[From Bruce Abbott (2018.03.08.1030 EST)]
[Martin Taylor 2018.03.07.10.51]
Martin, thanks for this.Â Letâ€™s see where it takes us.
[From Bruce Abbott (2018.03.06.1745 EST)]
[Martin Taylor 2018.03.06.11.55]
[From Bruce Abbott (2018.03.06.0935 EST)]
Bruce, the fact that you bring up your dissatisfaction with the idea of “intrinsic variables” is a sufficient disturbance for me that I would like to explore with you the notion of “intrinsic variable”, not as you or I now understand it, or even as Bill wrote about it, but as the concept of a self-organized perceptual control hierarchy may (or may not) require it to be.
To start this process, I will make a few comments on your message, both supportive and not.
To make this comment clear, I am going to omit a lot of very good (in my opinion) material from Bruce’s message. I may comment on some of it in a separate message. Or I may not, but if I do, it will be under the original subject line.
I must admit, the more I have thought about it, the less enamored I am of the concept of intrinsic variables. … … …
I propose that we are equipped with the capacity to experience such deviations (including the associated feelings such as discomfort) so that we can learn how to deal effectively with them – i.e., to bring the variables in question under behaviorral (as opposed to purely physiological) control.
Do you think that your last paragraph (the last of your full message) differs from the “standard” approach to intrinsic variables, as described in various places by various people, among them Bill P. and me (in a web page approved by Bill and in subsequent writings on CSGnet and elsewhere)? I don’t. I read it as a pretty good basic description of the way that errors in intrinsic variables are corrected by suitable perceptual control action, leading to reorganization when the existing perceptual control mechanisms persistently fail to act in ways that have side-effects that correct them. In that view, the key point about “intrinsic variables” is that they matter to survival and that when they go wrong, reorganization rates increase so that the perceptual control system can learn to “bring the variables in question under behavioral (as opposed to purely physiological) control.”
It seems to me that there are important differences between my proposal and Bill Powersâ€™ conception of reorganization. According to Bill,
â€œAshby dealt with this question by defining essential variables, top priority variables intimately associated with the physiological state of the organism and to a sufficient degree representing the state of the organism. Each essential variable has associated with it certain physiological limits; if a variable exceeded those limits, a process of reorganization would commence until the essential variables were once again within limits. â€œ
â€œThat is the essential character of the reorganizing system I propose: It senses the states of physical quantities intrinsic to the organism and, by means we will discuss shortly, controls those quantities with respect to genetically given reference levels. . . .â€?
Powers (1973, p. 179)
So, we have intrinsic variables (Billâ€™s relabeling of Ashbyâ€™s essential variables) that occupy a privileged position in the system: their values are being monitored and if exceeding certain limits, trigger the reorganizing process, which continues until the organism either dies or regains control over them. Their reference values and those limits are genetically specified. The reorganizing process operates by means of a biased random walk in which beneficial changes (those that improve control) reduce the rate of reorganization and thereby tend to preserve those changes.
A serious problem with this conception, one that Bill recognized and tried to resolve, is that it does not indicate how the reorganizing system is able to make the changes only where they are needed. Lacking this, such a system conceivably might reorganize the wrong systems, reducing or even destroying their abilities to control effectively while failing to address the error in the intrinsic variable that is animating the reorganizing process. Although Bill tried to solve this problem, to my knowledge the best he was able to come up with was to suggest that consciousness is somehow involved in targeting what gets reorganized and what is left untouched.
Bill and I discussed this issue at length, without resolution. Remember that Bill thought of reorganization rate as the important variable, not reorganization, yes or no. Even well-functioning areas were subject to reorganization, but slowly compared to areas of persistent and increasing error. That allows the system to get out of local optima that blocked routes to more global optima (Aside: MoL is based on this, together with the idea that consciousness could direct reorganization).
I probably have not kept in mind well enough the fact that, in Billâ€™s conception, reorganization is going on continually, even in systems that are currently functioning well.Â I did note later on that reduction of error slows (not halts) reorganization, thus allowing organizations that work well to persist for longer, but at other times I have treated the process as if it affects only systems that currently experience relatively large and persistent error.
