# The quality of control (is quite strained)

[Rick Marken 2020-10-29_15:02:24]

I have a question for those of you who believe that quality of control depends on how closely the perception being controlled corresponds to its environmental correlate: Could you please describe an experiment that shows that this is the case?

Thank you

Best

Portia Marken

It is not that the perception signal does not correspond to its environmental correlate, this is always true by definition of the environmental correlate. The problem is that the variable qi that is entering the input function is not always the environmental correlate of the perception signal.

In your helicopter following model, the variable alpha is an angle, the input function is a rate sensor, and the perceptual signal correlates with an angular velocity, not with the angle. The angle (qi) is not controlled, and does not correlate with the perception. I would call the angular velocity in the environment a controlled quantity, but not an input quantity.

For the illustration of quality of control depending on the sensor properties, I guess there could be many examples. Here is a common color blindness test. I suppose we can call the test an experiment. The subject is looking at this image, and is asked to read out the numbers:

The experimenter has the readout for a trichromat (normal-vision), dichromat and monochromat, and depending on what the subject says, he can place the subject somewhere on the “quality of control” scale. It might be a surprise for the subject to realize he is colorblind, he had no problems with recognizing colors, from his point of view.

Snellen charts are a similar example.

Depending on the size of the smallest letters you can read from distance, you get a number placing you somewhere on the scale from good to bad vision. If the third letter in the bottom row is C, and the subject sees a blob, then the environmental correlate (controlled quantity) of that blob is also a blob, not the letter C; while the input quantity would be the letter C.

RM: I had asked how one could demonstrate that the quality of control depends on how closely the perception being controlled corresponds to its environmental correlate. My goal was to get people to realize on their own that such a demonstration is impossible because no one has direct access to the presumed environmental correlate of a perception except via one’s own perceptual systems; it’s all perception. So the claim that the quality of control depends on how closely the perception being controlled corresponds to its environmental correlate is really a claim that the quality of control depends on how closely the perception controlled by a subject corresponds to a perception that an observer wants to see the subject control.

RM: This is illustrated extremely well by Adam’s examples of the Ishihara color blindness test and the Snellen visual acuity test. Both are tests of a person’s (the subject’s) ability to control a perception that the person giving the test (the observer) wants to see controlled. In the Ishihara test the observer wants to see if the subject can reliably control for naming the number traced out by the colored dots; in the Snellen test the observer wants to see if the subject can reliably control for naming the letters that appear at different visual angles. In both cases quality of control depends on the subject’s ability to perceive the numbers or letters that the observer perceives (either via direct experience if the observer has normal visual capabilities or via identification by another observer who has these visual capabilities – naturally or artificially).

RM: These tests are, therefore, not demonstrations of how the quality of control depends on how closely the perceptions being controlled correspond to their environmental correlates. They are demonstrations of how the quality of control depends on how closely the perception that the subject controls corresponds to the perception that the observer would like to see controlled.

RM: There is nothing wrong with measuring control in this way; it is important to know well a person can control for seeing the numbers in the Ishihara color blindness test, for example, if that person is applying for a job where it is important for them to be able to discriminate red from green. As Powers (1978) said about the Man-Machine approach to the application of control theory to human behavior: “If one’s primary purpose is to keep pilots from flying airplanes into the ground…the man-machine blunder amounts to nothing worse than a few mislabelings [of the control diagram] having no practical consequences”.

RM: So, if one’s interest in control theory is to use it to see how well people control perceptions that they should be controlling, then there is nothing wrong with seeing the measurement of quality of control as a matter of seeing how well a person controls what seems like an environmental variable, such as the number on an Ishihara plate.

RM: But Bill goes on to say “If one’s interest is in the properties of persons, however, the man-machine blunder pulls a red herring across the path of progress”. That is because in order to understand the properties of persons your interest is in determining the types of perceptual variables they control. The idea that people control perceptions of environmental variables is a red herring the diverts one away from this goal.

