LCS III: Review of Chapter 2

[From Rick Marken (2014.02.19.1330)]

···

At long last here are my answers to the questions on Chapter 2 of LCS III. Feel free to continue discussing this chapter as long as you like. I’ll distribute the “Study Guide” for the next chapter – Chapter 3 – shortly.

1a. What do you think is the point of the first demo? What is

illustrated by the fact that you can control different aspects of the

display?

For me, the main point of the first demo is that we control perceptions of variable aspects of the environment. In the “Choose Control” the controller can pick one of three aspects of the same physical display and control it. What you actually do – in terms of the observed actions that are used to control the perception-- depends on which perceived aspect of the display you chose to control. That’s because what constitutes a disturbance to be resisted by those actions depends on what perceptual aspect of the world is being controlled. So the apparent effect of the external environment (disturbances) on behavior (actions that oppose disturbances) depends completely on what perceptual aspect of the work is under control. This demonstration illustrates the central insight of PCT and shows what distinguishes PCT from all other version of control theory that have been proposed as explanations of behavior. It shows that behavior is the control of perception, not of objective events in the world. So understanding behavior, from a PCT perspective, is a matter of understanding what perceptions are under control. Only when that is known – only when we know what perception(s) a system is controlling – can we start trying to figure out how it controls those perception.

  1. The section of Chapter 2 on The Input Function is the most

important section in the chapter for modelers. Why?

Because it is the input function that defines what perceptual aspect of physical reality is under control. In the “Choose Control” demo, the difference between controlling the shape, position or orientation of the object in the display is the difference in the input function around which thew control system is organized. The input function corresponds to neural networks in the organism that presumably transform the sensory input from the physical display into a neural current (see Premises in B:CP) that represent the state of the perceptual variable computed by the input function: for example, one input function that takes the sensed display as input and puts out a neural current whose varying magnitude corresponds to the varying orientation of the object in the display. Understanding how these input functions work should be the focus of studies of the neurophysiology of perception. Indeed, PCT suggests that the really important (and difficult) neurophysiological work that is aimed at getting an understanding of the neurophysiological basis of behavior will be research aimed at understanding how organisms perceive (how their input functions work) not so much on how they produce actions (output functions). Some recent research pertinent to this point is described in the this paper:

https://dl.dropboxusercontent.com/u/31298693/Hierarchical%20Control.pdf

The research shows, again, that what living control systems (people in this case) control is perceptions of the reality in their environment; the environment described by the models of physics and chemistry. In the “Hierarchy” paper we show what is demonstrated in the" Choose Control"; that people can control different perceptual aspects of the same reality. The “Hierarchy” paper just goes on to show (by presenting “reality” at different frame rates) that the different perceptual aspects of reality that are controlled in this experiment are hierarchically related.

The fact that the perceptions we control seem to be hierarchically related relates to David’s first question:

  1.  What levels of perception are needed to perform the task of chapter
    

2? Recall that a display changes in three ways–left/right position, shape
(round/nonround), and axis orientation (pointing towards you/not towards

you). Don’t forget the levels of perception needed to understand and follow

the task instructions.

The answer to this question (as well as several other of David’s questions) requires knowledge of aspects of PCT that have not been discussed yet (Or are not discussed at all) in LCS III. I think the specific levels of the hierarchy proposed in B:CP is one thing that is not discussed in LCS III. Bill was always asking people to view his proposals about the hierarchy as a source of hypotheses for PCT based research. We did use it as a source of our hypotheses about the relative levels of perceptual control in the Hierarchy paper described above. So I’m reluctant to say what levels of perception are represented by the different perceptions that can be controlled in the “Choose Control” demo. Indeed, I’m not sure these perceptions actually represent different levels of perception; they all seem like configuration type perceptions to me. Though position might be a relationship type perception because you control it relative to some other point on the screen (“next to” or “above”, for example, which are relationships between). Same with orientation; it might be a relationship perception since orientation implies a relationship to frame of reference. We could use the techniques described in the hierarchy paper to see if this hypothetical hierarchical relationship between these perceptions actually exists.

  1.  On Page 22, Bill says: "The only thing that changed was which aspect
    

you decided to pay attention to and hold constant." Where does this happen

in the PCT model?

