[Vyv Huddy 9.2.17.2120]
[Martin Taylor 2017.02.07.09.26]
[MT] Every hunting animal and every prey animal has controlled various perceptions of distances and the relations between them, such as the predator-prey distance and the distance to safe cover. Predators who didn't control those very well died hungry, while prey who didn't died quickly. Whether those distances were in their environments or not, the animals that behaved as if they were in the environment would be more likely to have descendants than those that behaved as though the distances were not. We are those descendants, so it is highly probable that we are built to perceive that distances are properties of our environment. It doesn't matter whether that is really true, because it's unknowable. What matters is that we control other perceptions, and survive, better if we treat them as being properties of a real environment that exists.
VH: So what is the purpose of research in this field? This is a rather bleak comment from my perspective!
[MT] On the other hand, consider "x" and "y". You have to be taught about them in school, and nobody used them seriously as controlled perceptions before Rene Descartes (except maybe some Greek or Chinese or Arabian or Babylonian geniuses I have not heard of). Few, if any, animals compute "distance" by estimating an "x" and a "y", squaring them, summing the result, and taking the square root, as Rick says the subjects do in the rubber band experiment. Even if a few humans do that on some occasions when they are controlling a perception of distance, it isn't common, because a lot of humans don't even know that it is possible, and of those that do, it's much easier just to see that this stream is too wide to jump (or not) than it is to invent an x-axis and an orthogonal y-axis, project the path across the stream onto those imagined axes and do the arithmetic.
VH: My understanding is in his degrees of freedom in behaviour paper in 91 Rick demonstrated that a cartesian model worked better than a polar co-ordinates model of the CV in that study? This seems to support the idea that people are controlled the xy dimensions in that task? And probably in similar ones?
VH: I enjoyed reading the rest of your post Martin but am trying to stay focused, if I can.
Of course, that argument applies only if the stream exists in the environment, because if it doesn't, you won't perceive wetness when you jump it, no matter how great you perceive the distance across it to be. The fact you perceive the stream to be in the environment doesn't mean that it is. It's quite possible with current technology to produce the impression of a stream flowing over what you perceive to be a dry surface when you walk over it. And that is an example of a point you didn't mention but that has been frequently raised in this kind of discussion: it is only when controlling a perception of X affects one's (or better, someone else's) control of some other perception that we have evidence that X exists in the environment.
MT: One can demonstrate that it [lemonade] is in the environment, because you can get anyone else who has tasted lemonade to say whether the taste of a particular substance is lemonade or something else.
VH: That's true - it gives the appearance,doesn't it, of them recognising the same "thing"? But that's not how I understand the input function works. What is happening is they are perceiving enough lemonade-ness to say it's lemonade that out there? That perception is constructed from layers of lower level inputs. Given every person has different input function ... there will be differences between people in how they construct this perception. This means each "covariance" of p to qi will be slightly different? So there is no "lemonade" but a multitude of them.
[MT] I would prefer to say that there is a range of mixtures that create a conscious taste anyone would quite happily call "lemonade". Some are more "lemonady" than others, which is why we use adjectives as modifiers when we communicate. This glass has "really good home-made lemonade" while that one has "some canned stuff I call lemonade because it's all I have at the moment". If I offer those glasses to someone else, they might say the first is "lemonade, all right, but too tart" while the second is "real honest-to-goodness lemonade". But both glasses are lemonade to both of us.
[MT] Now do the same thing with lots of different mixtures in lots of glasses, and lots of people who have tasted what they call "lemonade". I think you will find that some glasses contain stuff nobody calls "lemonade", some have stuff that everyone calls "lemonade" and some have stuff on which there is disagreement. What there isn't disagreement on is that each of the glasses contains a liquid that produces a taste. The disagreement is on the labelling of the perception.
[VH] Given every person has different input function ... there will be differences between people in how they construct this perception. This means each "covariance" of p to qi will be slightly different? So there is no "lemonade" but a multitude of them.
[MT] To be able to say that, you have to be taking a Theoretician's (or Analyst's) viewpoint. No one person can tell how anyone else experiences a specific taste. What each person can tell is that there is a multitude of different mixtures that are called "lemonade" by different proportions of people, and possibly that there are certain mixtures almost everyone calls "lemonade". The qualia of "taste" is not in the environment, but the function of environmental variables that produces that taste for almost everyone is -- or, as with "distance", we control other perceptions better if we act as though it is. For example, If I am controlling for a stranger to say something equivalent to "That's really nice lemonade", my perception of their behaviour on drinking my mixture is more likely to be brought to its reference value if the mixture is near the one the largest proportion of people agree is "lemonade" than if is distant from the "ideal lemonade" mixture.
VH: I wonder in situations where there is a lot of agreement of when thing is being perceived does that mean that the function determining covariance of the qi and p might be more similar across people? Perhaps this is one for Rick to answer as I get the feeling he'd say we don't know.
