Real and Perceived Realities

Summary: Real Reality is what IS, Perceived Reality is what we perceive to be. They differ, but we can control only what we perceive. If that departs too far from Real Reality in the way it changes as we act on Real Reality, our control will not be very good. The ability to control well attests to there being something “out there” that corresponds to what we perceive in how it reacts to our actions. All we can know is that the “bundle of properties” we perceive an entity to have correspond more or less well to a “budle of properties” in Real Reality.

@@@@@@long message follows@@@@@

Every Living Control System LCS or (organism) lives in an environment. This environment is what offered Hamlet “the slings and arrows of outrageous fortune”. It is where a bacterium finds its food, and it is what allows a tree to take in CO2 and exhale oxygen (or so we perceive it if we believe what we are told). We cannot escape our environment. It is “Real Reality” (RR).

Every organism lives in another reality, a perceived reality (PR). The perceived environment of a bacterium or a tree is presumably much simpler than that of a human, and the perceived reality of a human is presumably a lot simpler than RR. The scientific tools we perceive to exist, even if we don’t ourselves use them, tell us for example that RR contains galaxies whose light was emitted billions of years ago, and that it contains invisible atoms and molecules, some of which emit radiation that could kill us, none of which we can perceive. They exist in our PR because other people who we trust tell us they exist in RR.

But do we have evidence that those people in our PR actually exist in RR? Good stage performers are able to allow is to have in PR persons who are not the actors who portray them. In our PR, on stage there is an actor, but also in our PR, somewhere not in the stage another person exists in a different version of RR. In fact, is there any evidence that anything, anything at all, in our PR corresponds to something in RR?

Yes, there is, and that evidence is provided by the possibility of controlling our perceptions. In the absence of control, it is just abstract philosophy. When we can effectively control a perception, the means of doing so are entirely in RR. But just what is in RR that allows us to control our perceptions? We cannot know more than that whatever it is, when we perceive ourselves to be doing X, something consistently affects our PR. Using this, we can learn how to make a perception consistently change in a particular way by changing X into X1, X2, … until our actions cause the effects in PR that we want.

We can learn in this way constellations of effects that we can manipulate in RR, even if we cannot know whether those effects occur because in PR there are Real Things that match things we perceive in PR, some omniscience and omnipotent being wills the effects to be thus and so or because a vast army of trained bureaucratic gnomes follows orders passed to them through a RR version of the internet.

The mystery then is why we perceive objects in PR, if we have no evidence for their existence in RR.

Norbert Wiener addressed a related issue in the preface to “Cybernetics: The human use of human beings” (at least in the second edition, 1965). The problem he addressed explicitly was one of electrical engineering, but it is easily applied to the RR-PR conundrum. Wiener postulated a “Black Box” (BB) to which an observer/experimenter had no access except through two sets of terminals, one (input) through which E could insert signals and one (output) that produced signals E could measure.

By judicious choice of signals to insert and to observe E could determine relationships among input patterns and output patterns. E could never discover how these relationships were produced inside the BB, but in a “White Box” (WB) E could perhaps design a network of processors that would produce the same set of relationships between its input and output patterns.

The analogy between the BB and RR should be obvious, as should the analogy between E’s problem and two PCT problems often considered independently: Reorganization and the “Test for the Controlled Variable” (TCV). In the TCV, E wants to find one small thing (the controlled perception) in one infinitesimal part of RR (S’s internal processing). Of course, before getting to this point, E has undergone a lifetime of reorganization on top of billions of years of evolutionary “reorganization”, and has built a WB network —a “theory” — of what is in S’s brain, namely a control hierarchy.

How did E reorganize to the point that it was possible to perceive a WB model that performs much as does S in tests? Wiener’s approach was the same as we usually take for early infant development. He first looked for simple, perhaps in inputs to only one terminal, patterns that resulted in consistent effects at one or more output terminals. A baby may wave its arms and legs around apparently at random, but the baby has sensors in the limb that allow it to connect “this” feeling in the limb (output from RR) to “that” movement of the limb (input to RR).

Wiener could readily describe in his WB simple processors (first-level mini-white boxes) that could produce relationship effects at its output (inputs to the experimenter) that matched the simple relationships the BB was observed to produce. The baby likewise could learn to produce some movements at will by linking the feeling with the observation (in the limb, of course, not by analytic thought).

The next stage of Wiener’s procedure was to examine relationships among relationships. When two or more input relationships altered in consistent ways with respect to observed output relationships, Wiener could produce much the same effects by designing a mini-WB that took inputs from his first-level mini-WBs and produced as outputs signals that matched those observed from RR. In reorganization, we would call this the development of a second level of control that used already reorganized first-level structures to produce actual perceptions and output actions, the latter being observable from outside the organism’s body.

Wiener described the procedure as a mathematical procedure to finding ever closer approximations to more and more complex effects produced in his BB and reproducing those approximations as mini-WB processors whose complexity was in large part in the connectivity of the network of mini-WBs, very much as we see the hierarchy of perceptual control structures reorganizing to ever higher-levels of the hierarchy. But NONE of Wiener’s mini-WBs, at any level, were supposed to do what they did because some mini-BB in the Black Box existed to do the same job. In fact, the bureaucracy of industrious gnomes could easily produce the same effects as the network of mini-WBs.

Earlier, I asked what was at that point a rhetorical question "is there any evidence that anything … in our PR corresponds to something in RR?", to which we can now offer a serious semi-answer: “Yes, there may be”. Whether that “may be” can be turned into an “is” depends on how one defines
“anything” and “something”. What we have in every mini-WB except those at the first level, and in every non-first-level perception, a complex of effects produced by the BB or RR respectively. As we perceive them, or as the mini-WB operates on them, those effects are a bundle of “properties” — what the higher-level mini-WB or perceptual “entity” does when influenced in an appropriate way. It is how the “entity” responds to input.

In the larger WB, the Analyst knows exactly what is going on, because the Analyst designed it to do what it does (whether a human analyst has the mental capacity to deal with a large network of mini-WBs is another matter). Some concatenation of signal paths is highly likely to exist in RR, whether it is in the form of an entity or a pile of bureaucratic orders that arrive on the desk of one gnome. A non-first-level mini-WB that produces effects that match something the B does that is not in the individual next-lower-level mini-WBs is very likely to have some corresponding processor in RR.

If Wiener is to be taken as any kind of a guide, reorganization of a perceptual control hierarchy is likely to create perceptions of entities that have some correspondence to something in Real Reality.

Likely but not necessarily. It may not have escaped notice that I have been describing not only reorganization of an individual hierarchy, but also reorganization of Science, which sometimes delves into quite low-level regions of the functional hierarchy.

The mini-WBs of Science may not do the same things as the processing in RR any more than are the mini-WBs of our individual reorganizations, even at levels on which much has been successfully built. For example, Gravity wasn’t a consideration until Newton. Objects had weight as one of their properties, but that had nothing to do with any other object. Newton introduced the idea of gravity as a force field that worked between objects rather than within one (creating a new lower, but not first, level mini-WB entity in the Science WB network). Newton’s gravitational force field modeled RR very well. Then Einstein came along and said that gravity was not a force field but a side-effect of warped space-time — causing a whole lot of disruption of lower-level Scientific entities.

To return to the implications of all this for the Science of PCT, what we could, in principle know with any desired degree of approximation is how effects in RR can be bundled into “properties” of “entities” that we perceive as objects, or as concepts such as the shape of a circle or the literary skill of an author. Those “properties” are very likely to be bundled similarly to permit effects in RR. Nevertheless, what we perceive in RR only behaves like something in RR, without necessarily corresponding to anything that exists in RR other than the effects bundle.

When we control, it is only by influencing some perceived thing in the environment that changes the way something changes in RR that we can do it. The “mini-WB” network we have built is crucial, and that network, we build by reorganization.

I hope that at least some of this makes sense1

2 Likes

It does.

Beautiful and thought provoking write up Martin.

MT: Summary: Real Reality is what IS, Perceived Reality is what we perceive to be. They differ, but we can control only what we perceive. If that departs too far from Real Reality in the way it changes as we act on Real Reality, our control will not be very good.

RM: This has nothing to do with the PCT model of control. In PCT there is no notion of Perceived Reality departing from Real Reality. And there is certainly nothing in PCT that predicts that control will “not be very good” in proportion to the degree to which perceptual reality “departs” from real reality. If this is predicted by your model of control, I’d like to see the data that shows that this is the case.