I don’t know whether he or I considered approaches to modularization based on Quality of Control (QoC), but my impression of what we both thought at the end of our discussions on the subject is that the combined magnitude and rate of change of QoC influences the rate of reorganization of the properties in and around that control unit. A recent diagram illustrating my idea of the process, as of some time last year, is near the end of the attached extract from PPC.
Again, I don’t know who demonstrated that QoC must be treated as an intrinsic variable, but there is, I think, a good evolutionary argument that it is, because only good control can produce consistent side effects in a consistent environment. In an inconsistent environment, one is likely to need to reorganize so as to find a variety of ways to control, effectively reducing the environmental inconsistency and enhancing the likelihood that the side-effects of control would have consistent effects on the intrinsic variables.
In most (all?) of his learning simulations, Bill used QoC as the only intrinsic variable, always with a reference value of zero error, and applied it in a non-modular fashion to the combined error of all the control units in the simulation, if I remember correctly. But that non-modular approach is not necessary, especially in a growing and maturing organism, in which the lower levels may reorganize to stable configurations before other levels are built on their foundation.
I believe that at least one of the demos in LCS III provides an option to choose â€œlocalâ€? or â€œglobalâ€? reorganization, but at the moment I canâ€™t remember which.Â In that context there didnâ€™t seem to be much difference in their effectiveness.
My post was written partly in response to the following from your post [Martin Taylor 2018.03.06.00.58]:
Both Eetu and Bruce hypothesized that as part of the syndrome of sickness effects the reference value for blood temperature might vary. To make your hypothesis functional, you need a source for this variation of the reference value. In the perceptual control hierarchy, reference variation is always a consequence of changing output from a higher-level control unit*. The temperature control unit is presumably not in the perceptual control hierarchy. At least so far as I know, nobody has suggested that it is not an intrinsic variable.*
My guesses as to what might be happening used the temperature control unit as a top-level intrinsic control unit, and the lassitude was a natural effect of output variation to balance energy dissipation and keep the temperature at its (fixed) genetically set reference value. Do you (Eetu and Bruce) have a hypothesis as to where the varying reference value comes from, and why it should come from there in an evolutionary sense?
Iâ€™ve italicized the most relevant parts. If Iâ€™ve understood you correctly, you are asserting (1) that body temperature in an intrinsic variable and as such lies outside the perceptual control hierarchy and has a genetically fixed reference value, and (2), this raises the problem of where, in my analysis of fever, the changes in reference value are supposed to originate.
Your point (2) is a correct interpretation of my intent, but your point (1) is not.
I didn’t take body temperature to be an intrinsic variable and say that therefore it lies outside the perceptual control hierarchy. It’s rather the other way around. Because we cannot act directly on body temperature to control it, therefore it lies outside the perceptual control hierarchy. But that’s not why I thought of it as an intrinsic variable. The argument went much deeper, to the energy “intrinsically” required to maintain the dynamic structure of the body, and the need to dissipate that energy in a low-entropy form, of which heat is a prime example.
The example of the refrigerator extracting entropy from the interior so that it could be exported using the energy flow to the room was the simple form of this. The body temperature typically being higher than the temperature of the environment allows the export of heat to be passive, but the internal biochemical processes have dynamic rates that are a function of temperature, which suggests that their cycle timing relationships have evolved to work most stably at a stable temperature. It is this that makes temperature an intrinsic variable, in my mind.
This question led me to suggest something that Iâ€™ve been thinking about for quite some time: that there may be no controlled variables with a privileged position in the system such that serious, persistent deviations from genetically specified reference values produce a biased random-walk reorganization process. We know that certain physiological variables are regulated by genetically specified control systems, but there is no reason to expect that their reference levels are genetically fixed, although in many cases they may be. Instead, their reference levels may be influenced by inputs from outside the control system.
