[Martin Taylor 2006.11.26.13.06]
[From Rick Marken (2006.11.26.0940)]
I think you still have to work a bit on understanding what a disturbance is in PCT. No offense, but I think you might want to be sure that you understand what a disturbance is before you start teaching others what it is.
No offence taken. I had come to the same conclusion myself, as you could see from my message. But I think the same applies to all of us (you and Bill P. included), right now.
I do (think I) understand the nature of a disturbance _as defined in B:CP_. Your "corrections" contribute nothing to that, since all you do is to complain about the verbal shorthand one must use if one is not to lost the content in the web of verbiage. OK, it's valuable to distinguish "disturbance source", "disturbance variable", and "disturbance variable value" when the context might make the meaning ambiguous. It is pedantic to do so otherwise.
The big problem I now have is how to assess disturbance from a viewpoint other than that of the analyst. The external observr's viewpoint is all we have when we act as experimenters, but the "girl walking" example illustrates that we can have profound disturbances to controlled perceptions that alter the input quantities substantially, with no -- zero -- observable output because of conflicts between different disturbed control systems.
That observation alone is sufficient to lead one to question the iconiv value of "The Test", which is the deliberate application of what is preumed to be a change in a putative disturbing variable.
Suppose you, as an experimenter, asked the girl to walk past the subject, as a Test to see whether it would disturb a variable such as a reference to perceive himself as being close to such an attractive girl. If he got up and made some action to bring himself to her notice, you might argue that it had indeed disturbed some such perceptual variable.
But suppose that the same subject was also controlling a perception of self-image that would depart from its reference value if he acted contrary to what he perceived as "proper" social behaviour, and that his perception of "proper" behaviour included not intruding on the privacy of people to whom he had not been introduced. That control system would conflict with the "get close to the girl" system. One would reduce error if he got up, the other would increase error if he got up. For the sake of the example, we assume the latter wins, and he sits still, making no overt action as she walks by.
You have applied "The Test", and obtained the result that application of the presumed disturbance has resulted on no output. According to all I have seen on CSGnet and heard in PCT discussions, you would conclude that the girl's passage disturbed no controlled perception. But it did.
The other problem I have with finding a precise definition of "disturbance" is the continuity of degree of control between zero and non-zero amounts. It's one thing if there is no pathway connecting the presumed disturbance source with the perceptual signal in question, but it's quite another if there is such a pathway and change in the value of the disturbing variable fails to result in a countering change in output. In that case, the external observer can't distinguish among four cases, at least not by observing the putative disturbing variable and the input quantity or (equivalently) the countervailing effects of control:
(1) The change in the putative disturbing variable does not influence the perception;
(2) The change in the putative disturbing variable influences the perception, but the perception is not controlled (i.e. there is either no reference value for the perception);
(3) The change...does influence the perception, which is controlled, but the control gain is near zero because controlling that perception is for the moment of low priority (I'll expand this below);
(4) the change .... influences a controlled perception, but the output does not influence the perception (e.g. a prisoner with a reference to perceive himself to be out shopping).
Case 3 needs a little more explanation than is reasonable in a list like the above. I'm thinking of a thresholded system, in which the input to the output function remains zero so long as the error is less than some threshold. In modelling the many tracks in my sleep-loss study, all the models with such a threshold fit better than the same models with zero threshold, so it's not an abstract possibility.
This case 3 situation makes the definition of "disturbance" rather problematic, even for a single isolated control loop that evinces good control for sufficiently large disturbance variable values. When the error value is near zero, small changes in the value of the disturbing variable result in zero change in output. One might easily say that such small changes do not constitute disturbances at all. But should one say that, when the only difference between that situation and one that produces good coutervailing output is in the value of the putative disturbing variable?
My thought at the moment is that it's fairly easy to find clear cases in which a putative disturbance can be shown to actually disturb some controlled perception, but it's not easy to show that a change in a person's environment is not a disturbance to any particular controlled perception, and it's not easy to determine (except in simple cases such as Rick's studies of geometric control) just what controlled perception or perceptions is/are being disturbed when it is clear that at least one is.
From the analyst's viewpoint, it's easy. The analyst knows all the connections, and if there is a control loop that goes from some point influenced by the putative disturbance source, back to that same point, then the source really is a disturbance source for that perception.
The analyst has that luxury, but nobody else does. And that makes it difficult to provide an operational (i.e. experimentally observable) definition of a disturbance.
Martin