Best regards

Rick

This is pretty radical, if I interpret it as you intend. But it is a rather ambiguous, so I would like to ask for clarification as to which of several possible interpretations you intended. I start with what I think are two agreements, and follow with a non-exhaustive list of six possible interpretations of your intent.

(Agreement 1)To start with, let’s be clear that you are not proposing solipsism, in which there is no “Real Reality” at all, because the entire Universe, including you, exists only in my mind. I do not (consciously) subscribe to solipsism, and in the past you have said that you don’t either.

(Agreement 2) I agree totally with what you say, if you refer only to the ultra-simplified minimalist “Bauhaus” model of a control loop that is typically used with great success in PCT experiments and simulations.

Since you refer not only to “the PCT model” but to Perceptual Control Theory itself, or perhaps to the use of PCT in the everyday world, I have to interpret your comment as applying more generally to all lesser levels of model simplification as well. So the following comments should be interpreted in light of these two agreements: that we are not being solipsistic, and that we are referring to more naturalistic PCT models than the minimalist “Bauhaus” control loop model, on which we agree. Some further questions then arise.

But first, we should get out of the way your next sentence: “And there is certainly nothing in PCT that predicts that control will “not be very good” in proportion to the degree to which perceptual reality “departs” from real reality.” Even the most minimalist PCT model says that if you control X perfectly and observe Y, you will find that Y will be controlled imperfectly if Y is not identical to X. That’s what you seem to be disputing, a position I think unsupportable.

Now for my questions about how you intend the ambiguous sentences to be interpreted.

(1) Are you saying that there IS no Real Reality? I think not, because that implies solipsism, which we both have assumed is not a viable hypothesis for the purpose of Science.

(2) Assuming you accept that there is a Real Reality, are you saying that PCT ignores its existence? If so, PCT describes only control of imagined variables.

(3) Are you saying that when we act, we do not act on real reality? If not, on what do we act?

(4) Are you saying that the influences that vary what our sense organs produce are not influences from real reality? If they are not from real reality, where do these influences come from?

(3+4) Are you saying that the environmental feedback path of a control loop does not pass through real reality? If it does not, and all control is not of imagined variables, what other non-solipsist possibility do you intend to imply?

(5) “In PCT there is no notion of Perceived Reality departing from Real Reality” seems to imply that you think there exists a Real Reality, but that nothing in it is hidden, because our perceptions are OF Real Reality. Is this your intended meaning?

(6) If you think our perceptions are OF Real Reality, how does that square with your usual claim that the CEV is only a projection into the environment of the Perceptual Function, and its magnitude (the CV) is only that of the perception?

There are other possible interpretations of these two sentences, but that’s, I think, enough to explain my puzzlement and my use of the word “radical” at the start of my message. Maybe if we could get the interpretation of these two sentences to be as you intended, we could begin a productive discussion within the larger community, rather than beginning an unnecessary escalating dispute. (At least if you are not controlling for having a dispute :slight_smile: )

RM: This has nothing to do with the PCT model of control. In PCT there is no notion of Perceived Reality departing from Real Reality.

MT: This is pretty radical, if I interpret it as you intend.

RM: No just basic PCT.

MT: I agree totally with what you say, if you refer only to the ultra-simplified minimalist “Bauhaus” model of a control loop that is typically used with great success in PCT experiments and simulations.

RM: I am referring to the model of behavior described by Powers in B:CP, a model that has come to be referred to as PCT.

MT: But first, we should get out of the way your next sentence: “And there is certainly nothing in PCT that predicts that control will “not be very good” in proportion to the degree to which perceptual reality “departs” from real reality.” Even the most minimalist PCT model says that if you control X perfectly and observe Y, you will find that Y will be controlled imperfectly if Y is not identical to X. That’s what you seem to be disputing, a position I think unsupportable.

RM: I presume that X and Y refer to what in PCT are typically called q.i (controlled variable or quantity) and p (perceptual variable). in PCT q.i and p are always identical; p is the theoretical variable that correspond to the observed fact that q.i is controlled.

RM: Your mistake, which is a very common one, is to imagine that Real Reality is made up of objects like trees, tables and people and that perception is a variably accurate mental representation of this reality. It’s the “through a glass darkly” model of perception. This is not the PCT model of perception.

RM: In PCT, perceptions are variables that are analogs of variable aspects of the sensory effects of Real Reality (the environment in the PCT Model). Whatever aspect of reality the perceptual variable is an analog of is the aspect of reality that will be controlled – if it can be controlled. If the perceptual variable is an analog of an aspect of reality that can’t be controlled – or can’t be controlled well-- it is still a perfect analog of that aspect of reality. That is why I say there is no such thing as a discrepancy between Real Reality and Perceived Reality. Perceived Reality is not a perception of Real Reality; Perceived Reality is (per PCT) an analog of aspects (functions) of Real Reality. Perceived Reality (perceptual signals, p) consists of theoretical variables that are exactly the same as the aspects of Real Reality of which they are an analog.

RM: If you don’t agree with this analysis (which I’m sure you don’t) and you would like to convince me that it is really the case that control is degraded to the extent that Perceptual Reality (the perceptual variable, p) is not identical to the aspect of Real Reality that is controlled (the controlled variable, q.i, then you’ll have to show me how you know this – show me with experimental; data, that is, not philosophical blather.

Best

Rick

Rick,

I presume you do not intend to answer my six questions. I do, however, have one comment of what you write.


rsmarken

              January 5


RM: I am referring to
the model of behavior described by Powers in B:CP, a model
that has come to be referred to as PCT.

I wish you would refer to the model of behaviour described by Powers in B:CP, but you don't... see the basic comment below.
            MT: But

first, we should get out of the way your next sentence:
“* And there is certainly nothing in PCT that predicts
that control will “not be very good* ” in
proportion to the degree to which perceptual reality
“departs” from real reality.” Even the most minimalist
PCT model says that if you control X perfectly and
observe Y, you will find that Y will be controlled
imperfectly if Y is not identical to X. That’s what you
seem to be disputing, a position I think unsupportable.

          RM: I presume that X and

Y refer to what in PCT are typically called q.i
(controlled variable or quantity) and p (perceptual
variable). in PCT q.i and p are always identical; p is the
theoretical variable that correspond to the observed fact
that q.i is controlled.

What you are saying here is that the PCT model of the single control loop
does NOT include a Perceptual Function (or Perceptual Input
Function). The model described by Powers DOES include such a
function, even in its minimalist (what I called “Bauhaus”) form. The
model described by Powers is, by Powers, described as a gross
simplification that he initially assumed would work much less well
in simulation than it has turned out to do. Powers discussed many
variants of less simplicity over the course of his published and
unpublished writing. Howsoever that may be, in the Powers model p
does NOT equal qi, though it does have a one-to-one functional
relationship with qi.

There’s no point in trying to comment on the rest of your message
until you answer the six questions. Actually, it would help if you
actually read the opening message on this thread. That might save
the kind of pointless back-and-forth we have so often seen on
CSGnet. The Forum allows you to go back in the thread very easily
and double-check what you may have thought you had read.

I have one more very short comment.

          RM: Your mistake, which

is a very common one, is to imagine that Real Reality is
made up of objects like trees, tables and people

As you are well aware from our CSGnet discussion, I do not imagine
Real Reality is anything like that, though I do not deny that it
could be. If I did think what you claim, why would I continue to say
that Real Reality might equally well consist of a big bureaucracy of
industrious message-passing gnomes, and that we can never know that
it does not?

Martin

MT: I presume you do not intend to answer my six questions.

RM: I might after you show me the data demonstrating that control degrades in proportion to the discrepancy between perception and reality.

RM: I presume that X and Y refer to what in PCT are typically called q.i (controlled variable or quantity) and p (perceptual variable). in PCT q.i and p are always identical; p is the theoretical variable that correspond to the observed fact that q.i is controlled.

MT: What you are saying here is that the PCT model of the single control
loop does NOT include a Perceptual Function (or Perceptual Input
Function).

RM: No, I am not saying that. I simply assumed that someone familiar with PCT would know that q.i = f(x.1, x.2…x.n) where the x.i are variables that make up Real Reality and that f() is a perceptual function.

MT: in the Powers model p does NOT equal qi, though it does have a one-to-one functional relationship with qi.

RM: I think you are referring to the fact that the perceptual signal is surely a non-linear function of q.i. So if q.i = f(x.1, x.2…x.n) and p= log(f(x.1, x.2…x.n)) then q.i has a one-to-one relationship to p but it does not equal p. But this doesn’t affect the ability to control. Or do you have some data that shows that it does?