Again, this is a question about an aspect of the PCT model that is not really discussed in LCS III. And it is not a very well understood aspect of the model because it has to do with consciousness. Attention seems to be about becoming aware of different parts of one’s perceptual world and awareness was described in B:CP as a consciousness phenomenon. How this process of shifting awareness works is not well understood by anyone; but at least in PCT we know that consciousness phenomena are different than control phenomena. Control is purposeful behavior, which can occur with or without consciousness.

  1.  Based on your answer to question 1, which aspect of the display do
    

you think a person should be able to control better and why?

Given my hypotheses about the type of perceptions controlled in the “Choose Control” demo I would say that shape should be easiest to control.

  1.  Consider Bill's statement on page 25: "A second way to find the
    

controlled variable is to look for the minimum correlation between the

disturbance and the aspect of the ball being influenced by the disturbance."

Can you explain this in terms of the properties of the correlation

statistic? (hint: what happens to the size of a correlation coefficient when

the range of one of the variables is restricted?)

Oops, this is not a result of restriction of range (which is a reduction in the size of the correlation between X and Y that results from looking at the correlation between X and Y for only a subrange (just as just the lower half of he range) of the X or Y variable. The low correlation between disturbance and “the aspect of the ball being influenced by the disturbance” which is the controlled perceptual variable (CV), results from the fact that the subject’s outputs are preventing the disturbance from moving the CV from it’s reference state. The range of the CV is being restricted by the actions of the system, not by the person computing the correlation coefficient. So the low correlation between disturbance and CV is a result of restriction of the range of the CV but this is quite different that the restriction of range that results in a reduced correlation coefficient. In the case of control, only the range of the CV is being “restricted” by the actions of the control system; in the statistical restriction of range case, both the range of both the X and Y variables is being restricted when you restrict the range of one variable (or the other).

  1.  From pages 27 to 35, Bill introduces the basic concept of a negative
    

feedback control system. On Page 28, he says: "It has to sense the variable

under control to know whether its magnitude is less than or greater than the

desired magnitude and of course the controller must affect the controlled

aspect so it can be brought closer to the desired value (which requires a

comparison). What order do you think is best for acquiring a new control

system using the three components of input, comparator and out functions?

Again, I think this is asking about something that is well outside the range of anything that has been dealt with in either LCS III or B:CP. It does sound like a good area for research though.

  1.  The addendum to chapter 2, from pages 35 through 40, has been
    

involved in our current discussion of Ashby and feedback versus feed

forward. I would simply like to direct you to a research project that Bill

and I did with adults which made use of the transfer function approach

rather than the modeling approach. Go to:

http://www.pctweb.org/Tracking.pdf

You are right that the addendum to chapter 2 is relevant to current discussions we’ve been having on CSGNet. It is particularly relevant to the discussion of “feedforward” and our perennial argument about whether the outputs of a control system are based on information about the disturbance. I’ll just let that section speak for itself.

Best

Rick


Richard S. Marken PhD
www.mindreadings.com
The only thing that will redeem mankind is cooperation.
– Bertrand Russell

Content-Type: application/vnd.openxmlformats-officedocument.spreadsheetml.sheet;
  name="RestrictionOfRange v Control.xlsx"
Content-Disposition: attachment;
  filename="RestrictionOfRange v Control.xlsx"
X-Attachment-Id: f_hrxsn72l0

RestrictionOfRange v Control.xlsx (22.7 KB)

Dear all,

has anybody seen the following article of February 18:

Shanechi, M. M., Hu, R. C. & Williams, Z. M. A cortical-spinal prosthesis for targeted limb movement in paralysed primate avatars. Nat Commun 5, 3237, doi:10.1038/ncomms4237 (2014).

The abstract is as follows:

Motor paralysis is among the most disabling aspects of injury to the central nervous system. Here we develop and test a target-based cortical-spinal neural prosthesis that employs neural activity recorded from premotor neurons to control limb movements in functionally paralysed primate avatars. Given the complexity by which muscle contractions are naturally controlled, we approach the problem of eliciting goal-directed limb movement in paralysed animals by focusing on the intended targets of movement rather than their intermediate trajectories. We then match this information in real-time with spinal cord and muscle stimulation parameters that produce free planar limb movements to those intended target locations. We demonstrate that both the decoded activities of premotor populations and their adaptive responses can be used, after brief training, to effectively direct an avatar’s limb to distinct targets variably displayed on a screen. These findings advance the future possibility of reconstituting targeted limb movement in paralysed subjects.