[MT] Rick may well say that, or he might take the other tack and say what he often does, that p "IS" qi, and the covariance is always perfect. But I think that's irrelevant to the question at issue, as you are mixing two concepts here. One is the Analyst's analysis of theoretical control loops, the other is the interpersonal consistency of labelling.
[MT] Let's imagine two sets of people. The people in set "S" have always bought a particular brand of canned lemonade concentrate to make lemonade, whereas the people in group "H" have always used lemons squeezed at home. When someone in group H tastes "lemonade" made by someone in group S, they are quite likely to say "I don't know what that is, but it isn't lemonade", and vice versa. For all we know, the qualia might be identical for both groups. What we do know is that their labelling is different, and if everyone controls with high gain for the label "lemonade" to be applied properly only to their own ideal mixture, we have a conflict of collective control.
[MT] "Real" world examples of such labelling conflicts are frequent. Think "whisky" and "whiskey", or "Champagne" or "Roquefort", and any court case in which the permission to use a trademark was contested. Or maybe none of that ever existed in the environment?
VH: It seems to me that the modelling approach helps with this problem.
[MT] I'm applying your quote to the entire issue, not to the restricted point you made on my comment that the issue was one of communication rather than process. I think modelling is great for certain issues, because it makes precise what is always fuzzy in textual language. That precision has a downside, however. If the model doesn't fit well, is it because the structure of the model is wrong or because the parameters are not correct? Even when we are dealing with control in a simple single-variable tracking experiment we really can't tell. The models never fit perfectly, and usually the fit improves if some prediction is incorporated (structure change). But the effect of prediction can be produced in at least three structurally different ways (direct prediction, adding a higher level of control, or Powers's Artificial Cerebellum). If modelling cannot directly answer even the simple questions "Is prediction generally or ever used in tracking" and "If prediction is used, how is it implemented", how is it going to helpful in addressing the questions we are talking about here? What kind of model would distinguish between there being an environmental correlate "qi" of the perceptual variable and there being no "qi" in the environment?
Long responses to short comments. Sorry about that.
Martin
PS. Partway through the writing of this, my mailer gave up the ability to "Paste as quotation" so the later quotes are improperly marked. I hope that doesn't cause confusion.
···
[Martin Taylor 2017.02.02.12.29]
MT: The real issue is not about words we use in our discussions, but about the underlying processes we are trying to describe, and my hunch is that we are closer on those than we are on the words used to talk about them.
VH: It seems to me that the modelling approach helps with this problem.
MT: One can demonstrate that it [lemonade] is in the environment, because you can get anyone else who has tasted lemonade to say whether the taste of a particular substance is lemonade or something else.
VH: That's true - it gives the appearance,doesn't it, of them recognising the same "thing"? But that's not how I understand the input function works. What is happening is they are perceiving enough lemonade-ness to say it's lemonade that out there? That perception is constructed from layers of lower level inputs. Given every person has different input function ... there will be differences between people in how they construct this perception. This means each "covariance" of p to qi will be slightly different? So there is no "lemonade" but a multitude of them.
VH: I wonder in situations where there is a lot of agreement of when thing is being perceived does that mean that the function determining covariance of the qi and p might be more similar across people? Perhaps this is one for Rick to answer as I get the feeling he'd say we don't know.
MT: But anyone is quite free to say that there's no "taste of lemonade" in the environment, just as they can say that if x and y are in the environment, x-y is not. I'm not clear, however, on what grounds they determine that x or y is in the environment.
VH: I don't quite follow the last sentence?
If the latter how should I square this with:
BP B:CP (p.273): “the output quantity and the *input quantity* are physical quantities and are expressed in physical units”
BP: Further BP (p. 273) says the perceptual signal “p [is a] neural event and must be expressed in signal units (impulses per second)”
VH: and p is a function (ki) of the Qi with BP “units impulses per second per physical units”. So all this would suggest there should be an “actual” position expressed in physical units in PCT models? If actual means consensual / verified / existing in fact.
VH: I’ve re-read the modelling section in the change blindness paper. The model in the paper and similar models elsewhere (e.g. Doing Research on Purpose p. 38) doesn’t include the terms ki / ko but only the overall term for gain.
VH: This is consistent with the approach Bill Powers (p. 275) describes where “the magnitude of ki is 1 which is the basic normalisation” and later (p. 278) “because of normalisation qi is numerically equal to p; hence we may substitute qi for p”.
VH: The consequence of this method though is that everything is “actual” in this type of modelling that estimates loop gain. All the variables are expressed in what BP B:CP (p. 275) terms of an “environment unit”? In terms of physical “actual” units, which are pixels in the rubber band model? So there is only actual distance?
VH: I’d appreciate your help (and others on here) in understanding this!