MT: There’s no point in trying to comment on the rest of your message
until you answer the six questions. Actually, it would help if you
actually read the opening message on this thread.That might save
the kind of pointless back-and-forth we have so often seen on
CSGnet.

RM: I don’t think conversations on CSGNet have been pointless; indeed, I think they have been quite productive, as has this one. I think I now know what you mean when you say that p can deviate from q.i. You are just saying that, from what we know of psychophysics and neurophysiology, the actual perceptual signal, p, produced by the perceptual function is likely to be non-linearly (but monotonically) related to q.i, which is surely true. We just don’t incorporate that fact into our models because it makes no difference; the model controls just as well when it controls log( p) as when it controls p. It doesn’t affect the fit of the model to the data or the ability to determine what perception is being controlled (because of the one-to-one relationship of p to q.i). But if you have some evidence that it does make a difference I’d be very interested in seeing it. You are allowed to present evidence in this trial;-)

RM: Your mistake, which is a very common one, is to imagine that Real Reality is made up of objects like trees, tables and people

MT: As you are well aware from our CSGnet discussion, I do not imagine
Real Reality is anything like that, though I do not deny that it
could be. If I did think what you claim, why would I continue to say
that Real Reality might equally well consist of a big bureaucracy of
industrious message-passing gnomes, and that we can never know that
it does not?

RM: I think there is pretty strong evidence that it is NOT a big bureaucracy of
industrious message-passing gnomes. The evidence is that Real Reality is very much like what the models of physics and chemistry say it is. That’s the model of Real Reality – the model of the environment – that we use in PCT.

Naive Realism and Constructivism

Traditionally, philosophers of perception have been divided between those who believe that “what we see is what we ge”, that we directly perceive real reality, and those who claim the the reality we perceive is one we construct for ourselves. In PCT, these claims can be translated into statements about perceptual functions. The “naive realist” would argue that the perceptual function simply maps a function in Real Reality (RR) and never deviates from what is directly enforced by RR, whereas the “constructionist” claims that what we perceive is merely a projection into an apparent environment created by the perceptual function itself.

I do not think PCT lends itself to either of these positions. Intsead, it leads to the conclusion that both are partly right. While I agree with the constructionists that what we perceive to be in the environment is a projection of the forms of our perceptual functions, I also partially agree with the realists (but not the naive realists) in saying that the nature of Real Reality determines how perceptual functions evolve toward becoming better and better representations of components of real Reality.

Two Parallel Problems of Living Control

This message continues to use the Wiener “Black-Box White-Box” process as an analogy. Message 1 of this thread discussed Real Reality (RR) as analogous to a “Black Box” (BB), and to our perceptual functions as Norbert Weiner’s “White Boxes” (WBs), each of which emulates something that happens inside a Black Box. Wiener was talking as an electrical engineer who wanted to investigate what the Black Box did, given that he could not measure anything about the Black Box except the signals sent to or emitted from two sets of terminals on its surface. All Wiener’s hypothetical investigator could do was to try to create a White Box with the same sets of terminals, that generated the same outputs as the Black Box when it was presented with the same patterns of signal inputs.

Wiener’s problem maps onto two separate problems related to living control systems. One is the problem faced by a psychologist or physiologist who wants to know what processes operate within a living organism, and does so by building models that can be realized in simulations. By successive refinement, the models produce results more and more like those of the living organism in their interactions with the environment. In this first problem, the BB is the living organism and the WB is a functional model of the organism made by the investigator, a psychologist or physiologist.

The second problem, which is the one that most concerns this message, is that of the living organism that has to survive in its interactions with Real Reality (RR). The White box in this problem is the set of processes inside the organism directly concerned with the organism’s interactions with its sensors and effectors, while the “investigator” is the reorganization process(es) that modifies those interactions and builds functions that use the previously built simpler functions.

In basic PCT, these mini-WB “functions” are Elementary Control Units (ECUs), each of which consist of a Perceptual Input Function that provides the perceptual variable to be controlled, a Reference Input Function that determines how the reference value is built from higher-level output variables (if there are any), a comparator and an Output Function. The internal part of every control loop has such an ECU as its operational component. All else, at lower WB levels, is, to this ECU only communication between the perceptual function or output function and Real Reality , eventually creating a loop whose environmental feedback path passes only through RR, not PR.

Wiener’s “investigator” engineer is, in principle, capable of discovering whether the functional properties of the Black Box change over time. If they do, the WB that the engineer builds must emulate that property of the BB just as it emulates how the BB relates signals at its inputs and outputs at any defined time. Wiener simplified the general problem by specifying a procedure whereby the engineer could build up a WB of increasing complexity, changing the interconnections and even the functions of a growing network of “mini-WB” components, beginning with very simple ones that work directly with small and consistent patterns of relationships among the signals between the input and output terminals, and building on those to produce mini-WBs that perform ever more complex functions and connections that change dynamically in tune with RR.

A “complete” WB, however, would have to match these complex functions without needing to be rebuilt by an outside engineer. It would reorganize itself to change along with changes in functions and functional connectivity it can detect in RR. The reorganization process would need to be an intrinsic part of the WB.

Wiener’s hierarchic network consists of levels of mini-WBs that take their inputs from lower-level mini-WBs and send their outputs to other (or possibly the same) lower-level mini-WBs. In other words, if the engineer does not use connections within a level of mini-WBs, Wiener’s investigator reorganizes the internal structure of the WB in a way that is difficult distinguish from maturation of the Powers control hierarchy.

We cannot know whether Wiener’s work might have had any influence on Powers, but it doesn’t matter either way. What does matter is that the structural similarity allows us to map the Wiener engineering problem more exactly onto the living organism problem. If Weiner’s BB exhibited control behaviour, the network of mini-WBs would probably come to look like Bill’s hierarchic network of control units, which the psychologist investigator investigates with the objective of producing ever more faithful emulations.

As with any WB-BB emulation investigation, even though the functions of the mini-WBs and of network in which they are embedded may come to match properties of RR increasingly exactly, there is no guarantee that the mechanisms are the same in the mini-WBs as in their corresponding mini-BBs. A hierarchy of mini-WBs might be realized in software, in hardware, in mathematical equations, in hydraulic microcircuits, or what have you. Functionally it makes no difference.

What one can say, however, is that if the mini-WBs in a part of the network faithfully emulate corresponding mini-BBs in a part of the unknowable Black Box, the connections among them must also correspond to connections in that part of a network in the BB. The two networks of interconnections would be essentially identical, but only in their functionality, not in their mechanism. The bureaucratic network of industrious message-passing gnomes would serve, as would a massive parallel computer or an organic neural network.

Much of the above applies mutatis mutandis to both of the problems identified at the beginning of this message, the problem of the psychologist trying to understand the organism or that of the organism trying to survive in a Real Reality of unknowable content. Both can approach a solution to their problem only by reorganization that produces mini-WBs that better emulate “stuff that happens” in their particular BBs, and also arranges the hierarchy among these mini-WBs so that their network better allows the mini-WBs individually to do their jobs.

The rearrangement of the hierarchy of mini-WBs will be ineffective if the lower-level mini-WBs inaccurately represent the current functioning of a corresponding mini-BB. One could not live in a reality that changes its “Laws of Nature” rapidly and capriciously. Some things must be stable over many generations, others over much or all of a lifetime, if they are to support a network of interacting functions that allows the organism to construct stable low-level mini-WBs on which a useably stable functional hierarchy can be built level by level to enable complex behaviours.

So what are the “objects” we perceive?

What this all seems to imply is that we can identify in the Black Box of Real Reality only functional relationships, linkages among properties that tend to occur together in stable structural patterns or “bundles”. If this is so, why would we perceive any objects at all in our perceptual reality? Consciously, our Perceptual Reality (PR) largely consists of objects, though we also perceive some relationships among the objects. We must ask what actually constitutes an object in our conscious perception. Within the control hierarchy, the perceptual signals are not of objects. They are only scalar variables, the values of various properties of the object, as filtered by many perceptual functions.

Objects, if the Powers hierarchy of scalar control loops is correct, must be conscious constructions built from sets of perceived properties that somehow persistently “go together”. Indeed, “objects” seem very much like vector-valued perceptions, outputs of perceptual functions that produce more than scalars.

Within the Powers hierarchy, there are no vector-valued perceptions. But there are “category perceptions” that segregate perceived structures of scalar property perceptions into classes such as “tree” or “penguin”. I think it is such “perceived property” bundles that create conscious perceptions of objects. So, we ask in light of the preceding WB-BB discussion, what is a “property”. I propose that it is a name for a mini-WB that takes a particular input pattern and performs some function on them, which it distributes to other parts of the network.

