Rick Marken,
"RM: Just going by the words that make up the phrase “collective control” I would say the phase refers to controlled results that are produced by a collection of control systems. But there are many different ways for a collective of living control systems to produce controlled results; I describe three general ways in the “Social Control” chapter of The Study of Living Control Systems.
RM: I purposely didn’t call that chapter “Collective Control”, by the way, because I get the impression that, in PCT circles anyway, that phrase has become associated with a particular kind of group behavior which I refer to as “conflictive control” in the Social Control chapter of my book. If this is the meaning of “collective control” that Bruce has in mind then I don’t see how it applies to the kinds of behaviors Bruce is talking about. I don’t even see it as an example of control. But I really don’t know what Bruce might have in mind when he invokes “collective control”.
RM: So I throw your question back to you: What do you think is meant by the term “collective control”?"
I can’t say what it means to anyone else, including Bruce, but I can try to answer your question.
I understand a taxonomy of types of collective control, which all have in common just one property, that the actions of two or more distinct controllers all have some influence on a variable that is the same as would be observed if that variable were perceived and controlled by some single controller with a virtual reference level. Kent and I conventionally call such a collective controller a Griant Virtual Controller (GVC). The TCV can be applied to a GVC just as it can to any conventional controller controlling a perception by acting on its environment.
That’s all I really need to say to answer your question as posed, but it needs some amplification. I won’t go into my taxonomy of types of collective control, but I will try to illustrate some of what “collective control” means to me.
For one thing, it immediately poses the question: "If a GVC produces exactly the same effect as a simple controller, how can we ever tell whether some controlled variable is or is not controlled collectively without tracing the actual signal paths using some non-PCT technology to do the tracing?
My thinking about collective control is all based on Kent’s 1993 CSG demo, in which he showed that if two control systems control their perceptions of the same environmental variable to different reference values, the result on the variable is the same as though there were only one controller with a gain the sum of the two individual gains and a reference value that is the weighted (by gain) average of the two individual reference values. This is true whether the two individual controllers are in conflict (current perception are between the to reference values) or are cooperating (pulling the environmental variable in the ame direction).
Kent’s demo is just a starting point in thinking about collective control. For example, if there are three controllers arranged in a triangle, all of them trying to move an object to some desired location on a plane on which X-Y axes cam be used to describe the location of the object in X and in Y (two perceptual values), you have three reference values but only two degrees of freedom for control. That’s another case in which control may be collaborative or conflicted in either or both directions, depending on the current perceived X and Y location of the object. In this case, the three individual controllers are “members” of two orthogonal GVCs with their own (virtual) gains and reference values based on the summing and averaging of the three individual gains and reference values.
Add a fourth, fifth, … Nth individual to this second case, all of them pushing and pulling this same object toward their individually preferred X-Y reference location, and you still have only two GVCs. To an external experimenter or observer who cannot see the individual actors, it is impossible to tell that there are not two real control systems, one moving the object to a reference X positon, one moving it to a reference Y position. I guess you might call this situation two-dimensional social control, but the same analysis applies to N-dimensional social control provided N is less than the number of individuals.
These examples apply whatever dimensions the individual perceptual functions trace out in the observer’s experimenter’s space. They aren’t limited to the one-dimensional lines of Kent’s demo or the Euclidean spaces of linear perceptual functions. They are more general.
Now we come to an apparent categorical division of types of collective control into continuous and stochastic. Continous control applies when the signal values in a control loop all change smoothly. Stochastic control applies when the action output is zero apart from occasional action events. These aren’t really different, if you consider how Powers came up with his concept of “neural current” by averaging spikes from a “neural bundle” of individual neurons over a time-span long enough to be sufficiently smooth as the averaging window since “now” moves over time. The actual effects are stochastic, but the smoothing is “good enough” to allow effective overall control. Early in B:PC Powers says he initially expected to get results within 10%, but had been pleased to find that actual experimental results were much better than that.
Stochastic collective control can be distinguished from continuous collective control if the time between events affecting experimental or observed is long compared to any averaging done by the observer. Elections are prime examples of stochastic collective control. Changes of “public sentiment” obtained by frequent enough polling could be seen as continuous collective control in N-dimensions, the N dimensions being the properties being polled, such as satisfaction with the government’s performance on fishing rights.
A GVC works the same as a conventional control loop. Without identifying the control mechanism, you can’t tell from external observation whether the controlled perceptual value exists anywhere in the hidden circuitry that creates the individual controllers, or indeed whether there are any individual controllers. This inability applies just as much to the perceptual values captured in the “neural current” as to the perceptual values controlled in any loop, virtual or in the single traceable circuit of a thermostat. So consideration of collective control leads to the unanswerable question: “Are our perceptions virtual?”
Now we come to a distinct area of collective control, which in a social context we might call “common cause”. Up to now, we have been tacitly assuming that all the individual reference values are random for all dimensions collectively controlled. Often they aren’t, but are biased and are very much more similar than a random distribution would expect. They have been biased by collective control since childhood, so that a whole bunch of people perceive X as “right and proper”, whereas a different bunch of people might not care about X one way or another, and yet another bunch might consider X to be very far from their reference values.
That way lies religious conflict, which is far beyond the scope of an answer to Rick’s question: “What do you think is meant by the term 'collective control’?”