RM: The discussion of the how quality of control depends on the correspondence between a controlled perception and its environmental correlate has made me realize that my interest in PCT is quite different that that of most others in this discussion group. I am interested in understanding the behavior of organisms by doing research aimed at determining the types of perceptual variables around which their behavior is organized.

RM: I am not interested in what we should call control or what PCT “really” says or how to explain some interesting mathematical side effect of controlling. I am interested in doing the kind of research that Powers wanted to get done, the kind of research I describe in my forthcoming book– research aimed at determining the types of perceptual variables around which behavior is organized.

RM: If anyone would like to help me out with ideas about how to go about doing do this kind of research, let’s discuss it in the existing Discourse Topic “Powers’ Model of a PCT-Based Research Program”. I really don’t want to keep arguing about the same stuff we’ve been arguing about for over 40 years. I would like to have a constructive discussion of how to advance the science of PCT along the lines Bill suggested.

Best

Rick

To make that easier to find, here’s a link to that topic.

You don’t need an external observer to test your vision. Put the newspapers at the maximum distance you can still read the letters, and you have a crude measure. You can compare to previous, or someone else’s measurement if you want. It does involve multiple loops, I guess, so you’re your own observer.

The color vision test is a pre-PCT test for the controlled variable. You find out the the exact form of the input function for transforming multiple wavelength light into color perception. There are probably good lessons to learn from the development and application of those tests.

RM: True.

AM: The color vision test is a pre-PCT test for the controlled variable.

RM: I think it’s a pre-PCT test of whether you can control a particular variable: the relationship between the character you see on the plate and the character you say.

RM: The TCV starts with a question about what variable is being controlled; the color vision test starts with a question about whether a particular variable can be controlled. The TCV ends with a description of the variable (or variables) that are being controlled when a person is performing a particular task; the color vision test ends with a description of the circumstances under which the variable can and cannot be controlled.

AM: You find out the the exact form of the input function for transforming multiple wavelength light into color perception.

RM: Yes, since you already know the perception to be controlled (the relationship between the character you see and the character you say) you can determine how variations in the relevant disturbances to that variable – the colors of the character and background – affect the ability to control that perception. What is found is that for most people, as long as the colors of character and background are different, they can control the relationship between the character they see and the character they say. But some people can’t control this perception even when the colors of character and background are different - some differences don’t make a difference.

RM: So you can learn a lot about why people behave the way they do – in this case, you can learn a lot about color perception – once you have a pretty good idea of what perceptions they control when they are carrying out various activities.

AM: There are probably good lessons to learn from the development and application of those tests.

RM: It would be nice if you could describe one or two of these lessons. I can think of any lessons from the color test work that would be relevant to doing the test for controlled variables. On that note, I think you didn’t quite get it right in your analysis of my helicopter following model. You said:

AM: In your helicopter following model, the variable alpha is an angle, the input function is a rate sensor, and the perceptual signal correlates with an angular velocity, not with the angle. The angle (qi) is not controlled, and does not correlate with the perception. I would call the angular velocity in the environment a controlled quantity, but not an input quantity.

RM: Actually qi is what is computed by the input function – it corresponds to the perception-- p – in the model. The angle that you call qi is itself a perception that is a lower level input to the input function.

RM: qi is the controlled variable – actually, a hypothesis about the controlled variable. Two possible qi’s were tested in the helicopter pursuit study: vertical optical velocity and vertical optical acceleration. These were two hypotheses about the perceptual variables – qi = d.alpha/dt or qi - d2.alpla/dt – controlled when intercepting the helicopters. These perceptions were defined in terms of a lower level perception, alpha – vertical optical angle.

RM: The test was done by seeing which hypothesis about the controlled variable – which qi – when used as the variable controlled in the vertical dimension, provided the best fit to the data. What we found is that the model that controlled vertical velocity – qi = d.alpha/dt – consistently fit the data better than the model that controlled vertical acceleration – qi = d2.alpha/dt.