Many “objects” can have the same “properties”, though at different levels, perhaps. For example. solid objects are perceived to move coherently when pushed. They have the property of “solidity”. Some solid object move as a unit more precisely than others; these have a high value of “stiffness” and a low value of “flexibility”. One can present myriads of other such examples, but in all of them the “property” defines a functional relationship between what the object does in its perceived environment when it is affected by some perceived influence from elsewhere — usually another perceived object. I think it is reasonable then to think of any object as being no more than a coherent bundle of perceived properties, the inputs to the perceptual function that creates the perception of the likeness of the object to a tree, a clock, or an eagle.

In this sense, and in this sense only, we may expect Real Reality to incorporate functional bundles resembling the property bundles of perceived objects that have reorganized into some kind of perceptual stability. In that sense, it may be not unreasonable to suggest that a perceived object is likely to have a counterpart bundle of properties in Real reality, though the constituent parts that implement the properties may be very different.

Summary of this Message 2

Reorganization in PCT suggests that, as a professor told our class 60 years ago: “When two schools of thought each claim the other is wrong, probably they are both right.” What we perceive as being in the environment is constructed by each perceptual function, but at the same time, each perceptual function is subject to the processes of reorganization, and tends over time to approximate reality ever more closely. Realism is not “Naive”, and works in tandem with constructivism to prolong the life of living control systems in a potentially hostile Real Reality environment.

There may be a future “Message 3” that extends the thread to dynamic and social environments, but no promises. Comments from the community on Messages 1 and 2 would be welcome, especially if they are rationally critical or ask reasonable questions.

Thank you Martin, this Message 2 answered most of the comments and questions I had after the first message. Some comments or thoughts are still left:

I would first like to remind about the concept of Umwelt by Jacob von Uexkull which just means the perceived environment of the organism (I have shortly studied it in http://ojs.utlib.ee/index.php/sss/article/view/SSS.2018.46.4.02). The Umwelt of an organism consists of its functional circles which roughly correspond to the control units or (sub)systems. Typically there are four important functional circles in the Umwelt of an organism, namely those of physical medium, food, enemy and sex. All other things, effects or properties in the environment remain unperceived for the organism - they have no meaning to it.

In spite of its visionarity Uexkull’s theory remained, however, very simplistic when compared to the later developed theory of Powers. But in relation to the topic or the title of this thread it has the same teaching: we do not perceive everything in our environment but only such “things” for which evolution and life - i.e. reorganization - has developed perceptual functions for us. Everything else remains unperceived independently of the reality of those unperceived “things”. I think this kind of filtering or screening phenomenon is more important and interesting division between the real and the perceived environment than the question of the correspondence between the q.i. and the p.

From the Powers’ theory we can also infer the that we “know” (in a certain meaning of tacit knowledge) more about the (real) environment than we can perceive about it. An assumed control system or unit perceives only the controlled (or controllable) perception (or rather the q.i.). It does not perceive the output it creates and the feedback path through which the output affects the (q.i. and the) controllable perception. However the succesful control of course requires that the output happens in a right way and that requires “know how” about those unperceivable aspects of the reality.

Good this year!
Eetu

Rick, you’re talking about a model of good control. What happens when control is not good? You’re overlooking PCT proposals about problem-solving and various forms of learning, the PCT model of reorganization, and the PCT explanation for the development of inherited control systems by a similar trial-and-error process over successive generations of surviving offspring (evolution).

We have little data on problem-solving, learning, and reorganization, and less for the extrapolation to evolution. Limited data did not stop Bill from proposing those aspects of the PCT model as a call for research that would produce data—or for that matter, limited data didn’t stop him from writing B:CP and calling for (and doing some) research that would produce data.

Here’s a way we might approach this. Consider the variety of ways in which control can be degraded. For example, the Asperger’s spectrum may involve limitations or lack of perceptual input functions for controlling social relationships. Then for each scenario for poor control, consider the changes a control system might make to improve control. Problem-solving at the planning level is one sort of change a control system can make within itself. Related to this are the various ways to resolve internal conflict. Reorganization is another sort of change a control system can make within itself.

An extremely important part of the PCT model has been given very little attention. In his account of emotions, Bill proposed that there are two branches of the hierarchy, the “somatic” branch and the “behavioral” branch, separating at the Event level. Here’s his diagram in the file EmotionModelV2.JPG.

To my knowledge there is no PCT research on the interactions between these two branches of the hierarchy. There may be quantitative data in the neuroscience literature. I haven’t looked at it since writing my LCS IV chapter. There I cited

  • Cardinal, R., J., Parkinson, J. H., & Everitt, B. (2002). Emotion and motivation: the role of the amygdala, ventral striatum, and prefrontal cortex. Behavioral and Cognitive Neuroscience Reviews 26.3:321–52
  • Lieberman, M. D., Eisenberger, N. I., Crockett, M. J., Tom, S. M., Pfeifer, J. H., & Way, B. M. (2007). Putting feelings into words: Affect labeling disrupts amygdala activity in response to affective stimuli. Psychological Science 18.5:421–428. [http://www.scn.ucla.edu/pdf/AL(2007).pdf]

This is the evident basis for a common experience that has come to be called an amygdala hijack. As a special case of conflict, interaction of the two branches can ‘hijack’ the means of control. We can restore better control by re-establishing appropriate communication between the somatic and behavioral branches, but this possibility is often neglected. My informal metaphor is “emotion is like water, it fills the available container”: if I’m unable to control some important perceptions—a medical crisis, for example, or that of a loved one—the somatic hierarchy (which operates without our conscious awareness) sustains a readiness for action. We may be conscious of a perception of anxiety or of frustration or anger as an interpretation of the body’s readiness for action. My experience, at least, suggests that this readiness for action all too readily results in high gain overkill control in some other conflict where I can control better, accompanied by expressions of impatience, anger, or the like that later I recognize were not appropriate. It’s kind of like grabbing low-hanging fruit just to grab something. The more important failure to control is still there. So people think they feel better when they “vent” (Sigmund Freud’s steam engine metaphor for cognitive process, still alive and kicking), but like the proverbial return of hunger after eating Chinese food it doesn’t last.

Now I’ve done it! Take a good look at old Sigmund scowling there with a black Fu Manchu moustache. Test the hypothesis: is a good chuckle good medicine?

MT: Traditionally, philosophers of perception have been divided between those who believe that “what we see is what we get”, that we directly perceive real reality, and those who claim the the reality we perceive is one we construct for ourselves…

RM: For some reason this jogged my memory and I remembered that Powers conceived of Real Reality as the world we experience – the world of birds, bees, flowers and trees, etc. The fact that what we experience is a perceptual representation of some other reality – what in PCT we call the Environment --is an inference based on other experiences, such as optical illusions (like the apparent bending of a stick in water which we can tell by other means to be straight), regularities of relationships between experiences (such as those found in scientific experiments) or the constraints we experience on how we can control what we experience (such as the fact that some jars require more torque to open than others). So from a PCT perspective, what we call perception is Real Reality and what we have been calling Real Reality is actually inferred reality.

MT:… the nature of Real Reality determines how perceptual functions evolve toward becoming better and better representations of components of real Reality.

RM: Since we don’t know what Real Reality (PCT’s “Environment”) is I don’t think we can say that evolution has made our perceptions better representations of its components (we don’t even know if really has components). I think what we can say is that evolution provides organisms adaptive ways of perceiving whatever is the basis of those perceptions.

RM: My recall of Bill’s conception of real reality as perceptual experience – it’s all perception – helped me realize that controlled variables exist only as perceptions in the experience of the observer of a control system. The idea that these controlled variables exist as perceptual signals in the controller is a theory that explains the existence of these controlled variables.

RM: What is important about this for me as a researcher is that I now understand that all controlled variables are perceptions that are defined in terms of perceptions. Even the lowest level controlled variables that we think of as being a function of variables in real reality – variables like brightness and loudness – are actually defined in terms of perceptions – the perception of the “meter readings” of devices that we presume are measuring variables in the environment, such as luminance and sound pressure level. So for a science of control, the data we need are descriptions of controlled variables in terms of our own perceptions of those variables. Once we have a nice catalog of descriptions of this sort we can start looking to see if controlled variables fall into the classes and hierarchical relationships hypothesized by Powers.

RM: In PCT there is no notion of Perceived Reality departing from Real Reality. And there is certainly nothing in PCT that predicts that control will “not be very good” in proportion to the degree to which perceptual reality “departs” from real reality.

BN: Rick, you’re talking about a model of good control. What happens when control is not good?

