Reorganization and the Hierarchy of Perception

[From Bruce Abbott (2016.09.26.1810 EDT)]

Chad Green (2016.09.26.1116 EST) –

CG: On p. 239 Powers writes: “If there can be any ultimate determinant of the way we learn to think and act, it must be in this set of inherited specifications for the state of the physical organism that calls for no change in behavioral organization.” Shouldn’t this metacontrol system serve as the highest level of HPCT?

I have always been troubled by the notion of “intrinsic variables” as the drivers of reorganization. I’ll try to explain why. I’ve also seen some difficulties with the proposed hierarchy. I’ll explain that, too.

Bill Powers borrowed the ideas of intrinsic variables and reorganization from W. Ross Ashby, who proposed them in his book, Design for a Brain. Ashby even built an electromechanical model he called the “homeostat” to illustrate the principle. When the homeostat lost control over a variable, the homeostat would begin to cycle through a set of changes in parameters until it found a set of values that would restore stability to the system. This ability to restore control by reorganizing connections within the system conferred on the system a property that Ashby called “ultrastability.”

One of Bill’s goals when developing HPCT was to make as few assumptions as possible concerning how the control hierarchy would come into being. Ashby provided an elegant solution – beginning from a relatively limited set of genetically-specified “intrinsic” variables and associated reference levels, the reorganizing system would act like Ashby’s homeostat to vary component connections and parameters whenever a relatively persistent error arose between the states of those intrinsic variables and their reference levels. Reorganization would occur at a rate proportional to the size of the error; thus, as reorganization managed to achieve an organization that better and better controlled the intrinsic variables, the rate of reorganization would slow until it effectively froze the new organization in place. Once the bottom-level control systems were in place, it became possible to further enhance control over intrinsic variables by combining bottom-level perceptions to produce more complex perceptions and developing control systems that would act by manipulating the reference levels of the systems immediately below. This process would continue, level by level, until the entire control hierarchy was in place.

It’s an elegant solution – the entire hierarchy of control systems above the intrinsic variables becomes self-generating with the aid of the reorganizing system, itself a control system. So what problems do I see with it? For one, the fact that development would depend so strongly on what happens to “work” at any given moment implies that there would exist a great diversity of nervous system organizations among the individual members of a species. Yet what differences there are, are minor (with the exception of pathological cases). Although your brain and mine may differ in the relative sizes of certain structures, the gross anatomy is identical – for every structure one can identify in my brain, one can identify the same structure in yours, and they are connected together in the same ways. This seems an unlikely finding if the development of our perceptual functions and control hierarchy depends on random reorganization and selective retention. Furthermore, not only are the brains of individuals within a species nearly identical, but even when making comparisons across species, one can find the same structures wired together in similar ways. For example, there is such a thing as a “mammalian brain”; the human brain and the rat brain have nearly the same structures and organization, despite the two species having been evolving along separate paths for millions of years. My conclusion is that brain development is much more strongly constrained by a genetically-orchestrated developmental plan than is suggested by the notion that the hierarchy of perception and control develop as a function of random reorganization and selective retention.

Another problem I have with reorganization as implemented in HPCT is that it has no mechanism that will allow it to target specific systems where control is failing. Bill was well aware of this problem but did not find a solution for it that he found acceptable. Without a mechanism that will restrict reorganization to those systems for which control is poor, reorganization will start to reorganize systems that are currently working well, leading to a reduced ability to control the perceptions those systems control. Bill suggested that reorganization might be targeted to the right places by attention, but attention itself is not currently given a mechanism within HPCT.

With respect to the HPCT hierarchy of perception and control, physiological evidence may only partly support Bill’s proposal, which was based mainly on his logical analysis and introspections and which Bill advised should be taken only as a starting point for research. At the level of intensities, for example, Bill argued for perceptions of intensity that were not associated with a particular sensation. The sensory experience (whether of light, sound, heat, etc.) was to appear at the Sensation level immediately above the Intensity level. Yet whether a given neural current is perceived as, say, a sound, depends on where in the cerebral cortex that neural current arrives. If it is conveyed to the auditory cortex, then it is perceived as a sound (with certain properties); the same neural current arriving in the visual cortex will be consciously perceived as light. Thus it appears that the kind of conscious sensation perceived depends not on combining certain intensities from the level below, but on conveying the signal to a particular analyzer in the cortex. (Aside: No one has any idea how activity in one cortical area leads to consciousness of sounds, as opposed to light or of any other kind of sensation.)

It turns out that our visual system responds to change; when a visual image is artificially stabilized on the retina, visual perception rapidly fades. This is not the simple Intensity sensation posited for Level 1. In the visual cortex, a separate system provides the conscious sensation of motion within the visual scene. One would think that having the image of an object changing position from one moment to the next would be enough to create the impression of motion, but in the rare cases in which this cortical area is damaged, people report only a sequence of still images. When we scan our eyes across a scene, the eyes move by tiny jumps (saccades) and suppresses the visual image while the image is in motion across the retina and would thus be blurred. Apparently the motion system fills in the gap during this motion by doing something like “morphing” the object between its serial still images. Without the morphing, motion perception ceases. Again, perceptual processing appears to be more complex than the simple scheme of combining perceptions from the next-lower levels to form new perceptions.

The take-home message I hope I’m conveying here is that HPCT should not be taken as a finished theory that only needs further support to validate its proposals. Rather, it should be viewed as a starting framework to be modified and elaborated as required by the evidence. To me, the core of PCT, what I like to call the “Powers conjecture,” is that the function of behavior is the control of perception. Everything else is just an elaboration of that core proposition. Bill worked to provide a theory concerning how control systems might be organized in the brain (hierarchically), how that organization might come into being (reorganization during development), and how existing systems might be reorganized to regain effective control over perceptions when control is compromised. He hoped to attract researchers who would put that theory to the test and modify it as necessary to fit new established facts. We shouldn’t fear making such alterations if they prove necessary and consistent with the evidence. No matter what changes may be required, so long at the Powers conjecture remains the core principle, it will still be PCT.