VS: VS: Controling observables (was Re:... long live William T. Powers)

[Eetu Pikkarainen 2017-06-15 3]

Thanks Martin, this was very informative. Some comments though, below…¦

[Martin Taylor 2017.]

I have had to rethink my earlier “quibble” about the use of the word “stabilize” [Martin Taylor 2017.], because I realize that stabilization is indeed the core of any kind of control process.
The question is the stabilization of what?

Here are the two canonical diagrams, one for engineering control and one for perceptual control. Functionally, they are the same except for the omission of an analogue to the PCT “Perceptual Function” in the engineering diagram. Even thought they are functionally
identical, they are usually interpreted differently, besides which they confusingly use the same labels for different parts of the loop.


We don’t need to explore the similarities and differences between these diagrams. All we need to do is ask just what is stabilized. In the Engineering diagram it is the difference between the “input” and the “feedback”, whereas in the PCT diagram it is the
difference between the “reference” and the “perception”. In both cases, that difference is called “error”. If the feedback loop gain is negative and the gain of the following element (“Plant” or “Output Function”) is appreciably greater than that of the rest
of the loop, then the circuit will tend to reduce the absolute magnitude of the error, stabilizing it near zero.

EP: That is certainly true and important. But if the concept is “stabilizing near zero� isn’t then the proper term for it “minimize�? (Especially where we do not expect
negative values.)

The perception isn’t stabilized. Nor is the engineering “output” (the PCT “input”). If control works well, engineering “output” changes as much as the engineering “input” and PCT “input” changes as much as the PCT “reference”. Indeed,
as I pointed out in a previous message, it may be extremely difficult or even impossible for an external observer to determine whether a control system is acting if the reference value is changing as unpredictably as is the disturbance or more so (PCT language)
and the observer can see only the PCT “input”.

EP: As a side thought I think that both �input� (what the subject perceives) and the “output� (what the subject does or affects, containing at least parts of the subject
itself) are usually perceived by the observer, but of course the latter is not necessary for inferring that there is control.

Of course, if the reference value changes only slowly or is temporarily fixed, an external observer will see the value of the externally observable “input” variable to be stabilized, which I think is why there is confusion over the use of
the term “stabilize” in PCT. Like the reference value, the error value that is the only truly stabilized variable can never be observed directly, but must be inferred from observations of the “input” variable in the “external environment” section of
the loop.

EP: (Side thought again) Yes, it is inferred from the changes in the input variable but doesn’t it require also knowledge about disturbances (either caused by something
else in the environment or by the observer/tester herself)?

So I think “stabilize” is actually the best word to use in conjunction with “control”, provided you realize that the essence of control is the relationship between the reference value and the perceptual value – the “error”, which is observable
in an engineered system, but not in a biological control system. All else in the loop, including everything that an external observer can observe, can vary rapidly and widely, depending on the loop parameters and on the ways the reference value and
the disturbance value change.

EP: This means that �stabilization� (or “minimization� as I suggested) is to be used to name that part of control which is inside the subject in the upper right area
of the canonical PCT diagram. Until now it has been used to point the environmental part of the loop, where often in some conditions happens apparent stabilization of some variable values. The concept referring to what generally happens to the CEV by the subject
during the control is still without a term.

[Aside: A point that always comes up in this connection is how to treat the top of the hierarchy, if the reference signal is derived from the outputs of higher level control units and there are none.
The answer given by Powers is that if the reference value is fixed, there doesn’t have to be any neural reference signal path, because that connection is used only to allow the reference value to vary. Every top-level reference value can be taken to
be permanently zero or any other value determined by the physiological environment. A reference value, however, must exist if there is to be control.]

EP: That is also important: the reference value does require reference signal! I think there is also a speculative possibility that any control unit can have a fixed
built-in partial reference value (typically zero perhaps). But even if the reference value of a high level control unit is partly or fully built-in it doesn’t mean that it were fully fixed. It can change as consequence of reorganization i.e. learning.