RM: Several possibilities. The controller may not be controlling the perception you think they are controlling, and controlling it perfectly well. Or the person may be trying to control the perception yo think they are controlling, they just haven’t learned to do it well yet. But there is nothing in PCT that says that good control results from controlling a perception that matches physical reality; nor is there anything in PCT that says poor control results when the perception controlled departs from physical reality. That’s because we have no access to physical reality; all we have is our perception.

In many places in CSGnet threads you have with admirable clarity described the relationship between perception and reality according to the PCT model, for example in

  • Rick Marken 2019-12-25_12:57:58 [to Martin Taylor]
  • Rick Marken 2019-09-04_13:59:57 & 2019-09-04_15:24:06 [both to Fred Nickols]
  • Rick Marken (2017.02.26.1650) [to Eetu Pikkarainen]

In these posts, you have restated the position that Bill reached, but the formulation is yours.

I’ll start with a brief, informal summary in your second post to Fred (cited in the list above):

Rick Marken 2019-09-04_15:24:06

Physical reality is the current models of physics; “level of water” is a perception that is a function of that reality. Actually, of the sensations produced by that presumed reality.

I think you would agree that the models of the physical sciences are not actually reality, they are also perceptions. (Elsewhere you say “physics and chemistry”, so I say “physical sciences” rather than physics.) We use the perceptual variables of the physical sciences as though they are reality because the work of science is to make models of reality that are as accurate as possible, and as far as we know the current models that are generally accepted in a field are the best that we have for those aspects of reality that are the concern of that field. The theoretical model is a white box in respect to the black box of reality, and insofar as inputs and outputs of the white box ‘behave’ like experimental inputs and outputs in the black box we believe that the structure of objects and relations in the white box represents the structure of the black box—the objects and relations and the structure that they constitute being perceptions, of course.

So what is special about these models of the physical sciences. To “do science” is to participate with other scientists in collective control of the perceptions that constitute one’s field. In addition to system concepts and principles of a particular field, these collectively controlled perceptions include system concepts of science generally. Among the important principles underlying (=controlled under) the system concepts of science are that reality is knowable, and that we can control the accuracy of science models by “interrogating nature” with experiments, because experimental results can tell us which theoretical models and other perceptions of the science most accurately ‘match reality’. That’s the point of experimental method. In an experiment we can control alternative perceptions and observe that one can be controlled successfully in the experimental environment and the other cannot be controlled as well or at all. One model accounts for the results and the other doesn’t.

But the reason we adopt the perceptions of physical scientists as a surrogate for reality is not solely because scientists have tested their perceptions more rigorously than we ordinarily test ours. It is also because the hierarchical PCT model tells us that inputs from the environment first produce only intensity-level perceptions, not relationship perceptions such as the level of water in a container. Each intensity signal comes from one sensor cell that is sensitive only to a particular kind of energy in the environment (light energy, heat energy, pressure, …). Only a subset of the variables identified in the physical sciences can directly affect intensity signals generated by sensor cells in a human. Only that subset of physical variables can be ‘aspects of the environment’ of which perceptions are a function. It may even be that most of the physical variables identified in the physical sciences today are not directly perceptible. We perceive heat and cold, which are explained as the movement of molecules, but we do not perceive molecules, much less their movement.

To rephrase your CSGnet post to Fred, quoted above: “‘level of water’ is a perception that is a function of intensity perceptions produced by that presumed reality, and of sensation perceptions constructed from them, and of configuration perceptions constructed from those in turn.” ‘Level of the water’ is a perception of relationship—for example, the relationship between two configurations (that of the container and that of the water in it) or a perception of relationship between the surface of the water and the surface on which the container and one end of a ruler both rest. The functions that transform influences from the environment into a relationship perception include all of the input functions at intensity, sensation, and configuration levels that contribute perceptual signals on the way up to the input function for the relationship perception. From the point of view of a comparator perceiving and controlling that relationship, it is as though all of the sensors and comparators below which contribute signals to its input are in a sense parts of its complex perceptual input function. That is the complex function that transforms intensity signals directly influenced by factors in the environment into a perception of the level of the water. Only in that view could we say that a perception of the level of the water is a function of physical variables identified in the environment by the physical sciences.

In your post to Eetu, cited above, you say

control always involves acting to influence how environmental variables—variables outside the control system—affect the sensory inputs to that system.

We keep coming back to the variables that are actually in the environment because the loop is closed through the environment. It is not closed through perceptions, not even the perceptions that constitute the current models of physics. In the PCT model, in a diagram of a control loop, and in a computer simulation, physical reality is represented by the environmental feedback function. In a simulation of a motor task, the mathematics of the environmental feedback function is generally the mathematics of Newtonian physics.

Continuing your reply to Eetu:

The term “aspect” refers to a feature or characteristic of something, in this case of the physical (environmental) variables that physics and chemistry say are the environment outside our senses.

“Outside our senses” could be confusing. The physical variables affect our senses by causing sensor cells to generate intensity signals. They also affect instruments that scientists, engineers, and inventors have created to observe and measure aspects of the environment. Such instruments are also sensitive to aspects of the environment, such as radio waves, that are outside our senses in the stronger sense that we cannot perceive them without using such instruments to transform their variation to perceptible variables. These imperceptible physical variables appear not to be relevant to this discussion.

Continuing with your post to Eetu:

Qi is variable that is a function of these variables; it is not itself an environmental variable

The status of Qi is developed more fully in your first post to Fred, cited above:

According to PCT, when you are controlling the level of the water, you are controlling a perceptual variable, p, that is the “level of the water”. This variable is a function of environmental variables, v.i: p = f(v1, v2,…vn).

So when you control the perception “level of water” you are controlling a function of environmental variables, which requires affecting those variables in such a way that the perceptual variable is bright [=brought] to and maintained in the state you want it. Because I have perceptual functions in me that can produce the perception “level of water”, I am able to see that you are controlling the level of water in the glass. Of course, the level of water looks as “real” and “out there i the environment” as it does to you. But in fact, “level of water” is a perceptual variable for both of us.

Once you understand this you don’t have to get involved in this question about what is controlled, the perception or the environment. It’s the same thing; controlling a perception is the same as controlling an aspect of the environment defined by a perception function.

Qi is perceptual variable from several points of view. It is a quantity that is measured by the experimenter. It is a quantity that is determined by program code in a simulation. As a symbol representing this quantity it is a perception controlled by the theoretician as part of the model, as in the above passage. However, Qi is not perceived as such by the subject organism that the model or simulation represents, unless of course the subject is controlling the same quantitative measurement that the experimenter is and perceiving that as ‘the level of the water’.

The subject who perceives the level of the water perceives relationship, configuration, sensation, and intensity perceptions. She may be said to perceive those aspects of the environment that impinge on sensors so that they generate intensity signals. A quantity Qi for ‘level of the water’ may be perceived and measured by the experimenter who (as you say) constructs the same levels of perception because of a shared evolutionary history and similar learning experience. A programmer may take the experimenter’s data and write simulation code that generates Qi. But centimeters or inches of height are not present in the environment, they are perceptions derived from what is present in the environment. Physical sciences may say there is H20 plus some dissolved substances inside a concavity composed predominantly of SiO2, but that is irrelevant to what the subject perceives, a glass of water.

It is clear that you have wanted people to understand and agree with these principles of the PCT. We may have different preferences how best to formulate and express Bill’s insights. Whatever way we choose, words are ambiguous because (being common parlance) they are also used in contexts that are uninformed by PCT. For the present, now, let us set aside these quibbles about how best to express the fundamental principles of PCT clearly. The purpose of all of this preamble has been to establish that I understand and agree with these fundamentals. I believe Martin understands and agrees with them as well—he can speak for himself if he doesn’t.

These concepts are essential for understanding how a control system works to control in real time. Your replies would qualify as responses to what Martin wrote if Martin had been writing about how control systems work in real time. But what Martin was talking about is how control systems change so that they control better in the future than they did in the past. Your complaints amounted to telling Martin that he didn’t understand these fundamental concepts of PCT. I know that you have seen a lot of evidence to the contrary.

Martin began his post with this summary:

Real Reality is what IS, Perceived Reality is what we perceive to be.

I think you have no quarrel here.

They differ, but we can control only what we perceive.

“They differ” means that perceptions are essentially different in kind from the realities that affect our senses.

If that departs too far from Real Reality in the way it changes as we act on Real Reality, our control will not be very good.

