[From Rick Marken (2016.10.05.1115)]
Martin Taylor (2016.10.01.22.47)--
RM: This is remarkable Martin. In this post you have proven (at least to your own satisfaction) that the Test for the Controlled Variable is impossible.
MT: Nope. I showed that the Runkel criteria were not sufficient in the general case. The Test for the Controlled Variable is both possible and useful under certain conditions.
RM: I agree that Runkel's criteria are not sufficient, but not for the reasons you say. The main problem is his final criterion: "If all preceding steps are passed you have found the controlled variable." This gives the impression that the TCV can end with a definitive identification of a controlled variable. In fact what we get with the TCV are closer and closer approximations to the actual controlled variable. Good illustrations of this are in Powers analysis of a shock avoidance experiment", described in the "Experimental Methods" section of B:CP, where he shows the probability rather than rate of shock is actually a better estimate of the variable controlled by rats. Also, Chapter 4 in my "Doing Research on Purpose" shows how the TCV is used to get closer and closer approximations to the perceptual variable that is actually under control.
MT: Actually, I should have linked to <<Control of Size; because it illustrates my point about the need to vary the reference value if you want to use the Test to choose among selected functions of the same set of variables.
RM: I don't understand where you got this idea. The control of size demo shows that there is no need to vary references in order to chose among selected (perceptual) functions of the same set of variables which is under control. Indeed, the demo works best if, as instructed, the subject adopts a fixed reference for the size of the rectangle.
MT: I only wanted to illustrate that there are quite normal situations in which they are insufficient, particularly number 9, which says that after testing just one hypothesis, if the criteria are met (criteria that can be met by any variable correlated with the actual controlled variable) the one you tested IS the controlled variable, even though a whole raft of other variables might fit the criteria and be better controlled than the one you first thought of.
RM: As noted above, I agree with this. It should always be made clear that the TCV is an iterative process and that the goal is to find the best approximation to the controlled variable.
RM: ...control systems in conflict don't control but, rather, lose control; they can't resist disturbances to the "virtual" controlled variable:
MT: That is exactly right. And totally irrelevant. That's what Kent neatly demonstrated at CSG '93, in the video to which I linked. Here's the link again: <<https://urldefense.proofpoint.com/v2/url?u=http-3A__www.mmtaylor.net_PCT_Movie_McClelland-5FCSG93.mp4&d=DQMFaQ&c=8hUWFZcy2Z-Za5rBPlktOQ&r=-dJBNItYEMOLt6aj_KjGi2LMO_Q8QB-ZzxIZIF8DGyQ&m=A-E_OXOguD89CGxh9CdtIVnlmOYWqgcosjElD9WdPJo&s=8bBf6ybWon5JvgRKJ7tXQ1XksT-utgjJMHWRP6ZE39Q&e=>http://www.mmtaylor.net/PCT/Movie/McClelland_CSG93.mp4>\.\.\.
MT: FYI, Kent's is an empirical demonstration, insofar as a computer simulation can be one. But you wouldn't know that, even though you were there when he gave his presentation. However, Bill P said of it at the end, and I quote: "That's great stuff. [You should] go on." (The words in square brackets are indistinct in the video).
RM: Yes, it was great stuff, particularly because, at the time, it was one of the few examples of using working models to test the predictions of PCT. Kent provided some really nice demonstrations of what can happen when two control systems are in conflict. But what struck me when re-watching this video (besides the fact that my hair was still mostly brown) is that these simulations were not (and, as far as I know, still have not been) compared to the actual behavior of people in conflict.
RM: The simulations showed that people in conflict over the state of some variable will keep that variable in a "virtual reference state". Apparently this is now taken to be an example of one kind of "collective control". But I can't think of any real world example of this kind of control. That is, I can't think of situation where two or more people persist in trying to keep a variable in different reference states so that the variable remains in a virtual reference state -- a state that is, in some sense, socially desirable. A real world example of this kind of collective control would be several cooks controlling for different states of a broth -- one wanting bouillabaisse, another tomato bisque, another cream of celery, etc -- and the result will be a delicious vichyssoise, something none of the cooks want. I don't think this ever happens; more likely that the broth is spoiled.
RM: So while Kent's models were a "great" demonstration of some interesting things that can happen when control systems come into conflict, I'm not clear about the relevance of these demonstrations to what actually happens when two or more living control systems come into conflict over the some variable. And I am particularly skeptical of the idea that anything good can come of this kind of conflict. Even if people do maintain variables in virtual references when they are in conflict, it seems that this could only be of interest to an observer; none of the people involved are getting what they want.
Best
Rick
···
--
Richard S. Marken
"The childhood of the human race is far from over. We have a long way to go before most people will understand that what they do for others is just as important to their well-being as what they do for themselves." -- William T. Powers