only one arrow leaves it. But if we think that every signal channel is a bundle or an array of neurons, and if comparators and other functions are built from synapses, and every neuron has its own synapses then it is quite improbable that there were just one node or crossing of single neurons. Rather both the signals and functions are throughout stochastic collective processes through the hierarchy.

Eetu

19

Here’s a short passage from the introduction to collective control in PPC Volume III.

[Edited 2022/12/05 by Bruce to include the figure.]

[begin extract from PPC] The same applies if many controllers rather than just two control their perceptions of the same environmental variable or variables whose control necessarily influences a common variable. Since we usually take the Analyst’s viewpoint, we give this environmental variable the name “Collective Corresponding Environmental Variable” or CCEV, to distinguish it from the CEVs of the individual perceptions. No matter how many individual controllers contribute to the GVC, each of them controls only its own perception of its own CEV.
When we are concerned only with a single dimension of variation, the CCEV is identical in effect to any the individual CEVs, so one might ask what is the point of giving it a different name. The answer is that very seldom, if ever, do any two individuals actually have identical perceptual functions. If one person is controlling a perception p=x+y, another looking at the same object from a slightly different angle might be controlling p=1.2x+0.8y. These are not the same, but if someone used the TCV by disturbing the object in different directions, that person might easily conclude that 1.1x+0.9y was a controlled perception, even though neither of the participants were actually controlling that (Figure III.1.1). Only the Analyst who could see what both were individually controlling would be able to detect that neither was controlling a perception of exactly 1.1x+0.9y.
III.1.1

When there are more participants in the collective control and more dimensions of variation, the independent nature of the CCEV as distinct from any individual CEV becomes more evident, an important consideration when we come to topics such as the internal dynamics of political parties in Volume IV.

20

Again, discourse censored most of my email. I added the missing part to http://discourse.iapct.org/t/ramifications-of-collective-control/15993/18 and below.

Whether that’s true of not, it sure doesn’t help the people in the collective for whom that variable is “out of control”; they are suffering constant error while you enjoy watching you virtually controlled variable behave like a real controlled variable.

Yes it can help them at least to understand why their situation is like it is. I believe it is a human condition to be suffering constant errors because the political decisions, the systems of our work places and other institutions, and the opinions and tastes of our neighbors are seldom just like we would like them to be. I don’t believe science could ever do much more.

I think the main problem is that each perceptual-signal-carrying neuron in your presumed array of neurons is influenced by the same output. So I don’t see the PCT control loop as being analogous to the situation where a virtual reference state emerges from the actions of a collective of conflicted controllers.

Seems like you were captured by the paper and computer models where the input function is neatly in a box and even though there comes many arrows to the box, only one arrow leaves it. But if we think that every signal channel is a bundle or an array of neurons, and if comparators and other functions are built from synapses, and every neuron has its own synapses then it is quite improbable that there were just one node or crossing of single neurons. Rather both the signals and functions are throughout stochastic collective processes through the hierarchy.

Eetu