Pct vs mct

CSGnet Archive CSG2014
1

[Philip 8/29/14]

I’m still trying to understand the exact distinction between pct and mct. So in pct, there’s a representation of input compared to a separate internal representation of what the input should be. Thus the behavior is described as input control. This stands against mct. In mct, theres a reference which specifies output. And the output of the system is ALWAYS(?) compared to a reference output which is NOT(?) an internal representation of what the output should be, but is instead a calculated inference based on an internal model of environmental parameters. If all this is true, I guess my question is this: Is the internal model associated with the mcs the defining characteristic which pct is trying to do away with? Or are there modern-engineering control systems which don’t use models and they’re basically the same thing as a pcs but they match an internal representation of an output value (reference) to the output value?

And another of my inquiries. In mct, there seems to be a concern over the measurement of an output value not being the true value due to uncertainty, noise, etc. but in pct this concern is fundamentally different. I detect a strange concern in pct regarding the effect of measuring a “stupid” variable, for lack of another word. For instance, if someone is trying to control an image on the other side of a glass, but they’re looking through warped glass. In such a case, the internal representation of the variable controlled will have an “incorrect” relation to the “true” value of the environmental variable. But this is not a measurement error as much as it’s a logistical error. So my question here is this: is there a fundamental,“Descartian” problem with our perceptual faculties that human perceptual control systems are trying to figure out how to overcome?

My final question is this: what happens in the case where there is no “environmental” disturbance except our own. For instance, on a sheet of paper, we can scribble any symbol or drawing we want and manipulate these symbols purely at will. There is no “wind” causing our pencils to stray such that we don’t draw the symbol we want. What’s going on here? Pure math is not ever compared directly to the environment. It seems that over the course of mathematical history, the entire endeavor can be described as ourselves applying disturbances to our own definitions of the symbols until we can no longer disturb them because they are robust. I don’t know, I’m perhaps rambling, but I think pencil and paper math, in particular geometric analysis is where pct has primary applications. Perhaps we can create a mathematical construct which behaves like a pcs and use it as a theoretical tool to solve complex problems. I’ve been thinking about this for a while and the bit about topological defects got me thinking.

Please don’t tear this post into tiny little pieces and answer piece each in turn, but rather give me a holistic evaluation of what I am or am not understanding.

2

[Martin Taylor 2014.08.29.14.49]

[Philip 8/29/14]

At the end of your message, you ask: "Please don't tear this post into tiny little pieces and answer piece each in turn, but rather give me a holistic evaluation of what I am or am not understanding." That's a little difficult without being inside you to learn what you are or are not understanding in the same way most students of PCT understand it. However, I'll try.

However, your three paragraphs seem each to deal with unrelated issues, and I don't see where they fit into a single misunderstanding, so I have to treat your message in parts but each paragraph holistically. Sorry about that.

I'm still trying to understand the exact distinction between pct and mct. So in pct, there's a representation of input compared to a separate internal representation of what the input should be. Thus the behavior is described as input control. This stands against mct. In mct, theres a reference which specifies output. And the output of the system is ALWAYS(?) compared to a reference output which is NOT(?) an internal representation of what the output should be, but is instead a calculated inference based on an internal model of environmental parameters. If all this is true, I guess my question is this: Is the internal model associated with the mcs the defining characteristic which pct is trying to do away with? Or are there modern-engineering control systems which don't use models and they're basically the same thing as a pcs but they match an internal representation of an output value (reference) to the output value?

Para 1: I see the core question as: "I guess my question is this: Is the internal model associated with the mcs the defining characteristic which pct is trying to do away with?" to which my answer is "No".

Internal models are not inconsistent with PCT. They are used in planning in imagination, for example. There's an example in my comment on your third paragraph. The defining difference with MCT is in the point you make earlier in the paragraph, that PCT is (apart from control in imagination) totally unconcerned with the output. The output just varies until the input is what it "should" be to make the controlled perception match its reference value. The key difference is no more than that. PCT deals with control of input, MCT with the problem of how to get the output right.

And another of my inquiries. In mct, there seems to be a concern over the measurement of an output value not being the true value due to uncertainty, noise, etc. but in pct this concern is fundamentally different. I detect a strange concern in pct regarding the effect of measuring a "stupid" variable, for lack of another word. For instance, if someone is trying to control an image on the other side of a glass, but they're looking through warped glass. In such a case, the internal representation of the variable controlled will have an "incorrect" relation to the "true" value of the environmental variable. But this is not a measurement error as much as it's a logistical error. So my question here is this: is there a fundamental,"Descartian" problem with our perceptual faculties that human perceptual control systems are trying to figure out how to overcome?