Yes. They couldn’t be “fixed” in the sense that their numerical value remains unchanged forever, but they could be “fixed” in the same way that dynamic components of the perceptual control hierarchy are fixed to produce desired dynamic patters of activity, such as performing a trill or an arpeggio on a piano. Their reference values would be changing as a consequence of the dynamic activity of other intrinsic variables, and one would expect that external inputs would influence those reference values in the same way they do in the perceptual control hierarchy, by disturbing other controlled variables – higher-level variables in a hierarchical control system.
“Biased random walk” seems to me not to describe the e-coli process of reorganization. Maybe you did intend the apparent deprecation of e-coli, but I think of it as an effective hill-climbing process in a situation that does not allow the steepest hill direction to be determined.
I think you misunderstand what I mean by â€œbiased random walk.â€?Â It is not a depreciation of e. coli, it is a technical description of how the e. coli method works.Â To take the literal e. coli example, the tumbles select new directions of travel at random.Â However, if the new direction yields a decreasing signal of nutrient concentration, a new tumble takes place fairly quickly, but if the signal is increasing, this delays the next tumble.Â The result is a â€œbiased random walk,â€? in which the system is more likely to move up the concentration gradient than down it.
For some physiological variables (e.g., blood pressure), we seem to have no conscious awareness of significant deviations from normal reference levels. You can be afflicted with high blood pressure and never know it. For others there seems to be associated with the error a particular experience: we perceive what variable is involved, the intensity of the error, and a feeling of unpleasantness associated with that experience. You feel excessively and uncomfortably hot or cold or hungry or thirsty or in pain, etc. If you already have a behavioral means with which to correct the error, you engage in that behavior. If you donâ€™t, you begin trying out different things, varying your behavior until something succeeds. You donâ€™t just randomly start reorganizing, inadvertently reorganizing away your ability to control the tying of your shoestrings along the way.
Except for the last two sentences, I was with you. I even agree with them, but not with the implication that they describe the reorganization process. Bill was clear that reorganization never stops, but its rate depends on the quality of control. The problem was not whether reorganization was localized, but how that localization was produced.
OK, but what Iâ€™ve proposed does act locally on the system for which control is poor or has been lost.Â If successful, it produces a system reorganized to control better.
My suggestion is not quite the method of reorganization that Bill proposed, but it may be more similar than it appears. Bill suggested that we do not experience a control systemâ€™s error signal.
I disagree with Bill on that, as I have mentioned here on CSGnet. It seems to me that we do experience, say, both wanting the glass to be full of water and seeing that it is not. We experience both the reference value and the error. That was what led me to see that a hierarchy in which a perceptual function took as its inputs, not lower level perceptual values but lower level reference and error values, would behave exactly like the Powers hierarchy, except that a higher level unit would “perceive” the lower level unit as being equal to its reference value if the perception was within tolerance bounds, instead of taking the actual lower level perceptual value as its input.
Interesting.Â This does seem to agree more with experience than the idea that error signals are not perceived.Â
In my conception, errors in certain controlled physiological variables arouse perceptual effects in consciousness – the perceptions and feelings previously desscribed, that allow the individual to identify what the problem is and initiate a process that typically leads to a behavioral method (or methods) that effectively counters disturbances to those physiological variables. Control systems are developed or improved via this process, or in other words, reorganization takes place. In its most primitive form the process may involve no more than blind trial and error, but more sophisticated methods may be available based on the organismâ€™s previous experience with solving similar problems.
I suspect this “most primitive form” is unlikely to be found even in the simplest extant life forms. As Bill pointed out, it would take longer than the life of the Universe to reorganize even a fairly small system effectively, whereas e-coli can do a pretty good job in a time that could be useful to an organism.
E. coli is a trial-and-error process:Â if what you are doing is not making things better, try something else, and keep varying until you succeed in getting the variable under control.Â As for the time required, a given environment, coupled with the capacities of the organism as currently organized, may have only a limited number of options that will not require anything like the age of the universe to try out.Â The hungry rat in the operant chamber usually discovers fairly quickly how pressing down on that thing on the wall makes food pellets appear in the cup.