As they are focusing on the intended targets of movements instead of their intermediate trajectories, isn’t this the first physiological proof of this kind that behavior is the control of perception?

Best, Frans

···

[From Rick Marken (2014.02.21.1100)]

David Goldstein disputed my criticism of the “restriction of range” explanation of the low correlation between disturbance and controlled variable in a control system. Here’s the relevant part of the discussion.

RM: It certainly looks like the low correlation between disturbance (d) and controlled variable (c) is an example of “restriction of range”. In statistics, “restriction of range” means that the correlation between two variables, like c and d, goes down as the range of one of the variables is restricted; that is,when only a subset of the entire range of one of the variables (and the values of the other variable that are associated with that subrange) is included in the analysis.

This sounds like what is going on in control. In the control situation the range of one variable, c, would be equal to the range of the another, d, if the control system weren’t controlling c. And controlling c definitely reduces the range of variation of c. With no control, the range of c is the same as the range of d and the correlation between d and c is 1.0; with good control, the range of d is still the same as it was without control but the range of c is considerably restricted and, sure enough, the correlation between d and c goes down, ultimately to 0 when control is perfect.

But the reduced correlation between c and d when there is control is not a result of the restriction of range of c; it’s a result of the active cancellation of the effect of d on c by the output of a control system. I’ve cobbled together a spreadsheet (attached) to try to illustrate this point. Using a simulated control system I’ve demonstrated how you get a small correlation between d and c when the range of c is reduced by the actions of a control system and a high (indeed, perfect) correlation between d and c when the range of c is reduced by just multiplying the the c values by fraction.

The main results are in the upper right corner of the sheet. “d range” is the range of the disturbance - 200 in this case. “c range” is the range of c, the variable controlled by the control system (12.54, pretty good control); “c’ range” is the range of c values computed as .0001*d (.02). “r d-c” is the correlation between d and c (.19). This is the low correlation we always see when we correlate d and c from a tracking task done by a relative skilled controller."r d-c’ "is the correlation between d and c’ (1.0). Note that even though the range of c’ is much smaller than the range of c the correlation between d and c’ is much higher than that between d and c.

So it’s not the reduced (“restricted”) range of c relative to the range of d that is responsible for for the low correlation between d and c in a control task. It’s the fact that c = d-o, with o nearly exactly equal to d.

I’ve also included an example of the effect of “restriction of range” on correlation as it is meant in statistics. I looked at the correlation between two variables, X and Y; the correlation between X and Y (rX-Y) was .84. The range of the X variable (X range) was 0-200. I then selected for further analysis only the X scores (and associated Y scores) that feel in the range 0 - 100. In other words, I restricted the X,Y pairs to be included in the analysis to those with X scores in the lower half of the X range of scores. The X and Y scores from this subrange are called X’and Y’ and the resulting correlation between these scores (r X’-Y’) is only ,6. So by including scores from only a restricted range of one of the variables involved in a correlation analysis (X in this case) you end up with a reduced correlation between the values of the variables in this restricted range. This is the “restriction of range” effect described in statistics texts.

So hopefully you can see that the statistical “restriction of range” reduction in the correlation between two variables is nothing like the reduction in the correlation between disturbance and controlled variable that is seen when control “restricts the range” of variation of the controlled variable.

Best

Rick


Richard S. Marken PhD
www.mindreadings.com
The only thing that will redeem mankind is cooperation.

                                               -- Bertrand Russel

DG: > 4.Consider Bill’s statement on page 25: "A second way to find the

controlled variable is to look for the minimum correlation between the

disturbance and the aspect of the ball being influenced by the disturbance."

Can you explain this in terms of the properties of the correlation

statistic? (hint: what happens to the size of a correlation coefficient when

the range of one of the variables is restricted?)

