Seems a fair plan. So the virtual reference level doesn’t counteract disturbances to maintain itself?
The justification for terms like virtual reference level, collectively controlled variable, giant virtual controller, and so on is given by this statement in Behavior: The control of perception:
“The level of detail one accepts as basic must be consistent with the level of detail in the phenomena to be described in these basic terms. One can always, for other purposes, analyze further.”
There, in the chapter on Premises, Bill invokes this fundamental principle (you don’t take the temperature of the water by measuring the Brownian movement of each molecule and carrying out Boltzmann/Helmholtz kinds of calculations) to justify the notion of a unitary ‘neural bundle’ represented by a line in a block diagram, where the details of brachiation, manifold synapsing, proximity or not, degree of myelination, etc. are not relevant to the chosen level of investigation. The ‘neural signal’ in a PCT model which is represented by unitary values p, r, e in control loop formulae is a Virtual Giant Neural Rate of Firing in a Virtual Giant Neuron.
After the two sentences that I quoted, Bill goes on to elaborate the consequences of this principle for PCT. I will modify that continuation to show the application to collective control.
If we wish to describe the activity of the participating members of a public which correlates with the phenomena of direct experience, and constitutes the inner components of observable stabilities of a human-built perceptual environment such as tools kept available in a workshop, roadways and signage, taxation and government policy, electrical power generated, transmitted through a grid, metered to individual businesses, homes, and public facilities such as street lights, and so on, then it would be inappropriate to begin with control of aspects of these observable stabilities by individual controllers in that public. No control effort by one isolated individual has any discernible relationship to observations (objective or subjective) of the aggregate result. Even if we knew the values of the control inputs and outputs of all the stakeholders in a given public resource at any given instant, the listing of their locations, values, and effects on the state of that resource would convey only meaningless detail, like a halftone photograph viewed under a microscope. If we want understanding of relationships, we must keep the level of detail consistent and comprehensible.
It is necessary to elaborate on this to show its application more clearly. Individual stakeholders counteract disturbances to their control of their perceptions. If a public resource is part of the feedback paths by which they do this controlling, then the availability and efficacy of that resource for their individual purposes may be among perceptions that they individually control. Public resources are of enormous variety, tools in a workshop, roadways and signage, taxation and government policy, constituting a human-built environment of great complexity. All of these public resources are perceived, in one aspect or another, by all the individuals who use them to facilitate control as individuals going about their daily affairs in this built environment perceptually overlaid on the perceived natural environment. Many individuals controlling perceptual aspects of a public resource maintain that resource in a condition at levels (plural, for the various aspects which are perceived and controlled) which for convenience are called virtual.
For comparison to my periphrastic reframing of Bill’s discussion of the principle of appropriate level of detail, here is the original quotation with additional context.
We know now that electric current is not a flow of continuous substance, but the drift of tiny individually charged particles (electrons) through the atomic matrix of a conductor. That was not the initial picture, however, nor is it the concept that & used in modern electronic and electrical engineering (except at levels of current so low that effects of individual charges can be seen). For purely practical reasons, engineers treat electric current as a smoothly variable continuous quantity and ignore what they know to be too detailed a representation of reality for their purposes.
The level of detail one accepts as basic must be consistent with the level of detail in the phenomena to be described in these basic terms. One can always, for other purposes, analyze further. If we wish to describe the activity of the nervous system that correlates with the phenomena of direct experience, and constitutes the inner component of such behaviors as walking, talking, and execution of action patterns in general, then it would be inappropriate to begin with an individual neural impulse. No one neural impulse has any discernible relationship to observations (objective or subjective) of behavior. Even if we knew where all neural impulses were at any given instant, the listing of their locations would convey only meaningless detail, like a halftone photograph viewed under a microscope. If we want understanding of relationships, we must keep the level of detail consistent and comprehensible, inside and outside the organism.
NEURAL CURRENTS
As the basic measure of nervous-system activity, therefore, I choose to use neural current, defined as the number of impulses passing through a cross section of all parallel redundant fibers in a given bundle per unit time. The appropriateness of this measure depends on the maximum neural current normally expected to occur in a given bundle of fibers. If the maximum in a bundle of 50 fibers is 200 impulses per second in each fiber, the maximum neural current will be 10,000 impulses per second, and statistical variations will not be important at any level of neural current in proportion to the whole normal range of operation (they will be roughly 1 percent of the maximum, or less).
The use of neural current is appropriate, for example, in considering the stimulation of a whole muscle, especially in terms of the forces thereby developed on the tendons and thus on the bones. The impulses going to the muscle arrive via hundreds of individual pathways, each terminating on one tiny contractile fiber, but the net force developed depends on all these parallel events, not on any one of them. The individual random twitches are averaged out.