This being so, I am left wondering why you wrote the first sentence of your message, the first quoted here. But perhaps in your final paragraph you did not mean what I interpreted you as saying.
I can interpret what you say in another way, which is that what are usually called “intrinsic variables” are simply properties in the external environment of every Elementary Control Unit in the perceptual hierarchy, special only in that they are in the interior of the organism, and that perceptions of discomfort (and I might add, of good feelings as well) are simply the perceptual correlates of input patterns from that environment.
This really changes nothing but the definition of an intrinsic variable, which would change from being an internal variable that has been effective in assisting one’s ancestors to survive long enough to produce descendants to a definition that adds something along the lines of "and the value of which may contribute, along with inputs from sensory organs responsive to variations in the external world, to the value of a perception in the perceptual control hierarchy’.
Perhaps I have misunderstood the way in which the Powers reorganization system is supposed to work or the distinction that separates intrinsic variables from other kinds, and consequently, that my proposal attempts to solve a problem that is already adequately dealt with in the present system. If so, I await enlightenment.
I don’t think the issue is solved at all, which is why I suggested up front that it is an aspect of PCT worth some thought, and if feasible, experiment.
Earlier I talked about temperature as an intrinsic variable for thermodynamic reasons, but let’s invert our appreciation of the organism. We, as users of the body, think of it as an entity carried around by a physical world, buffeted by the actions of the physical environment and by the actions of other bodies. But we could take the opposite view, and think of our bodies as structures built by microorganisms to provide them with the necessities of life. After all, there are orders of magnitude more of them than our bodies have cells, so why should the health of our bodies have priority in our analysis of how we function?
When a “bad” bacterium invades our microbiome, there may be a conflict between its armies and those of the bacteria types that had long-established efficient trading relationships. The effect on our bodies are what we call “illness”, but which they might call “getting God to be on our side” by altering the temperature of the medium in which the war is conducted. Their “prayers” might be the production of toxins that change the reference levels of intrinsic variables as you suggest. Are those external or internal inputs, as seen from the gross viewpoint of the body? Hard to tell. Fanciful notions, for sure!
How does reorganization work? I don’t know, but I am sure that it is not a “random walk” through the entire control hierarchy’s multitude of tunable properties. Perhaps looking upward from the viewpoint of a bacterium might help in understanding it.
Neat!Â I like the idea that I might be protected by an army of â€œgoodâ€? bacteria – ones with which we have established mutually beneficial relationships – that help us to figght off destructive ones.Â Of course, our own cells have evolved their own defensive mechanisms, which often involve self-sacrifice â€œfor the good ofâ€? the organism of which they are a part.
Getting back to reorganization, a major feature (and advantage) of Billâ€™s conception is that it provides the mechanism by which the perceptual control hierarchy self-organizes.Â An important question in this regard is the extent to which the systemâ€™s development requires such a mechanism, as opposed to being structured primarily through genetically orchestrated means (e.g., neurons growing toward sources of certain biochemical agents being released by the target structures).Â Taking the reorganization process as the primary developmental mechanism would seem to suggest that individuals would differ widely in their ultimate neural organization, yet despite individual differences, individual brains are highly similar, having developed the same basic structures and connections, so much so that one can identify and label them in any given example (e.g., you and I both have a hypothalamus, similar in structure and function to the hypothalamus of every other normal human brain).Â And I can go further: there are homologous structures in the brains of every other mammal (e.g., mice have a hypothalamus and it performs much the same functions as ours).
Having said that, one cannot entirely discount the influence of environment on development.Â For example, the normal development of the visual system requires structured input from the eyes.Â Those who have been born with cataracts that have not been corrected until adulthood may never learn to see properly, because the analytic structures of the visual brain that create our perceptions of recognizable objects etc., either fail to develop or deteriorate when the developing system fails to receive the right kinds of inputs.Â But development may still be primarily under genetic control, although the developmental mechanism may require at least to a certain degree certain structured sensory inputs to guide the formation of â€œusefulâ€? connections once the gross structures are in place.Â (Could this be where reorganization works its magic?)