Rm: Oops, this is not a result of restriction of range (which is a reduction in the size of the correlation between X and Y that results from looking at the correlation between X and Y for only a subrange (just as just the lower half of he range) of the X or Y variable. The low correlation between disturbance and “the aspect of the ball being influenced by the disturbance” which is the controlled perceptual variable (CV), results from the fact that the subject’s outputs are preventing the disturbance from moving the CV from it’s reference state. The range of the CV is being restricted by the actions of the system, not by the person computing the correlation coefficient. So the low correlation between disturbance and CV is a result of restriction of the range of the CV but this is quite different that the restriction of range that results in a reduced correlation coefficient. In the case of control, only the range of the CV is being “restricted” by the actions of the control system; in the statistical restriction of range case, both the range of both the X and Y variables is being restricted when you restrict the range of one variable (or the other).

[Fred Nickols (02.22.2014.0817 EST)]

Frans:

Many thanks for this. I think you are correct. I am especially glad they used the term “target” as it fits with my Target Model of Human Behavior and Performance which is based on PCT.

Thanks again.

Fred Nickols

···

From: Frans Plooij [mailto:fplooij@KIDDYGROUP.COM]
Sent: Saturday, February 22, 2014 7:25 AM
To: CSGNET@LISTSERV.ILLINOIS.EDU
Subject: Re: LCS III: Review of Chapter 2

Dear all,

has anybody seen the following article of February 18:

Shanechi, M. M., Hu, R. C. & Williams, Z. M. A cortical-spinal prosthesis for targeted limb movement in paralysed primate avatars. Nat Commun 5, 3237, doi:10.1038/ncomms4237 (2014).

The abstract is as follows:

Motor paralysis is among the most disabling aspects of injury to the central nervous system. Here we develop and test a target-based cortical-spinal neural prosthesis that employs neural activity recorded from premotor neurons to control limb movements in functionally paralysed primate avatars. Given the complexity by which muscle contractions are naturally controlled, we approach the problem of eliciting goal-directed limb movement in paralysed animals by focusing on the intended targets of movement rather than their intermediate trajectories. We then match this information in real-time with spinal cord and muscle stimulation parameters that produce free planar limb movements to those intended target locations. We demonstrate that both the decoded activities of premotor populations and their adaptive responses can be used, after brief training, to effectively direct an avatar’s limb to distinct targets variably displayed on a screen. These findings advance the future possibility of reconstituting targeted limb movement in paralysed subjects.

As they are focusing on the intended targets of movements instead of their intermediate trajectories, isn’t this the first physiological proof of this kind that behavior is the control of perception?

Best, Frans


Dr. Frans X. Plooij
Director
International Research-institute on Infant Studies (IRIS)
Zijpendaalseweg 73
6814 CE Arnhem
The Netherlands
Mobile: +31 6 460 888 20
Email: fplooij@kiddygroup.com
Tel.: +31 26 389 4841
Fax: +31 26 389 4493

Op 21 feb. 2014, om 20:04 heeft Richard Marken rsmarken@GMAIL.COM het volgende geschreven:

[From Rick Marken (2014.02.21.1100)]

David Goldstein disputed my criticism of the “restriction of range” explanation of the low correlation between disturbance and controlled variable in a control system. Here’s the relevant part of the discussion.

DG: > 4.Consider Bill’s statement on page 25: "A second way to find the

controlled variable is to look for the minimum correlation between the
disturbance and the aspect of the ball being influenced by the disturbance."
Can you explain this in terms of the properties of the correlation
statistic? (hint: what happens to the size of a correlation coefficient when
the range of one of the variables is restricted?)

Rm: Oops, this is not a result of restriction of range (which is a reduction in the size of the correlation between X and Y that results from looking at the correlation between X and Y for only a subrange (just as just the lower half of he range) of the X or Y variable. The low correlation between disturbance and “the aspect of the ball being influenced by the disturbance” which is the controlled perceptual variable (CV), results from the fact that the subject’s outputs are preventing the disturbance from moving the CV from it’s reference state. The range of the CV is being restricted by the actions of the system, not by the person computing the correlation coefficient. So the low correlation between disturbance and CV is a result of restriction of the range of the CV but this is quite different that the restriction of range that results in a reduced correlation coefficient. In the case of control, only the range of the CV is being “restricted” by the actions of the control system; in the statistical restriction of range case, both the range of both the X and Y variables is being restricted when you restrict the range of one variable (or the other).