RM: I did this analysis to show how to do the test for the controlled variable using modeling. I actually tested three hypotheses about the variables controlled when intercepting the helicopters – LOT, optical velocity and optical acceleration. LOT was tested in the “old-fashioned” way – by showing that linear optical trajectory (LOT) Is not protected from disturbance, such as irregular trajectories of the pursued object. I was actually surprised to find that it was possible to distinguish between optical velocity and acceleration as possible controlled variables since I thought these variables would be too highly confounded. But the results were pretty clear – optical velocity – not optical acceleration-- is the best hypothesis about the optical variable controlled when intercepting moving objects.

RM: This is the kind of PCT- based research that I would like to see being done. If you are interested – and I wish you were because you could do this stuff far better than I can – let’s continue the discussion over at the Powers’ Model of a PCT-Based Research Program topic.

RM: This is the kind of PCT- based research that I would like to see being done. If you are interested – and I wish you were because you could do this stuff far better than I can – let’s continue the discussion over at the Powers’ Model of a PCT-Based Research Program topic.

I’m super interested, and baseball research is good. It is just that I can’t see whether you are more dishonest or ignorant. That is a hard pass from me.

RM: It’s ignorance, but I might be lying.

I have been thinking about this question. I have many first-hand experiences about how the quality of control has worsened in controlling for understanding the discussant’s message because of missing hearing aids or too much background noise; or in controlling for understanding the written text because of missing eyeglasses or just too low light; etc. etc. But I understand the problem of testing: we cannot compare the perception straightly to its environmental correlate.

In a way, there are two “things” we must perceive for to control something. First, we must perceive that there is something to control. I would like to say that we must perceive some object, but perhaps it is enough to say that there must be the controlled variable and some context to it. This side could be called the what side of perception or control. Second, we must perceive the value of that variable. This side could be called the how much side. The first is necessary because we cannot perceive AND control just speed, size, or color but always the speed, size, or color of something. The accuracy of perception relates to these both sides of perception. Usually I have thought that especially with higher perceptions it is important for the controller to perceive accurately what it is what she wants to control because the type of output should be suitable for just that.

This what-side can, however, be especially difficult to test, but this text from Making Sense of Behavior (1998, p. 7) opens a nice possibility to test the how-much-side:

“… When you drive a car,
particularly in traffic or on a mountain road, you look where
you’re going almost continuously. On that twisty mountain road
with cars coming around every blind corner you don’t look away
for even one second.

The general rule is that if you want to control something you
have to perceive it. This doesn’t mean just perceiving that
something exists, as in looking out the windshield and noticing
that there’s a road out there. It means perceiving exactly the
aspect of the world that is supposed to be under control. You
don’t care what color the road is or whether it’s four lanes wide
or sis; when you’re trying to stay in your lane, you’re looking at
the relationship of the front of the car to the lane, and trying to
keep that visual picture in a certain configuration that you know
means you’re in your lane. You can’t do that blindfolded; you
can’t do it without watching what’s going on nearly all the time.”

This opens a possibility to test the effect of the accuracy of perception to the quality of control very simply. It could be a normal tracking study where the position of the target changes quasi randomly (smoothly) AND in addition the target (ad/or the cursor too) vanishes from the screen for short periods. The calculated position of the target and the position of the cursor determined by the controller simulate here the environmental correlate of the controller’s perception. You can easily change the accuracy of the perception by changing the time how often and for how long times the controller loses the sight to what is going on. (Of course also a disturbance could be added to the position of the cursor if it is affected via mouse or joystick.)

This way I think it could be easily tested whether the quality of control depends on how closely the perception being controlled corresponds to its environmental correlate. What do you think?

Best
Eetu

Rick:

To assess how closely a perception being controlled corresponds to its environmental correlate strikes me as a theoretical impossibility. Why? Because to assess their correspondence you can’t do any more than compare one perception with another. That said, I think we can do something similar.