Para 2: The "concern" you detect in PCT regarding the effect of measuring a "stupid" variable is not a concern of PCT, but of the way various among us (recently Rick and me) understand some naive philosophical consequences of PCT. I don't think Rick and I differ very much in what actually happens. If there's a difference (and I'm not sure there is), it's along the lines of disagreements about how many angels can dance on the head of a pin. In other words, for you to worry about it is an unnecessary difficulty put in the way of your understanding of PCT.

What PCT does say is that if you can control a perception successfully, then your output to the environment is influencing the environment in such a way as to compensate for all the distortions that may or may not be "out there".

Bill Powers made a wonderful demonstration that we don't control, or sometimes even perceive our output, using a kind of distortion. the "Square-Circle" demo in Chapter 9 of Living Control Systems III. Try it. You are quite sure you traced a pretty good circle, but when you see your trace, it's actually a pretty good square.

My final question is this: what happens in the case where there is no "environmental" disturbance except our own. For instance, on a sheet of paper, we can scribble any symbol or drawing we want and manipulate these symbols purely at will. There is no "wind" causing our pencils to stray such that we don't draw the symbol we want. What's going on here? Pure math is not ever compared directly to the environment. It seems that over the course of mathematical history, the entire endeavor can be described as ourselves applying disturbances to our own definitions of the symbols until we can no longer disturb them because they are robust. I don't know, I'm perhaps rambling, but I think pencil and paper math, in particular geometric analysis is where pct has primary applications. Perhaps we can create a mathematical construct which behaves like a pcs and use it as a theoretical tool to solve complex problems. I've been thinking about this for a while and the bit about topological defects got me thinking.

Para 3: I'm not going to comment on the history of math. Just on the issue of action in the absence of disturbance.

Remember that the error value in a control loop is a difference between the reference value and the perceptual value, and that the output depends on the error value. If the perceptual value hasn't changed, and there's no disturbance, why might the output have changed from zero?

Going up a level, and assuming you answered the above question, we ask the same question again.

And again...

Some levels up this recursion, we come to a level where the person we will call "the writer" has a reference to perceive on the page something such as "an explanation of why output changes when the disturbance does not".

The writer perceives no such explanation on the blank sheet of paper, which means that there's an error in the relevant control system. To correct the error, she must at least perceive some first word on the page. She doesn't perceive that word as an acceptable explanation so it must be followed by a second word, and so forth until the she perceives the result that she now sees on the page as being a satisfactory explanation.

Each of the control systems that "wants to perceive" a particular word on the page is in error until that particular word is on the page at the desired place. To make that happen involves controls of perceptions such as pencil in hand, movements of the pencil over the paper, and so forth. And when all the words are on the paper, the writer may perceive that the explanation is not very good (error in the highest level perception mentioned above) and correct the error at that level by altering the sequence of words by means that you can guess.

All the above sounds as though there is modelling going on, and in a sense there is, but mostly it isn't happening during this exercise. It happens in prior reorganization that has developed output connections that make more or less correct muscle commands when the writer want to make an "a" or a "q". On-line, the writer perceives whether the letters are coming out properly and adjusts, erases, modifies on the fly, and otherwise produces the desired input. Similarly at the high level of "an explanation", if the writer is practiced in teaching, word patterns will have been reorganized (e.g. lay a foundation first). If not, the explanation will probably have to be thought out, planned, modelled as MCT might suggest.

Please don't tear this post into tiny little pieces and answer piece each in turn, but rather give me a holistic evaluation of what I am or am not understanding.

I hope this isn't too many little pieces. And I don't know what you aren't understanding. I just hope that the foregoing helps a bit.

Martin

3

  1. Do MCT controllers always use an internal model to calculate&control their output? What I’m trying to understand is this: what logically prevents the reference value for the output of a MCS from being internal to the system, as it is in PCT. Why can’t an MCS simply control it’s (inputted) output for no regard to anything else, just as a PCS controls its input without regard for anything else. If I’m being retarded, please try to help me understand why. My inkling is that if an MCS controlled its output like a PCS controls its input, then it just doesn’t work right. I’m wondering why.