The only quantity inside the nervous system that correlates with the net force exerted by, say, the biceps muscle is the neural current obtained by counting all the impulses reaching that muscle per unit time. That is essentially the same as counting the impulses passing a cross section of all the parallel motor-nerve fibers running from the spinal cord to the biceps muscle. I am doing nothing more here than formalizing a measure that is commonly used in neurology and physiology, even if not instrumented with just this definition in mind.
The single-impulse model of neural activity treats the convergence of separate trains of impulses on a single nerve cell by instant-to-instant evaluation of the summation of excitatory effects. The neural-current model handles the same situation in terms of continuous average summation effects. Unlike the time-state analysis, the neural current analysis handles this summation effect easily whether several small (low repetition rate) signals are converging, or one large (high repetition rate) signal is present. Binary computing elements do not distinguish the rate at which input events occur (in normal operation), and so a digital-computer model cannot handle the common situation in which one high-frequency current can cause a nerve cell to fire, while two low-frequency currents cannot. Neural impulses may either increase the tendency of a nerve cell to fire (excitatory effects) or may decrease that tendency (inhibitory effects).
The difference seems to reside in the type of nerve cell in which the impulse originates: Renshaw cells are apparently specialized to emit an inhibitory substance at the end of the outgoing impulse-conducting fiber (Wooldridge 1963). Therefore, in the summation of neural currents, some currents contribute positively to the net excitation of the receiving nerve cell, while others contribute negatively. Some connections seem to involve such low thresholds that only a few impulses arriving at once—even just one—can fire the following nerve cell. In a case where two impulses are required, and at relatively low neural currents, the chance of two independently originated impulses arriving nearly enough at the same instant is the product of the probabilities of arrival of an impulse in each incoming path per unit time. Thus the neural current generated by the receiving cell is proportional not to the sum of incoming neural currents, but to their product.
Thanks Bruce
I only wish peple would follow it!!
The virtual reference level doesn’t do any countercting. Rather, it is the state of a variable that appears to be under control but is not. (The virtual reference level is probably better described as the illusory reference level). This appearance is the result of conflict between control systems that are controllng (actually trying to control) a variable relative to different reference spcifications.
A variable will appear to be controlled – maintained at a reference level – if the systems involved in the conflict are producing maximum output and the disturbances to the variable these systems are trying to control push it outside a dead zone – the range of values of the variable that corresponds to the difference between the reference specifications for all control systems involved in the conflict.
All this was explained (and demonstrated) in my presentation to the IAPCT conference in 2023 (click to download; it’s a Power Point file and it’s safe to “keep”; NB. Slide 11 on the “Illusion of Control”).
The problem with the concepts of “giant virtual cotroller” – or just “virtual controller” – and “virtual reference level” is not the level of detail but the level of truth. A virtual controller doesn’t control and a virtual reference level is not the reference level of a controlled variable.
Oh, and Bill’s discussion of “level of detail” in modeling control systems is (as usual)) right on target because the detailed components of the control systems in people are causally connected. So, for example, the detailed neural conections that make up human perceptual functions, I(), are causally connected to the detailed neural components that make up the perceptual signal, p, etc.
So we can model p = I(v) knowing that the detailed entities that make up I() and p are causally connected in a way that makes this functional relationship hold. But this is not true of “collective control” phenomena, where the people that make up the functional components of “collective” control organizations are not causally connected like the neurons that make up the functional components of human control systems.
So, to the degree that the reference levels of the (many) systems are close in value, they will work together to maintain the virtual reference level, but to the degree that they are distant in value, they will form a dead zone that cannot be controlled. So, there must a set of values of number of systems, reference values, gains and delays of each system, and differences in each system’s feedback function, over which these emergent illusions have a continuum of effects?
Two things. First, I think going from 2 systems to many systems in conflict is not as straightforward as it has been thought to be. I recall Kent having a paper where he describes models with more than 2 systems in conflict. I can’t seem to find it in the disorganized mess that is my file system. Could you (or someone) please send me a copy of that paper if you know which one I’m talking about?
Second, your comment seems to be speculating about the kinds of “effects” you might be able to get out of a giant virtual controller model. I presume the effects you would like to see are behaviors of the model that correspond to social behaviors that have been observed. If that’s correct, I would really like to see the social behavior data that you are trying to explain.
By the way, a great example of using a PCT model to explain social behavior data is the CROWD model. The social behaviors produced by variations in the parameers of the CROWD model correspond, at least qualitatively, to the observations of CROWD behavior that have been made by Clark McPhail and Chuck Tucker. The CROWD model was Bill’s effort to show sociologists how to do PCT sociology. I have no idea why PCT sociologists didn’t didn’t run with this – indeed, they seem to have run away from it. Go figure.