RM: It certainly looks like the low correlation between disturbance (d) and controlled variable (c) is an example of “restriction of range”. In statistics, “restriction of range” means that the correlation between two variables, like c and d, goes down as the range of one of the variables is restricted; that is,when only a subset of the entire range of one of the variables (and the values of the other variable that are associated with that subrange) is included in the analysis.

This sounds like what is going on in control. In the control situation the range of one variable, c, would be equal to the range of the another, d, if the control system weren’t controlling c. And controlling c definitely reduces the range of variation of c. With no control, the range of c is the same as the range of d and the correlation between d and c is 1.0; with good control, the range of d is still the same as it was without control but the range of c is considerably restricted and, sure enough, the correlation between d and c goes down, ultimately to 0 when control is perfect.

But the reduced correlation between c and d when there is control is not a result of the restriction of range of c; it’s a result of the active cancellation of the effect of d on c by the output of a control system. I’ve cobbled together a spreadsheet (attached) to try to illustrate this point. Using a simulated control system I’ve demonstrated how you get a small correlation between d and c when the range of c is reduced by the actions of a control system and a high (indeed, perfect) correlation between d and c when the range of c is reduced by just multiplying the the c values by fraction.

The main results are in the upper right corner of the sheet. “d range” is the range of the disturbance - 200 in this case. “c range” is the range of c, the variable controlled by the control system (12.54, pretty good control); “c’ range” is the range of c values computed as .0001*d (.02). “r d-c” is the correlation between d and c (.19). This is the low correlation we always see when we correlate d and c from a tracking task done by a relative skilled controller."r d-c’ "is the correlation between d and c’ (1.0). Note that even though the range of c’ is much smaller than the range of c the correlation between d and c’ is much higher than that between d and c.

So it’s not the reduced (“restricted”) range of c relative to the range of d that is responsible for for the low correlation between d and c in a control task. It’s the fact that c = d-o, with o nearly exactly equal to d.

I’ve also included an example of the effect of “restriction of range” on correlation as it is meant in statistics. I looked at the correlation between two variables, X and Y; the correlation between X and Y (rX-Y) was .84. The range of the X variable (X range) was 0-200. I then selected for further analysis only the X scores (and associated Y scores) that feel in the range 0 - 100. In other words, I restricted the X,Y pairs to be included in the analysis to those with X scores in the lower half of the X range of scores. The X and Y scores from this subrange are called X’and Y’ and the resulting correlation between these scores (r X’-Y’) is only ,6. So by including scores from only a restricted range of one of the variables involved in a correlation analysis (X in this case) you end up with a reduced correlation between the values of the variables in this restricted range. This is the “restriction of range” effect described in statistics texts.

So hopefully you can see that the statistical “restriction of range” reduction in the correlation between two variables is nothing like the reduction in the correlation between disturbance and controlled variable that is seen when control “restricts the range” of variation of the controlled variable.

Best

Rick

Richard S. Marken PhD
www.mindreadings.com

The only thing that will redeem mankind is cooperation.
– Bertrand Russel

[From Rick Marken (2014.02.22.1400)]

···

Hi Frans

On Sat, Feb 22, 2014 at 4:24 AM, Frans Plooij fplooij@kiddygroup.com wrote:

Dear all,
has anybody seen the following article of February 18:

Shanechi, M. M., Hu, R. C. & Williams, Z. M. A cortical-spinal prosthesis for targeted limb movement in paralysed primate avatars. Nat Commun 5, 3237,…

As they are focusing on the intended targets of movements instead of their intermediate trajectories, isn’t this the first physiological proof of this kind that behavior is the control of perception?

RM: Nice find. It’s kind of hard to tell, just from the abstract, what they’ve done. It sounds to me like they are using the target location information as the basis for computing the “spinal cord and muscle stimulation parameters” that move the limb to the target. So it sounds more like a computed output rather than a control of input type of system. I would consider it a control of input system if it could track a moving target. Doing that would require constant revision of the “target -based” calculations. Indeed, these target based calculations would become unnecessary if it were controlling a perception since the appropriate outputs could be produced on the basis of the current difference between target and pointing limb. Now that I think of it, since they didn’t mention revising their calculations based on a perception of the difference between target and actual pointing position I would say that this is almost certainly not a control of perception type system.