  2. Would you kindly provide me with an additional example of how modeling fits into the PCT framework. I still don’t get it :frowning:

  3. I want to try to delve a little deeper into the math example. Everybody knows flat-space Euclidian geometry. Now, suppose we declare great circles on the sphere to be the lines in our geometry. Then, one of the first axioms of Euclidian geometry fails, because there is no longer a unique line containing any two distinct points (this is one of the axioms, and it’s broken because if you consider the special case where you take two points on a sphere’s diameter, then there is more than one unique line - great circle - containing them). To get rid of this, we change the definition of a point to be a pair of enpoints on a diameter. The axiom can thus remain. In this manner, we SEEM to have countered the effect of our own disturbance to our own “axiom” or “belief” or “initial assumption”. In math, we continue along like this until we can completely describing some geometry in exactitude with as few initial assumptions as possible. Let’s not mention anything about Godel yet. So starting from this simple example, what do we consider as the “disturbance”? Does the Euclidian axiom disturb our redefinition of the line, or does the redefinition of the line disturb the Euclidian axiom? Either way, we’re talking about one of our actions disturbing another and there is no distinction between system and environment.

This is a rant:

If we can advance this discussion as efficiently as possible, if we can get to the heart of how math and beliefs are connected, then we can easily develop PCT into a powerful academic tool, rather than just a tool for therapy or computer control system design. Right now, PCT is way too much a node of tempestuous internal debate. Between Rick, Martin, and Boris, I see more debate about the CEV than I see between christianity and terrorism. And I think you guys are soon going to find yourselves at Bill’s age when he was writing LCS III, trying to minimize the amount of debate and maximize the output of fact. In my opinion, if we step into the world of geometry and discuss the very fundamental concepts of proof and argument and axiom as analogies of input and output and reference, then we’ll be handling a task that no thinker other than a PCTist has the facility to describe. PCT, as Bill left it in LCSIII, seems to me like a fully excavated gold mine. You guys basically got it all! phenomenal work; but now we’re hunting for scraps and gold dust and imaginary things like the breaking of the laws of thermodynamics in cursor tracking demos. PCT has explained animal behavior. Cool, we see the crowd behavior, we see the optical illusions, we see the squaring of the circle, we see the reorganization. Now, I think there’s two major goals left in PCT. One is to develop the mathematical concept of reorganization into a monstrously all-encompassing mathematical tool. The other is to understand the concept of intelligent design. Right now, thanks to PCT, we have a picture of what it feels like to be a penguin rolling an egg around in the snow, or what an angry commuter looks like as a green circle. Now we need to understand what it’s like to be a homo sapien doing math. And not just the numbers game. Not just with those big piles of numbers that PCT uses when it generates 60 samplings a second - no, that stuff only works when you have a computer shooting out the computations or the triangulated graphics. No, all that will come, but only later. Right now, I’m talking about the “stick to the dirt” / “quill to the scroll” mathematics - ancient style.

End of rant.

4

Content-Type: application/pdf; name="WaterJar.pdf"
Content-Disposition: attachment; filename="WaterJar.pdf"
X-Attachment-Id: f_hzlnh5vk0

WaterJar.pdf (1.28 MB)

5

Rick,

Did you know that one of the most brilliant contributions of Faraday, which helped him figure out how to conceptualize the experiements was his meticulous hand drawings of what the magnetic field literally looks like as a geometric curve? What Maxwell always had to do before experimenting was to imagine shapes he drew off the impressions he gained over periods of observation and experimentation with the forces. Nikola Tesla’s hero also was, not surprisingly Faraday. Tesla also made use of a vivid, “cartoon”, imagination to guide his experimentation and invention. He invented most of the wireless energy transmission technology we have today before 1900. But his absolute entire operation was destroyed by Edison and the US government. He was the only person in all of western history, even the history of mankind, to singlehandedly developed technology sophisticated enough to transmit global signals.

···

On Tue, Sep 2, 2014 at 1:21 PM, Richard Marken rsmarken@gmail.com wrote:

[From Rick Marken (2014.09.02.1320)]

On Fri, Aug 29, 2014 at 4:40 PM, PHILIP JERAIR YERANOSIAN pyeranos@ucla.edu wrote:

Hi Philip

I’ll try to answer these as best as I can.

PY: 1. Do MCT controllers always use an internal model to calculate&control their output?

RM: I believe that is the essence of the MCT control system; they calculate outputs based on model-based predictions of the future states of input.

What I’m trying to understand is this: what logically prevents the reference value for the output of a MCS from being internal to the system, as it is in PCT.