This is a poor statement of a notion that is difficult to express simply. There is a misleading ambiguity in the word “depart”, which you may have interpreted to mean that a perception is gradually becoming different from the corresponding reality. The idea rather is as though looking across a number of pairings of perception with reality and seeing that some are more discrepant than others. This is of course impossible in any literal sense. You have identified the difficulty: there is no non-perceptual way to see the state of reality and see how much a perception differs from it. One way of expressing the idea behind this sentence is with the example that we control better with scientific concepts than with unscientific ones. If you believe that one’s economic and social fortunes are governed by invisible spirits that might help and might hinder, but you don’t know, and you can’t placate them or enlist their aid because they don’t care (a system concept of Voudun), you’re not going to control economic and political perceptual variables very well. Larry Harrison gives the case history of Haiti, where the Voudun religion is pervasive, vs. the Dominican Republic, with a heritage of British rule of law (The Central Liberal Truth: How Politics Can Change a Culture and Save It from Itself). The material evidence of relative ability to control economic and political perceptual variables, individually and collectively, on two parts of the selfsame island in the Caribbean, is quite stark. While controlling a system concept and principles of Voudun it makes no sense to plan for the future.

The ability to control well attests to there being something “out there” that corresponds to what we perceive in how it reacts to our actions. All we can know is that the “bundle of properties” we perceive an entity to have correspond more or less well to a “bundle of properties” in Real Reality.

In what follows this summary, Martin was not talking about experimentally causing a perception to depart from real reality. He was talking about processes by which the organization of the hierarchy is adjusted or optimized so that control improves, so that the perceptions that are constructed from intensity signals and at successively higher levels up through the hierarchy are easier to control well. These change processes are of two kinds: learning processes within the individual, and the evolutionary consequences of individuals that control less well tending to be less successful bringing offspring to reproductive maturity. Because of evolution we humans perceive a far richer world of visual configurations than frogs do. Because of learning and now controlling heliocentric concepts published by Copernicus in 1543, when we see the sun going down the “eyes in our head see the world spinning round”. Because of learning, the radiologist at Massachusetts General Hospital looks at my x-ray image and sees clearly what’s going on where I see a murky cloud, and a properly trained and experienced physicist identifies specific particles and their interactions by their tracks in a cloud chamber where you and I see what might be a mid-20th century composition of lines and spirals by Paul Klee.

We usually assume that humans are equally endowed by evolution. (This assumption may be wrong in detail, for example in color blindness.) An example of this is where I quoted you above saying

Because I have perceptual functions in me that can produce the perception “level of water”, I am able to see that you are controlling the level of water in the glass

Setting aside possible differences of genetic and epigenetic endowment, due to learning some humans are able to construct and control perceptions that other humans cannot. The reason we conclude that the perceptions of heliocentric astronomy correspond better to the way that reality works is because that system concept sets reference values for principles, plans, and so forth that enable us to control perceptions at those lower levels more successfully than we can control with reference values that originate from a geocentric model. So likewise for other advances in science. A major task in spreading acceptance of PCT is solving the “so what” problem: demonstrating to people how the reference settings that descend from the PCT system concept enable better control of perceptions that matter to them more immediately. (Despite disturbance to their control of career and reputation.)

In Rick Marken 2019-12-25_12:57:58 you admitted “I got so excited about the correctness of the first sentence that I failed to carefully read the second.” Maybe you got “excited” in this case about the ‘incorrectness’ of Martin’s first summary paragraph and didn’t read carefully any farther.

Here’s a later excerpt from what Martin has written:

You replied:

Relevant data would be a measure of how well people control before and after learning, where it can be shown that changes in goodness of control are due to changes in perceptual input functions.

This accounts for people tolerating relatively poor control when they haven’t learned better (= haven’t developed perceptual input functions that would enable them to control better).

But I think your term ‘aspect of reality’ is in serious trouble here. The perceptual signal p that we experience as the level of the water is a theoretical variable that is exactly the same as a single aspect of Real Reality of which it is an analog? The perceptual signal p that we experience as the system concept Copernican astronomy is a theoretical variable that is exactly the same as a single aspect of Real Reality of which it is an analog? This is asserting the thing-ness of perceptions against which you have inveighed.

No disagreement about non-perceptual access to physical reality, but for that very reason it is irrelevant to bring it up in a discussion of changes to perceptual input functions brought about by evolution and learning. What is relevant is in your first sentence here. The person hasn’t learned to control well yet. Learning makes changes in their control hierarchy. If the changes are in the perceptual input side, then the improved ability to control is evidence that the changed perception accords to reality better than the perception before learning did. On a small scale this is the same reasoning that says that Copernican heliocentric astronomy accords better to reality than Ptolomaic heliocentric astronomy did. If you don’t like the reasoning as applied to an individual learning e.g. to perceive a task correctly, then please explain why your objections do not apply to learning due to advances in science.

1 Like

Bruce, I think we are in fairly good, but not perfect, agreement. I won’t go into any detail about minor disagreements, but there is one big one that derives from my most recent thinking, as expressed in this thread and perhaps a little further developed since my last message. But first I should get out of the way one comment on something else that Rick wrote. Then I will add what might be new here.

I will call this long message my “Message 3”, though it is nothing ike the one I suggested at the end of message 2 that I might send at some future time.

—begin comment—

[RM] . .I remembered that Powers conceived of Real Reality as the world we experience – the world of birds, bees, flowers and trees, etc. …

[MT I think you are mixing up two different memories. The first is Powers’s habit of pointing out that what we perceive is all we can know for sure, with which I agree. The other is his claim that we can know nothing of what is in Real Reality. The environment we perceive contains “birds, bees, flowers and trees, etc.”, but there is neither guarantee nor evidence that Real Reality contains those things or “atoms and molecules” as opposed to there being an omnipotent power that decides exactly what should happen when, or Real Reality containing my huge bureaucracy of message-passing gnomes.

—end comment—

[MT] Bill’s second claim is what my two major messages are intended to dispute. My alternative is that though we cannot know what are the contents of Real Reality, we can create, both through individual reorganization and “scientifically”, perceptual and imagined constructs that mutually influence each other in ways that ever more accurately model the ways influences interact in Real Reality.

[MT] Staying with the “gnomes” fantasy, to represent a tree, a bee, or a molecule would be the job description of a particular gnome, in that it is what the gnome does with incoming messages to turn them into outgoing messages to other gnomes. The messages passed by the bureaucracy are all that we can approximate by individual reorganization and collective science, but we can approximate them very well, given time in which the gnomic organization chart doesn’t change. We would see that lack of change as the tree looking and behaving the same in its responses to externam influences.

[MT] The “tree-gnome’s” Inbox is filled with what we perceive as the forces of the wind on the tree, the forces of a logger’s saw, energy supplied by the sun, the nibbling of caterpillars on the leaves. All of these do change how the tree behaves, changing the tree-gnome’s job description as to how to deal with incoming messages.

[MT]There are myriads of such “gnomic messages” that we perceive as effects on the tree, but they are not reported individually to the tree-gnome. They come in compact packages, perhaps one package from the “north lower tree-branch gnome” another from the “sap-pressure gnome”, another from the “foliage integrity” gnome, and so forth. There are lots of packages, too, each of which contains packages the lower-level sending gnome received from other yet lower-level gnomes.

The “tree-gnome” also sends packages of messages to less senior gnomes whose jobs are to cause effects within their particular domains. For example, perhaps the tree-gnome has received a package from the air-temperature gnome and another from the leaf health monitor gnome. On reading those messages, the tree-gnome sends a package of messages to different “effector” gnomes. For example, the sap-delivery gnome for the leaves gets a message to stop work (we perceive that the season is autumn and the leaves turn red or brown and soon fall off the tree), and so on with messages to other effector gnomes.

Why do I conceive this way the structure of the messages passed in the gnomic message network? It is because we have, by evolution, reorganization, and science, found that just such a message-passing structure has kept us and our ancestors functioning reasonably well. In this community, we call that structure a “Perceptual Control Hierarchy”. This hierarchy is full of “White Boxes” we call “Elementary Control Units”, the individual gnomes. Each ECU “White Box” is built from two other WBs we call perceptual and reference input functions plus two others we call comparator and output function. Each mini-WB corresponds to a gnome. Something in Real Reality performs the functions I ascribe to a “gnome”. Whatever it may be, it is a mini-BB and we cannot know how it works internally unless we can design a set of micro-WBs that together perform the functions of the mini-WB.

We can know what the White Boxes do, and how they do it, because we built them in our theory. When we don’t have a good understanding of the micro internal WBs that we need for a particular WB, we just say that the mini-WB does thus and so, and we will figure out later how it does what it does. “What it does” is a write-up of one gnome’s job description, consisting of the message packages that may arrive in its inbox and how from them it should create a message to send on. If we have a correct understanding, even an incomplete one, of the gnome’s job description, our experiments should give the same or nearly the same results as do our enquiries of Real reality.