Best
Rick
That is, it is not k-d (the physical distance) that is controlled; it is f(k-d) that is controlled; f() is the trigonometric function that gives the optical angles of k and d from the point of view of the subject.
RM: f(k-d) is what is called the controlled variable (or controlled quantity, q.i in Bill's diagrams). It is the aspect (or function) of the physical variables in the environment that the system is controlling. PCT assumes that the perceptual signal varies in proportion to f(k-d). That is p = f(k-d). So the perceptual signal (or, just "perception") that is controlled is the same as the controlled variable (or controlled quantity). That's why I say that an observer who is perceiving f(k-d) is perceiving what the controller is controlling.
RM: This is the basis of the TCV. When an observer perceives the aspect of the environment that the controller is controlling -- in the rubber band case, when the observer perceives f(k-d) -- the observer will see that this variable is protected from disturbances; it's the controlled variable. When the observer perceives an aspect of the environment that is not what the controller is controlling -- say f(k+d) -- then the observer will see that this variable is not protected from disturbances (not very well, anyway) and the observer will try to come up with a new hypothesis about what the controller is controlling.
RM: The process isn't as mechanical as it may seem. In the rubber band demo, for example, my first hypothesis about the controlled variable was that it was simply k-d (taking f() to simply be 1* (k-d)). And this variable was indeed controlled. The only reason I was able to realize that it was a trigonometric function of k-d that was controlled was because I noticed that the model was keeping the knot well to the side of the dot.
RM: My research on object interception provides a better example of how the TCV works. In that case there were three clearly different aspects of the environment (the object's physical trajectory) that might be controlled: 1) relative change in the vertical (v) and horizontal (h) optical angles (dx/dy),
2) vertical velocity (dx) and horizontal displacement from visual center, c (h-c) and 3) vertical acceleration (d2x) and horizontal displacement (h-c). The winner, determined by modeling, was 2). So my calculations of dx and h-c are my perceptions of the controlled quantity that correspond to the perceptions (controlled variables) controlled by the controller.
Best
Rick
On 1 Feb 2017, at 22:40, Warren Mansell <<mailto:wmansell@gmail.com>wmansell@gmail.com> wrote:
Hell if you feel that strongly about it I'll just roll over and go with your eloquent explanation! 
On 1 Feb 2017, at 19:46, Richard Marken <<mailto:rsmarken@gmail.com>rsmarken@gmail.com> wrote:
[From Rick Marken (2017.02.01.1145 PST)]
On Wed, Feb 1, 2017 at 12:20 AM, Warren Mansell <<mailto:wmansell@gmail.com>wmansell@gmail.com> wrote:
RM: I think it allows one to slip off the slippery slope of science. But if you can show me how to demonstrate that the perceptual signal is controlled better than the aspect of the environment to which it corresponds, I'll turn in my PCT decoder ring on the spot.
WM: Sounds like a deal. I'll put it on my long list of research plans!
RM: This is not good enough for me. You (and Martin and other people on CSGNet) seem to be convinced (sans any evidence) that the perceptual signal -- a theoretical construct -- is controlled better than the observable aspect of the environment to which it corresponds. It seems to me that this represents a fundamental misunderstanding of the relationship between theory (PCT) and the phenomenon it explains (control) that should be addressed immediately, rather than left for future research. This is especially true for someone who is the editor of a set of papers that pay tribute to William T. Powers, the developer of PCT, and a true scientist who was very clear about the relationship between fact (control) and theory (PCT).
RM: As I've said before, the idea that the perceptual signal is controlled better than the aspect of the environment to which it corresponds is not just a harmless misunderstanding of the relationship between fact and theory. It's a red herring pulled across the path of progress (to paraphrase Powers from his 1978 Psych Review paper) -- progress in research on PCT. The idea that the perceptual signal is controlled better than the aspect of the environment to which it corresponds implies that the perceptual signal is not an accurate representation of that aspect of the environment. This points research in the direction of measuring how accurately perception reflects what is going on in the environment. This kind of research is called psychophysics -- research that is already a major part of conventional psychology -- and it doesn't advance our understanding of the controlling done by living systems very much at all.
RM: What PCT needs is research aimed at understanding what aspects of the environment (what perceptions) organisms control and how they control them. And this kind of research can only be done by people who understand the relationship between fact and theory in PCT; people who understand that the perceptual signal is a theoretical construct that corresponds to the observable (factual) aspect of the environment that is being controlled.
Best
Rick
--
Richard S. Marken
"Perfection is achieved not when you have nothing more to add, but when you
have nothing left to take away.”
--Antoine de Saint-Exupery
--
Richard S. Marken
"Perfection is achieved not when you have nothing more to add, but when you
have nothing left to take away.”
--Antoine de Saint-Exupery
--
Richard S. Marken
"Perfection is achieved not when you have nothing more to add, but when you
have nothing left to take away.”
--Antoine de Saint-Exupery
