Martin
···
From: Martin Taylor (mmt-csg@mmtaylor.net via csgnet Mailing List) csgnet@lists.illinois.edu
Sent: Tuesday, May 22, 2018 5:52 AM
To: csgnet@lists.illinois.edu
Subject: Re: Varieties of PCT
[Martin Taylor 2018.05.21.23.23]
[Rick Marken 2018-05-21_18:01:05]
[Martin Taylor 2018.05.18.12.46]
RM: An excellent treatise on the theory-only approach to PCT!
HB : What exactly was excellent ?
MT : Thank you. I will accept “excellent treatise” but I do not think it is or was an approach to PCT of any kind, theory-only or scientific.
It is just a reminder of the wide scope of Perceptual Control Theories, nothing more Such a reminder from time to time seems to be needed, because people write as though they forget that the fundamental claim that behaviour is the control of perception can be the basis of many theories, and that no experiment of which I am aware has yet distinguished among even some major classes of them. The Test for the Controlled Variable, for example, works equally well with all of them, as does any analysis that is concerned with only a single control loop. The proper study of PCT is the discrimination of the correct theory among the great range of possible theories (assuming as I do that the correct theory is a Perceptual Control Theory).
To make that discrimination requires either experiments that would give results predicted differently by different theories, or analyses of the theories that shows one or other of them to predict results that are not in accordance with the body of science. Your “hierarchies of perception” is such an experiment, since it argues quite strongly against a “flat” version of PCT in which all controlled perceptions are at the same level as each other. But there aren’t many such experiments.
HB : Martin. Could you specify more precisely where exactly you don’t agree with Bill ?
Boris
Martin
Much, but not all, of the problem of long unresolved threads on CSGnet comes from misunderstandings about the intentions of the writers when using certain words and phrases. Other problems arise because each of us, novice or self-appointed expert, has a personal perception of what is and what is not “PCT”, and of what consequences arise from that personal perception when it is applied to the real world. I doubt that any taxonomy could incorporate all the varieties that different CSGnet readers think of when they think “PCT”, but it is possible to describe some variations that deserve the name.
A long time ago, in a mailing list far away, we had three classifications for variants of PCT. I forget the names we used, but in principle they were: Generic PCT, Generic Hierarchic PCT, and Bill Powers’s specific version of Hierarchic PCT (HPCT). Some contributors to CSGnet now write assuming that Bill’s HPCT was “PCT”, as though the road to enlightenment was only through studying Bill’s writings, despite the unavoidable self-contradictions that occur between his early and late writings. As should be the case for any good scientist, Bill’s understanding of PCT developed, over the half-century and more that he studied the control of perception. To base discussions of PCT only on fine interpretations of Bill’s writings, especially the earlier ones, is a sure way to create disputes. Just as an exhaustive study of any ancient writings by several people will do, the various Koranic, Biblical, or Talmudic scholars come to different conclusions about the truth that is hiding behind the Holy Writ of PCT.
My own preference is to work from the other end, to consider Nature, using my own interpretation of Bill’s powerful insights as a valuable but not infallible guide. I ask questions such as: Supposing this or that variant form of PCT happened to be correct, what would we observe? What limitations does Nature impose on control, and is this or that version of PCT consistent with those limits? How could we know whether what we perceive has a counterpart in properties of the environment in which we live? Etc., etc… One class of questions concerns whether observed effects depend on Generic PCT, Generic hierarchic PCT, HPCT, or some other variant, of which there may be many.
What did we mean by these three levels so long ago? Generic PCT simply means a theory that organisms create internal variables that correspond to some patterns of sensory data based on things and events in the real environment, and act to influence those patterns to conform to internal “reference variables”. Generic PCT specifies nothing about the form of these control loops or how one relates to another. They might all work independently, they might have all sorts of cross influences or network structures of interaction. Generic PCT allows almost anything that involves control of internal variables by action on the environment outside the organism.