None of the above means I think there is a shred of likelihood that Real Reality is filled with gnomes. What it means that if there were such gnomes in Real Reality, we could not possibly tell that they were (or were not) there. But in contrast to the claim that we can know nothing about Real Reality, the PCT construct of “reorganization” suggests we can in principle know to any desired level of precision what influences pass from our effectors to our sensors and how those influences interact in “bundles” to define the “properties” of the entities we perceive to exist in our experienced world.

It is worth observing, however, that the construction of sets of smaller White Boxes within already described WBs describes the progress of Science all the may from imagining the spirits that animate every rock, river, or cloud to imagining the Higgs Field and Boson and computing to exquisite precision its consequences. But at no point in this progress is it possible, or will it ever be possible, to say that the nano-WBs are not implemented by spirits or Gods and Goddesses. Or even by a single omniscient omnipotent God who could turn the whole thing off on the spur of the moment.

Martin

Hi Bruce

Most of this is fine. I’ll just quickly reply to some points with which I disagree.

RM: …there is certainly nothing in PCT that predicts that control will “not be very good” in proportion to the degree to which perceptual reality “departs” from real reality. If this is predicted by your model of control, I’d like to see the data that shows that this is the case.

BN: Relevant data would be a measure of how well people control before and after learning, where it can be shown that changes in goodness of control are due to changes in perceptual input functions.

RM: This only shows that people can learn to control variables that you can also see. It does not mean that they have learned to control a variable that is more consistent with external reality. When a student learns to control the distance from cursor to target in a tracking task it means that the person has developed a control system that can perceive that perceptual variable, not that they have learned to control a variable that corresponds better to a variable in physical reality.

BN: The perceptual signal p that we experience as the level of the water is a theoretical variable that is exactly the same as a single aspect of Real Reality of which it is an analog?

RM: Yes, if I see, via the TCV, that you are controlling a variable that could be called “the level of water in a tank” then p is exactly the same as that variable.

BN: The perceptual signal p that we experience as the system concept Copernican astronomy is a theoretical variable that is exactly the same as a single aspect of Real Reality of which it is an analog?

RM: Yes, of course! That’s how control theory works.

BN: This is asserting the thing-ness of perceptions against which you have inveighed.

RM: Not at all. It is just saying that what an observer sees as a variable being controlled – the variable q.i in PCT, such as the “level of the water” or “Copernican astronomy” (that’s actually more like a reference state of a variable called “astronomy”) – then p is that same variable in the person who is found to be controlling that variable.

BN: Learning makes changes in their control hierarchy. If the changes are in the perceptual input side, then the improved ability to control is evidence that the changed perception accords to reality better than the perception before learning did. On a small scale this is the same reasoning that says that Copernican heliocentric astronomy accords better to reality than Ptolomaic heliocentric astronomy did. If you don’t like the reasoning as applied to an individual learning e.g. to perceive a task correctly, then please explain why your objections do not apply to learning due to advances in science.

RM: I don’t think finding a model that “accords better with reality” (by which I presume you mean, fits the data better than another model) is equivalent to finding a way of perceiving that allows you to control better. One evidence of this is a demo Bill wrote where he had something like 500 control systems controlling different perceptions of an environment made out of 500 scalar variables. Each system’s perception was a different linear combination of the same 500 scalar variables. Bill used the E.coli reorganization algorithm to adjust the weights of of these 500 perceptions so that each control system controlled a perception that was different enough from the perceptions the other systems were controlling so that there was no conflict between the systems; after reorganization all 500 systems were successfully controlling a different aspect of the same environment.

RM: None of the perceptions that were being successfully controlled “accorded with reality” better (or worse) than any of the other perceptions. They were just perceptions that “worked”. I think it’s very likely that the perceptual variables that organisms have evolved (and/or learned) to perceive and control are simply those that “work”, not necessarily those that “accord best with reality”.

RM: This is quite different than what goes on in science. There, the scientist is controlling for inventing a model that accounts for the data. It is “assumed” that the model that does best is the one that “accords best with reality” but, as you know, the models that fit data best can change so that the “reality” with which the new model accords can change rather dramatically; witness how different the reality with which Einstein’s model accords is from that with which Newton’s model accords.

RM: But the bottom line is that this notion of control being better the closer perception accords with reality is something that is irrelevant to doing PCT science. The goal of PCT science isn’t to see how well controlled perceptions accord with reality; it’s to figure out what kinds of perceptual variables organisms control, why they control them and how. So this Real versus Perceived Realities thread really doesn’t belong in the PCT Science category. But I’m not the category police here;-).

Best

Rick

It is clearly possible to control in consciousness any arbitrary variable without worrying about Real Reality. We control in imagination all the time. We can control through the environment using the evolved and reorganized control hierarchy, at least for sufficiently low-level variables. For these, the environmental feedback path through Real Reality influences something in RR that behaves very like some variable that has been embodied in a Perceptual Input Function.

Issues of control quality if Real and Perceptual Realities differ only come to light when the controlled perception creates a variable in PR that behaves differently from what in RR is influenced by the output of the control unit (acting through the so-far reorganized hierarchy) to become actions on the environment.

As I have tried to clarify in my longer messages above (the ones I have called Messages 1, 2, and 3. What we cannot know about Real Reality is its content — how it does what it does.

In contrast, the processes of reorganization building on evolution (and of Science building on both) allow us to learn an arbitrarily close approximation to the intertwined network of influences in Real Reality, We embody them in the network of influences among the White Boxes we have constructed (a.k.a the learning and maturing control hierarchy). For example, if some perceptual variable is a function of sensory variables X and Y, and the control output influences a Real Reality variable that is a similar function of X, Y, and Z, control will be improved if reorganization incorporates Z appropriately into the Perceptual Input Function.

Bottom line: Reorganization continually improves the influence map between the White Boxes in our Perceptual Reality and Black Boxes in Real Reality. Improved influence maps result in improved control.

Martin

Your term ‘aspects of reality’ is misleading and confusing. As an experimentalist, of course, it makes perfect sense that Qi is what you perceive, and the TCV verifies that it is “the same” as what the subject is controlling—that is, they co-vary.

Yes, Qi is p in the experimentalist, it is not a Really Real quantity in the environment (though it is what the experimentalist perceives as being in the environment). The TCV demonstrates that p in the experimentalist is an analog of p in the subject (they co-vary). The subject may be estimating distance to the top while the experimenter is reading a ruler, or one may be reading the centimeter side of the ruler while the other is reading the inches and fractions side of the ruler–like the coin game example of Z and N.

However, success controlling the TCV leads the experimenter and observers to conclude that we know what is Really Really going on inside the subject out there in our environment. We may measure it as Z or N or 3.729 cm, and the subject may gauge it as about enough to water the petunias, but however we measure or gauge it, we have identified a controlled variable and we have demonstrated that the subject is controlling it.

The disparity of perceptions and labels, by the way, suggests something Really Real is behind the discrepant perceptions and labels, by something vaguely analogous to parallax. I do not claim that is proof, nor am I seeking it.

If you’re a scientist controlling the fit of data to the model then finding a model that does that enables you to control better. But a successful model does more than that. It enables engineers to control a great many applications, which enable other people to gain and maintain better control of a wider and wider variety of perceptions. For many scientists—every scientist that I have known and know—these social benefits are also among their own controlled perceptions. So you may not think so, but there’s a variety of evidence supporting the claim.

I remember Bill’s reorganization demo in LA in 2003, an array of 3 to 500 systems each controlling the same array of 3 to 500 variables. I’ll cite that at the end of this post.

Of course not. They had no sensors, only programmatic inputs. This has no bearing on the question.

I understand, yes, that your focus is on the application of PCT methodology in experimental work, and the presentation of the results of that work. I applaud and support that. We haven’t yet figured out how to apply it to e.g. the data of language change in a community; that will come.

The relation of perception to reality is important in communicating to people who almost universally assume that their perceptions are reality. Control of perception gets construed as control of mere appearances, as in public relations spin. Or they say, OK, we’re controlling what matters to us but that means we’re controlling that which matters, not just our perception of it. If I put on a VR headpiece the sandwiches are not going to satisfy my survival needs.