For example, when driving I want to keep my car in the center of its lane. I use the alignment of the front left fender with the center line and/or the alignment of the right front fender with the edge line on the right side of the road. Now, let’s suppose I want to check that correspondence. I find a rarely traveled road and stop in what I believe to be the center of the lane. I get out and go to the front or back of my car and check how centered the car is in the lane. If it’s fine, no problem. If it’s off, I get back in my car and adjust it’s position then check to see if it’s centered. If it is, I get back in the car and note the new position of the fenders relative to the center line and right side of the road.

As I said at the outset, all I’ve got to go on are my perceptions but this sounds good enough for me and for practical purposes.

RM: What you are doing in your driving example is determining how accurately you are controlling one perceptual state of affairs – how centered the car looks from outside the car – when you are controlling a different perceptual state of affairs – how centered the car looks from inside the car. You probably consider the outside view to be the objective (or close to the objective) state of affairs, but, like the inside view, the outside view is also a perception. Powers (1978) called that outside-view perception an objectivized side effect of control – “objectified” in the sense that a perception (in this case the outside-view perception of the car relative to the lane) is taken as an objective (real reality) state of affairs.

RM: In this driving example, the inside (driver’s) view perception must correspond to the outside view perception (the “objectivized state of affairs”) or there will be an accident. The correspondence is obviously not one where the driver’s view is the same as the outside view. Rather, it is one where control of the driver’s view perception has the same effect on the outside view perception as it would if the driver were controlling the outside view. In order to keep a car centered in its lane, drivers must learn to control their own perception of the relationship of the car to its lane so that doing this has the effect of keeping the outside view perception – the objectivized state of affairs – in the state “centered”.

RM: What you are describing above, Fred, is not an example of finding the environmental correlate of a controlled variable. What you are describing is control of an objectivized side effect of a driver controlling his or her own perception. A person’s ability to do this – to produce the desired objectivized side effect of control – is the purview of engineering psychology – the study of man-machine interaction.

RM: An example of the use of engineering psychology to help a person avoid undesirable objectivized side effects of controlling the “wrong” perceptions is it’s application to prevention of the death spiral in aviation. The death spiral occurs when a pilot, deprived of the usual visual perceptions controlled when flying (due to fog or darkness) ends up controlling the wrong perceptions – those being the perceptions of orientation and movement provided by the pilot’s vestibular systems. When the pilot controls (or tries to control) these perceptions, the objectivized side effect of this control is the death spiral – uncontrolled spiraling descent into the ground (or ocean, as in the case of JFK Jr). In order to prevent this, pilots have to be trained to control the perception of their instrument readings instead, even if those readings conflict with their vestibular perceptions of orientation and movement.

RM: The difference between control theory as it is used in engineering psychology and PCT is that, in the former the controlled variables are known and in the latter the controlled variables are to be found out. Another way to put it is that PCT aims to find out how people work while engineering psychology aim to find out how to help people work better (at using their machines). Or as Powers put it:

RM: The progress in understanding the properties of persons provided by PCT centers around the discovery of the types of variables people control when they do the things they do.

Best

Rick

RM: As Mr. Bennett said to Mrs. Bennett in Pride and Prejudice, “I have not the pleasure of understanding you”. In particular, if the calculated position of the target and cursor simulate the environmental correlate of the controller’s perception then what simulates (or is) the controller’s perception?

Best

Rick

EP: This way I think it could be easily tested whether the quality of control depends on how closely the perception being controlled corresponds to its environmental correlate. What do you think?

RM: As Mr. Bennett said to Mrs. Bennett in Pride and Prejudice, “I have not the pleasure of understanding you”. In particular, if the calculated position of the target and cursor simulate the environmental correlate of the controller’s perception then what simulates (or is) the controller’s perception?

EP: No, the idea is NOT to compare the perception with it’s environmental correlate. (Perhaps it could be possible. but I don’t think it is necessary.) Instead, the idea is just to try to make the controller’s perception varyingly inaccurate by hiding the environmental correlate temporarily from the sight and then studying how that caused inaccuracy affects the quality of control.