Generic Hierarchic PCT (GHPCT) is a specific refinement of Generic PCT, one of many possible structures of interaction among control loops. The specific refinement is that the internal (perceptual) variables individually represent identifiable entities or structures in the environment. Those that are controlled come in simple and more complex forms, the more complex being built on the simpler to form a kind of “tower of complexity” with very complex perceptual forms at the top and raw sensory data at the base. GHPCT allows any kind of relationship among the functional processes that lead to control of these perceptions.
Powers’s Hierarchic PCT (HPCT) is a specific refinement of GHPCT, in which different levels of complexity and a specific form of interconnections among the functional processes is defined. Although Powers did not insist on it (at least at the lowest levels), one might include as a necessary component of HPCT the eleven types of increasingly complex perception Powers proposed. Of course, I am not saying that Powers developed his version of PCT by refining GHPCT. Far from it. The reverse, a generalization from Powers’s HPCT to GHPCT, would be nearer the truth.
Another way one might categorize varieties of PCT is according to how the PCT structures evolve over generations and within individuals. Among other possibilities, we might include e-coli reorganization, reinforcement learning, Hebbian cell-assembly and random connection making and breaking, all globally, in modules, or in hybrid form. Powers’s HPCT incorporates e-coli reorganization generically, but not in any formal and specific way. Different variants of Powers’s HPCT would depend on different understandings of how reorganization functions to change the control hierarchy.
Already we have listed over a dozen possible varieties of PCT, and these are far from exhaustive. For example, there are other species of GHPCT, and forms of Generic PCT that are highly structured but not hierarchic. I don’t know whether anyone has seriously proposed the following as a PCT variety, but in the silico-acoustic world the inverse Fourier transform of the log Fourier transform is a useful tool. Imagine a model in which the controlled perceptions were quasi-spectral relationships among vectors of logarithmic values. Structurally, this kind of thing was proposed (a long time ago) for the distributed character of perceptions to protect them against focal brain damage, and maybe control of spectral representations might be functional. Many years ago, when I visited the Swedish Defence Research Institute I was shown how re-inverting the phase spectrum of a spatial Fourier transform after losing the amplitude information produced a good outline drawing of the original picture. Neural systems seem well suited to doing this kind of thing, but is it well suited to control in a dynamic environment? One could actually consider the plethora of specific perceptual functions of Powers’s HPCT at a given level of the hierarchy as a kind of quasi-spectral transform, mathematically.
One specific variant of GHPCT that I do like is based on the apparently different functional tendencies of the brain hemispheres. This variant is like Powers’s HPCT except that there is no category level as such. Instead, each of the other perceptual types may interface with a category process that produces category values and performs logical operations on them ( (left-brain function)) interacting with analogue values (right-brain function) at each level. Again, there are probably many variations on this theme, all producing different variant forms of PCT.
My main point is not to advocate for any changes to Powers’s version of HPCT, but to show how easily variant forms can be imagined. This being the case, how many more subtly different varieties of PCT may exist in the hidden assumptions each of us has when we discuss “PCT” on CSGnet and get into unnecessary arguments. In most cases, so long as the assumptions remain hidden, arguments are unresolvable about what is or is not PCT and about what is or is not true of Nature viewed through a PCT lens. Even if by some lucky chance the conflicting assumptions are revealed, it is as like as not that no experiment has been done that could distinguish between them, and the only rationale for choosing one over the other would be Ockham’s Razor – and that particular razor shaves differently for different people (see http://www.mmtaylor.net/Academic/ockham.html for the explanation).
My second point is simply to try to shake up a bit the idea that some people seem to hold, that “PCT” means exclusively the constellation of concepts about which Powers wrote. That happens to be the version of PCT that is by far the most studied and tested, but it is far from the only one that is possible, and many of the others would automatically have provided identical results to any tests that have as yet been applied to the Powers version, including all studies such as the Test for the Controlled Variable, which examine only a single isolated control loop.
Martin
“A good scientist always tries to find what might be wrong about what s/he firmly believes”
–
Richard S. Marken
"Perfection is achieved not when you have nothing more to add, but when you
have nothing left to take away.â?
–Antoine de Saint-Exupery