You introduced a rhetorical device that could actually help bridge this communication gap, so long as no one looks too closely. I mean your talk of f(v1, v2, …, vn). Now we know those ‘aspects of reality’, the ‘physical variables’ (v1, v2, …, vn), are no more than perceptual variables that are collectively controlled by physical scientists. If you reject the notion that the work of science is to make science perceptions accord as closely as possible with reality, then these ‘aspects of reality’ have no more claim to being Really Real than the naive perceptions that are said to be a function of them. The rhetorical device ‘aspects of reality’ is an empty gesture. Sorry, no cigar.

The PCT model includes more than the behavioral hierarchy which is tested and demonstrated in every PCT experiment that I’ve ever heard of. On good evidence Bill postulated the parallel somatic hierarchy from the Event level down, and proposed how emotions arise from interconnections between the two branches. Emotions are perceptions. They can be controlled in a number of ways. For example, the somatic branch is amenable to suggestion according to where the higher cortical functions direct one’s attention, deliberately evoking memory and imagination to which somatic systems respond. Talking about feelings reduces the traffic between the amygdala and cortical functions (I’ve cited a paper about this). Finding out “what kinds of perceptual variables organisms control, why they control them and how” is more difficult here, because the somatic branch is (for most people) not available for conscious control. Even in investigations into the behavioral branch, the interactions between the two branches often obscure what is really being controlled.

Very much related here are the astonishing phenomena of hypnosis. Figuring out how in heck the brain and body do those things is a rich field for PCT research. I recommend The collected works of Milton H. Erickson. Rather expensive now, I’m fortunate to have them from my family therapy training in the 1970s, and they can be found in a library.

Scientists generally are also very much interested in the interpretation of their science, not only for prior sciences on whose results they depend and in respect to fields which depend upon their own results, or for philosophy, especially philosophy of science, but also for its social and political contribution. There’s a good argument that these belong under subcategories of the Science category, such as ‘philosophy of science’, ‘science and society’, ‘science and politics’, and under a category like ‘communication of PCT’.

At the end of his 2003 paper which we’ve been citing for a description of the reorganization demo (though its stated topic is a comparison of PCT with engineering control theory), Bill described an unfortunately severe limitation of PCT simulations:

It is possible to set up control-system simulations
for single levels of perception. Tracking tasks, for
example, involve control of a relationship between
a target and a cursor that is supposed to stay on it,
or near it. However, this requires assuming that a
perceptual input function exists and produces a signal
corresponding to a measure of the relationship that
we can compute in other ways. We can propose that
there is a control system that perceives the distance
between the cursor and the target, and we can say that
the magnitude of this perceptual signal corresponds to
the measured or otherwise known distance between
these entities in laboratory space—but we can’t put
into the simulation the mechanism by which this
physical situation is turned into a perceptual signal.

The predictive power of such simulations can be
quite impressive. We can often match the performance
of a simulation to that of a real person within
one percent or better over a continuous 60-second
experimental run. But there is no way, in most cases,
to build a two-level simulation involving either a
higher or a lower system while accounting for the
way the higher-level perception is derived from the
lower. Those dependencies we still have to get from
informal observations.

As matters stand now, therefore, we have to conclude
that we do not know how perceptions of higher
than order two are computed, and our only way of
guessing what these perceptions are is through careful
observation of subjective experience—which is to say,
the only kind of experience there is.

  • W. T. Powers (2003) PCT and engineering control theory. (Presented at the annual meeting of the Control Systems Group at Marymount College in Los Angeles, July 23-27, 2003.)

As I suggested earlier, the perceptual input function of a relationship perception like ‘level of water’ could be thought of as comprising all the control loops at lower levels that provide perceptual input signals to it, right down to the intensity perceptions. In this passage, Bill is saying that the perceptual input function in a simulation is essentially a fudge factor summarizing what the net effect must be of the imputed lower levels of the hierarchy.

These are outstanding puzzles that are of great importance for PCT in respect to neuroscience:

  • The weighted combination of lower perceptual signals by a perceptual input function
  • The combination of error signals by a reference input function
  • The branching and weighting of signals that are so combined
  • The role of memory, which we are told resides at every synapse

In robotics and artificial intelligence we can do whatever works; we don’t have to be concerned with how the nervous system appears to work—unless it turns out that is indeed just what works and fudge factors run into blind alleys.

For reference, here are links to the reorganization demo cited above, and Bill’s associated paper, together with your 2017 email commentary. I quoted and linked the 2003 paper earlier. As you know, but other readers may not, it documents the reorganization demo, or rather testbed, in context of contrasting PCT to engineering control theory and early attempts to apply it to psychology.

Rick Marken (2017.10.17.0840)

RM: This reminded me of a demo Bill Powers did involving N control systems simultaneously controlling perceptions, each of which is a function of N environmental variables. I managed to find the program (which unfortunately runs only on PCs). But here it is:

Dropbox - MultiControlPrj.exe - Simplify your life

RM: And I highly recommend running this program (if you can) after reading the write up associated with it:

RM: I won’t go into an explanation of the demonstration here but I’ll just say that it’s a great test bed for testing models of reorganization of perception. Indeed, Bill says as much in the write up:

BP: Since the initial perceptual weightings are selected at random, there is no guarantee that the perceptions are independent of each other. Thus there can be considerable conflict between some pairs of control systems. Also, the weightings can add up to a small or a large effect on the perception. For both reasons, some control systems will control more accurately than others. This, indeed, is one basis on which reorganization could occur: the weightings for a given control system’s input function could be randomly shuffled until the control error is minimized. This is a promising topic for further investigation. [italics mine]

RM: I think I recall there being a reorganizing version of this demo but I can’t seem to find it. If anyone has access to it please let us know about it. But I think it would be pretty easy to write up some perceptual learning models for this test bed program here. The goal would be to reorganize the weights of the linear combination of environmental variables that produce each of the N controlled perceptual variables so that these perceptions are as orthogonal as they can be. When this is true, you will see all N perceptions coming very close to their respective reference signals. I’m pretty sure that it would be found that RL algorithms don’t work, unless they are actually versions of E. coli reorganization by a different name.

RM: This demo also illustrate a few points I made earlier about the relationship between perception and environmental variables and about the nature of perceptual learning from a PCT perspective. The demo shows that perceptual variables are functions of environmental variables; they don’t correspond to variables in the environment. There are no CEV’s in this demo, just N environmental variables. The perceptual functions of the N control systems define the aspects of these N environmental variables that are controlled. The demo also shows that perceptual learning involves variation and selective retention of the parameters of “built in” types of perceptual functions. In this case, the type of perceptual function that is “built in” to all N control systems is a linear weighting function of the type p = w.1q.1+w.2q.2+… w.N*q.N where the w.i are the weights of the environmental variables q.i.

RM: As usual, you can learn a lot about PCT by reading Powers and, more importantly, following Powers’ demonstrations of how the model works.

Of course there are no environmental variables (CEVs) in this demo. As in most of Bill’s demos of which I am aware (Arm 2 is an exception), the environment is empty of them by design. In the Arm demos, the Arm’s physical properties are properties of the environment of the various controllers.

Bill’s typical demo environment is not an environment in which you might be blocked by a wall or struck by something coming at you from you know not where. Nor, for that matter, does the external environment of even the Arms contain arbitrary constraints on the the movements of the Arms, or disturbances from outside to the multiple individual control systems. Apart from the constant physical properties of the Arm, control is only in imagination, where all is possible, and reorganization orthogonalizes the hierarchy.to minimize mutual disturbance among control units.

Real or perceived, our environments are not like that.

Thanks Bruce. This was a very helpful post, inasmuch as it pointed me to the demo I was talking about that was at my Dropbox site (!) and because of the nice quotes from Bill. The last part of Bill’s quote basically described where I have gotten to as far as understanding what we are trying to find out with PCT research:

As matters stand now, therefore, we have to conclude
that we do not know how perceptions of higher
than order two are computed, and our only way of
guessing what these perceptions are is through careful
observation of subjective experience—which is to say,
the only kind of experience there is. – Powers, 2003

I think all we really need to assume to do psychological research based on PCT is that perceptions are based on physical reality. We don’t have to know how they are related to physical reality; that is, we don’t have to know how they are computed by the perceptual functions. What we want to know about perceptions is what they are – that is, we need precise descriptions of what perceptions are controlled and a way of objectively determining what types of perceptions are they. So I take Bill’s quote as saying something like: the difference between perceived and real reality is not relevant to PCT; it’s certainly not relevant to PCT research.

Since PCT research is the topic I’m interested in I’ve decided to create another topic heading where we can discuss what a research program in PCT might look like. If anyone interested in PCT science has anything to contribute on this topic feel free to join the discussion at research-on-purpose.

Best

Rick