Eetu

RM: You will find that the ability to control the variable you were controlling (the distance between cursor and target, in this case) will deteriorate considerably. This has nothing to do with the correlation between controlled perception and its environmental correlate; it has to do with removing the environmental "raw material ‘’ from which the controlled perception (distance between cursor and target) was constructed.

Best

Rick

RM: This has nothing to do with the correlation between controlled perception and its environmental correlate; it has to do with removing the environmental "raw material ‘’ from which the controlled perception (distance between cursor and target) was constructed.

EP: According to our environmental model the distance between cursor and target is continuously existing and quasi randomly changing in the computer. (The changing is not continuous in a digital computer, but that does not matter. The existence of the relation can anyway be thought to be continuous.) The perception can be thought to be either continuous or sampling. If we temporarily and repeatedly cut the connection between the perceiver and the perceived, then the perception will become sampling. The rate of sampling affects the accuracy of the perception. During the connection, the perception is as accurate as it can be, but between the connections the accuracy is insecure. The blinking of eyes and watching aside cause this kind sample insecurity to visual perception. (The input function can try to correct this kind of insecurity by anticipation / predicting. The simplest way of prediction is the assumption that the value of the perceived variable will remain stable between the samples. Even if there is no prediction in the input function, about the same effect will be caused by the leaky integrator output function. A little more complicated method of prediction is to assume that there will be similar change during next period as there was in the previous one.)

EP: This accuracy of the perception or its missing has always to do with the correlation between the perception and its environmental correlate: accuracy of perception = correlation between perception and its environmental correlate. The most simple and brutal way to affect the accuracy of the perception is to temporarily remove the environmental "raw material ‘’ from which the controlled perception was constructed. More delicate ways to affect the accuracy of perception in the test situations could using distorting glasses or utilizing optical illusions. Then the problem is that if these tricks will lessen the quality of control then the reorganization tends to correct the accuracy often quite quickly.

EP: A more generally interesting case is such where all the low-level perceptions are normally accurate (senses are functioning normally without external or internal distortions) but some higher-level perceptions is yet inaccurate. It means that even though the quality of control (QoC) of those low-level perceptions were good but still the QoC of the higher-level perception, which is constructed from those lower-level perceptions, is not good. An example of this could be a child who has learned to impose her will in her family but then does not success in a different context in the society, workplace, or neighbor family. These kinds of situations are difficult to imagine easily tested; and even in these the reorganization may quickly correct the problems of control.

EP: Anyway, that most simple and brutal test setting should prove that the accuracy of perception can and will affect the quality of control, doesn’t it?

RM: Yes, it will show the quality of control of a variable that you knew the subject was able to control before you introduced the disappearance of a component of that variable. Your demonstration is equivalent to demonstrating the effect of having a degraded sense of acidity on the ability to control for the taste of lemonade. Depending on the degree of degradation, the taste of what others perceive as lemonade will not taste like lemonade to you. This doesn;t mean that there is an environmental variable “out there” called lemonade that some can perceive accurately and you can’t. It just means that you can no longer create the perception that others call “lemonade” from the same sensory variables.

RM: I think the idea that the quality of control depends on the controller’s ability to correctly perceive environmental variables comes from viewing control from the point of view of a teacher, consultant or coach. These people presumably know what variables a student has to control in order to be able to carry out the task being taught. For example, a driving teacher tries to teach the student what variables to control in order to do parallel parking. The teacher sees those variables – such as the position of the student’s car relative to the car in front of the parking space, the angle of the car as it backes into the space, etc – as “out there” in the environment, and the teacher can see how well those variables are being controlled. But, as I said in another post, the job of finding out how well a person can control what are taken to be “environmental” variables is the purview of engineering psychology. The job of PCT is to find out what variables are controlled well.

Best

Rick