But your post reminded me that my sister-in-law pointed me to a report of a recently developed sensory-based prosthesis that does illustrate the importance of control of perception in motor behavior. A press report on the system is here:

http://www.latimes.com/science/sciencenow/la-sci-sn-sensorized-prosthetic-hand-20140204,0,4329957.story#axzz2sTjlxwzu

And the abstract of the research report is here:

http://stm.sciencemag.org/content/6/222/222ra19

This is a remarkable prosthesis that makes it possible for the hand amputee to control perceptions of the force exerted by his prosthetic fingers when doing things like grasping various objects. The amputee doesn’t describe it this way but what he is clearly exited about is that he is finally able to use his prosthetic hand in the same way as he uses his actual hand: to control a perception which is the one he is used to controlling with his hand – the perception of the force exerted by the muscles of the fingers when grasping, typing, etc. With the usual non-sensory prosthesis these motor activities had to be controlled visually. That is, the amputee had to control visual perceptions of gripping, pushing, etc. With the sensory prosthetic the amputee could do these things without looking, just like non-amputees.

This sensory prosthesis is, indeed, a very good “proof” that behavior is the control of perception.

Best

Rick


Richard S. Marken PhD
www.mindreadings.com
The only thing that will redeem mankind is cooperation.
– Bertrand Russel

[From Frans Plooij (2014.02.23.1458)]

Thanks Rick for the paper on the sensory based prosthesis. That is the kind of ‚proof’ I was looking for.

Best, Frans

···

Hi Frans

On Sat, Feb 22, 2014 at 4:24 AM, Frans Plooij fplooij@kiddygroup.com wrote:

Dear all,
has anybody seen the following article of February 18:

Shanechi, M. M., Hu, R. C. & Williams, Z. M. A cortical-spinal prosthesis for targeted limb movement in paralysed primate avatars. Nat Commun 5, 3237,…

As they are focusing on the intended targets of movements instead of their intermediate trajectories, isn’t this the first physiological proof of this kind that behavior is the control of perception?

RM: Nice find. It’s kind of hard to tell, just from the abstract, what they’ve done. It sounds to me like they are using the target location information as the basis for computing the “spinal cord and muscle stimulation parameters” that move the limb to the target. So it sounds more like a computed output rather than a control of input type of system. I would consider it a control of input system if it could track a moving target. Doing that would require constant revision of the “target -based” calculations. Indeed, these target based calculations would become unnecessary if it were controlling a perception since the appropriate outputs could be produced on the basis of the current difference between target and pointing limb. Now that I think of it, since they didn’t mention revising their calculations based on a perception of the difference between target and actual pointing position I would say that this is almost certainly not a control of perception type system.

But your post reminded me that my sister-in-law pointed me to a report of a recently developed sensory-based prosthesis that does illustrate the importance of control of perception in motor behavior. A press report on the system is here:

http://www.latimes.com/science/sciencenow/la-sci-sn-sensorized-prosthetic-hand-20140204,0,4329957.story#axzz2sTjlxwzu

And the abstract of the research report is here:

http://stm.sciencemag.org/content/6/222/222ra19

This is a remarkable prosthesis that makes it possible for the hand amputee to control perceptions of the force exerted by his prosthetic fingers when doing things like grasping various objects. The amputee doesn’t describe it this way but what he is clearly exited about is that he is finally able to use his prosthetic hand in the same way as he uses his actual hand: to control a perception which is the one he is used to controlling with his hand – the perception of the force exerted by the muscles of the fingers when grasping, typing, etc. With the usual non-sensory prosthesis these motor activities had to be controlled visually. That is, the amputee had to control visual perceptions of gripping, pushing, etc. With the sensory prosthetic the amputee could do these things without looking, just like non-amputees.

This sensory prosthesis is, indeed, a very good “proof” that behavior is the control of perception.

Best

Rick


Richard S. Marken PhD
www.mindreadings.com
The only thing that will redeem mankind is cooperation.
– Bertrand Russel