RM: I don’t think anything prevents the model from being any way the modeler wants it to be. I use modeling to account for observations. I try to account for the data with as simple a model as possible. I have used what could be considered predictive calculations in some models, but they were very simple predictions abnd I’m pretty sure they could be modeled as control of a higher order perception. So I’m not convinced that predictoin is needed in control models of behavior, except to model situations where people are actually making predictions (which is a control of imagination process in PCT).

PY: Why can’t an MCS simply control it’s (inputted) output for no regard to anything else, just as a PCS controls its input without regard for anything else. If I’m being retarded, please try to help me understand why. My inkling is that if an MCS controlled its output like a PCS controls its input, then it just doesn’t work right. I’m wondering why.

RM: I’m not sure I understand this. But, again, models, for me, are just tools for accounting for data. If what is called an MCS model could account for the data better than a PCT model then that would be evidence that it’s a better model than PCT. My evaluation of a model is based on how well it accounts for the observed phenomena; it’s not based on whether it seems plausible to me or not.

PY: 2. Would you kindly provide me with an additional example of how modeling fits into the PCT framework. I still don’t get it :frowning:

RM: I’m not sure I understand this either. But let me try. Modeling fits into PCT in terms of using the PCT model to account for data. So modeling is something people (like me, who are modelers) do; so the behavior called “modeling” has to be part of the PCT model, and it is. Modeling (in the PCT model) involves controlling, in imagination and then in perception, for a set of equations or a computer program that is presumed to act in relevant ways like the system under study.

PY: 3. I want to try to delve a little deeper into the math example. Everybody knows flat-space Euclidian geometry. Now, suppose we declare great circles on the sphere to be the lines in our geometry. Then, one of the first axioms of Euclidian geometry fails, because there is no longer a unique line containing any two distinct points (this is one of the axioms, and it’s broken because if you consider the special case where you take two points on a sphere’s diameter, then there is more than one unique line - great circle - containing them). To get rid of this, we change the definition of a point to be a pair of enpoints on a diameter. The axiom can thus remain. In this manner, we SEEM to have countered the effect of our own disturbance to our own “axiom” or “belief” or “initial assumption”. In math, we continue along like this until we can completely describing some geometry in exactitude with as few initial assumptions as possible. Let’s not mention anything about Godel yet. So starting from this simple example, what do we consider as the “disturbance”? Does the Euclidian axiom disturb our redefinition of the line, or does the redefinition of the line disturb the Euclidian axiom? Either way, we’re talking about one of our actions disturbing another and there is no distinction between system and environment.

RM: You are describing some pretty complex mathematical/geometric thinking; in PCT this would be control of imagined (self produced) perceptions of configurations and relationships and sequences in terms of higher order goals for meeting rules (of, for example, logic) and principles (of, for example, parsimony). This is a very challenging type of thinking to analyze in PCT terms but I think it would be a very good exercise to develop some simple experimental tests to determine what kinds of perceptions are being controlled when someone proves a geometrical theorem, for example. Such a demonstration would show very clearly what kinds of changes to the problem are and what are not disturbances to the variables controlled when people do this.

PY: This is a rant:

If we can advance this discussion as efficiently as possible, if we can get to the heart of how math and beliefs are connected, then we can easily develop PCT into a powerful academic tool, rather than just a tool for therapy or computer control system design.

RM: I believe that the only way to develop PCT into what it should be – the default basis for understanding the nature of living systems in general and humans in particular–is to keep testing it empirically.

PY: Right now, PCT is way too much a node of tempestuous internal debate.

RM: I think debate about PCT can be a healthy thing. But I do agree that the debates on CSGNet often seem to get us nowhere. My own opinion about why this is the case is that these debates are too often about what the PCT model says and how it differs from what other models say. I think these debates would be far more productive if they were about ways of testing the predictions of PCT and alternative models of behavior. You can only get so far at understanding a model by describing it verbally or mathematically or even implementing it as a computer program or physical device. For me, the true test of one’s understanding of a model is being able to demonstrate that it behaves just like the real thing. So I would like the debates on CSGNet to focus more on how to test the model; more lab work and less lecture;-)

RM: But, of course, it would also be nice to have discussions about how the model applies to real world behavior. This kind of discussion would be less contentious, I think, if it were about how the PCT model applies to some example of behavior rather than what model is better. If we want to discuss (debate) which model is better, PCT or something else, then that should be done in the context of developing experimental tests that would discriminate between the models. But when it comes to using a model to understand “real life” behavior, I think we should just assume that PCT is right (since there is no evidence I know of that it’s not).

PY: Between Rick, Martin, and Boris, I see more debate about the CEV than I see between christianity and terrorism.

RM: In the back of my mind is always “how to test the model”. The debate about the CEV was, for me, about making it clear that we have to test to determine what aspect of the environment corresponds to the perception a person is controlling. I thought (mistakenly, apparently) that Martin’s concept of the CEV suggested that such testing was not that important since the environmental correlates of controlled perceptions were “out there” for all to see. The debate ended with this misunderstanding being cleared up, which seems better than what happens with most ideological (ie. religious) debates.

PY: And I think you guys are soon going to find yourselves at Bill’s age when he was writing LCS III, trying to minimize the amount of debate and maximize the output of fact.

RM: I think Bill was there well before LCS III. I hope I’ve been there since I got involved in PCT. For me it’s all about facts, which means phenomena: observations of purposeful behavior.

PY: In my opinion, if we step into the world of geometry and discuss the very fundamental concepts of proof and argument and axiom as analogies of input and output and reference, then we’ll be handling a task that no thinker other than a PCTist has the facility to describe.

RM: These are not the kind of facts that I am interested in. Geometry and mathematics are tools for understanding what I consider to be facts. We certainly have to understand math in order to understand the model that purports to ex[plain the facts. But what’s more interesting (and important) to me are the facts themselves; in our case, the facts of control (purposive behavior).

PQ: PCT, as Bill left it in LCSIII, seems to me like a fully excavated gold mine. You guys basically got it all! phenomenal work; but now we’re hunting for scraps and gold dust and imaginary things like the breaking of the laws of thermodynamics in cursor tracking demos. PCT has explained animal behavior. Cool, we see the crowd behavior, we see the optical illusions, we see the squaring of the circle, we see the reorganization. Now, I think there’s two major goals left in PCT. One is to develop the mathematical concept of reorganization into a monstrously all-encompassing mathematical tool. The other is to understand the concept of intelligent design.

RM: I think it’s good to want to continue to develop the PCT model but I am not interested in developing the model for its own sake. I would only see a need to expand or change the model based on the results of experimental test. Since the model in its present form has survived all experimental tests so far I am comfortable using the model as a heuristic for helping me understand everyday human behavior. PCT in its present form is a sufficiently useful tool for me. But that doesn’t mean it’s perfect. We have to keep testing it; and there is plenty to test. I believe that doing this science is what will ensure that PCT becomes fundamental to psychology (and the life sciences in general).

PY: Right now, thanks to PCT, we have a picture of what it feels like to be a penguin rolling an egg around in the snow, or what an angry commuter looks like as a green circle. Now we need to understand what it’s like to be a homo sapien doing math.

RM: I agree completely! How about thinking up a way to study what a skilled mathematician is doing (controlling for) when solving a math problem. I found an article describing an approach to studying how people solve water jar problems, which are kind of like math problems. The research wasn’t done from a PCT perspective but their experiment might suggest ways to test to see what people are controlling for when solving math problems. In the study of water jar problems what they find is that people have particular difficulty hen problem solution involves putting the problem into states that look very different than the solution state. So this implies that at least one thing people are controlling for is for the problem state to keep looking more and more like the solution state. Maybe you can figure out a way to do some experiments with math problems that get at what people who can solve these problems are controlling for. I think that would be an enormous contribution to the development of PCT; it would certainly get the attention of cognitive psychologists.

PY: And not just the numbers game. Not just with those big piles of numbers that PCT uses when it generates 60 samplings a second - no, that stuff only works when you have a computer shooting out the computations or the triangulated graphics. No, all that will come, but only later. Right now, I’m talking about the “stick to the dirt” / “quill to the scroll” mathematics - ancient style.

End of rant.

RM: I think it’s great that you want to develop PCT mathematically. But I have neither the inclination not the skill to do that. What I can do reasonably well is develop little tests of the model. My current scientific hero (other than Bill Powers) Is Michael Faraday. He became my hero just recently because I found out that he was as math challenged I am: he knew little more than basic algebra; not even trig. But he became one of the most respected and distinguished scientists of his time by making some of the most important discoveries about how electricity behaves by doing experiments! Of course, he had moderately quantitative models in his head of what was going on but Faraday’s brilliant discoveries were non-mathematical until Maxwell came along. Maxwell was terrific but it’s Faraday who was the captain of the ship of discovery. I’m more of a Faraday kind of guy; like Faraday, with me it’s phenomena phirst!

Best regards

Rick

Richard S. Marken, Ph.D.
Author of Doing Research on Purpose.
Now available from Amazon or Barnes & Noble

–