Control of Perception (was Re: Powers reference wanted)

[From Rick Marken (2014.07.15.1050)]

···

Martin Taylor (2014.07.14.14.40)–

              MT: True, but since in the absence of control, var(o)

= 0, cv = d.

            RM: I think you must know that this is not the case.

But if not, you can demonstrate to yourself that it’s
not true using, once again, the “Control of Size” demo (http://www.mindreadings.com/ControlDemo/Size.html).

  MT: As usual, we are talking at cross-purposes. I am comparing

open-loop with closed-loop conditions, whereas you are introducing
an added disturbance that is a side-effect of controlling
something else. It’s not surprising we come to different
conclusions.

MT: I think I didn’t word that very well. Let’s try again.

MT: A side-effect is an effect on something that is not in the control loop of the perception being controlled…when in Rick’s demo you control the area, a side-effect is that the perception of perimeter length may be influenced.

RM: I think you miss the point of the demo, which is that there are two perceptual aspects of the same physical situation that can be controlled. The physical situation is the rectangle on the screen which has height (h) and width (w). The computer varies h and mouse movement varies w. The two aspects of this situation that can be controlled are area (a) and perimeter (p):

a = w * h

p = 2* (w+h)

So a and p are two different aspects of the same physical variables, w and h. In order to control either a or p the subject must move the mouse to vary w in order to appropriately compensate for changes in the computer generated variations in h. So variations in w are the outputs (o) of the system controlling either a or p; and variations in h ara the disturbance to a and p. So we can write the equations for the two possible perceptions controlled in this demo as:

a = o * d

p = 2* (o + d)

If the subjects want to control a they must vary their outputs so that:

o = a’/d (1)

where a’ is the desired value of the area. If subjects want to control p they must vary their outputs so that

o = p’/2-d (2)

where p’ is the desired value of the perimeter.

Which perceptual aspect of the display the subject controls is selected by the subject. But regardless of which aspect of the display the subject chooses to control, a or p, there output will be varied as the means of controlling it. So o will be varying whether the subject is controlling a or p; output (o) will be varying as per eq. (1) if the subject is controlling a and as per eq. (2) if the subject is controlling p. Also, regardless of whether a or p is being controlled, both a and p will be varying, although a will be varying much less than p if a is under control and p will be varying less than a if p is controlled.

The demonstration shows that you can determine which perceptual aspect of the same physical situation is being controlled by the subject be computing the stability factor for both possible controlled perceptions, a and p. The stability factor for a is

S = 1 - var(log(a))/(var(log(o)+var(log(d)))

and that for p is

S = 1 - var(p))/(var(o)+var(d))

var(log(o)) and var(o) will be non-zero whether the subject is not controlling a (and, thus, controlling p) or not controlling p (and thus controlling a). So it’s not true that var(o) is zero when a perception is not under control. Outputs are always affecting many different perceptual aspects of the world simultaneously (as in the demo where variations in w are affecting both a and p at the same time). The goal of testing for controlled perceptions is to determine which perceptual aspects of a situation are being controlled by those outputs. The “Control of Size” demo is another approach to showing what is meant by control of perception: that we control various aspects of physical reality and that we can control different aspects of the same physical reality. Because of the latter fact, it is often difficult to determine what an organism is doing (what perceptual aspect of its environment it is controlling).

MT: Since, by the definition of the problem, the perception of perimeter is not being controlled, the output of the perimeter control system is zero, and any changes in the perimeter perception are due to the joint influences of the disturbance to, and any side-effects of, the area control system.

RM: Hopefully you can see now that this statement is incorrect.

Best regards

Rick


Richard S. Marken PhD

www.mindreadings.com

[Martin Taylor 2014.07.15.23.12]

You are saying that th subject simply cannot see the area when

controlling the perimeter, and vice-versa? If he cannot see it, how
could he possibly ever later choose to control it?
So does mine, at the level of the environment. But not at the level
of the control unit.
True of both diagrams.
So the neurophysiology says that if you choose to use your hand to
lift a cup to your mouth, you cannot see the pencil that you might
use your hand to write with? Lord Nelson may have used that trick,
but he controlled it by putting the telescope to his blind eye so
that he could correctly say he saw no signal. In the present case we
have not presumed the subject to have imposed blinders such that he
cannot see the perimeter when he chooses to control area, and
vice-versa.
What a strange thing to say. Normally we expect different
perceptions to be controlled by different control systems. Are you
now going to remake your 3-level spreadsheet so that there is only
one control unit for all the top level controlled perceptions, to
conform to this remarkable new claim that no two higher-level
perceptions can use the same lower-level control units in their
feedback paths?
Yes, that is what I said.
Yes, that is what I said.
Yes, if you use the words in a constant way, but instead you do the
Philosopher’s trick of using the same word in two different ways.
“The system” can mean the hierarchical complex, some subset of it,
or a single ECU, and “outputs” can mean externally observable
effects or the outputs of a single ECU. If you make sure to specify
which you mean when you use the terms, you won’t get into the mess
you find yourself in this unnecessary argument.
In this case, the “Subset of the hierarchical complex” “output to
the external environment” is the “o” in your expressions. On the
other hand, the “single ECU” controlling whichever perception is
being controlled has its own output to the lower levels of the
hierarchy, while the “single ECU” not controlling its perception has
no output to the lower levels and no influence on the “Subset of the
hierarchical complex” “output to the external environment”. The
“single ECU” outputs are the ones that matter when quantifying QoC.
The disturbance to the uncontrolled variable, let’s say area in this
case, is the product of the externally imposed disturbance to the
height and the variation of the width caused by the control of the
perimeter. That’s “d” in your expressions, when the cv is area. When
the cv is the perimeter, “d” is the externally imposed disturbance
to the height. If you wanted to make a better diagram, you should
probably include the lower-level perceptions of height and width,
and show that only the width is controlled at that level, regardless
of which of the higher-level combinations of width and height is
being controlled.
Again you use the same word – symbol in this case – to mean two
different things. “d” is used once to mean the disturbance to a CV,
which is any influence on the CV other than that produced by the
output of the controller of the perception of that CV. “d” also
means an experimenter-imposed variation of a component of the
perception. It’s unhelpful to use the symbol both ways in the same
pair of expressions.
That’s rather an irrelevant comment, when your objective is to argue
that I was misleading Boris about the meaning of Quality of Control.
I imagine that by now, Boris is rather confused. So I reiterate what
I said initially, adding “open loop” as a clarification not intended
to change the sense of it in any way: “.”
Martin

Re Control of Perception (was (108 Bytes)

Re Control of Perception (was1 (109 Bytes)

···

[From Rick Marken (2014.07.15.2000)]

            Martin Taylor

(2014.07.15.16.39)–

            MT: Here's a picture

of the situation. The dashed line means the two switches
are connected so that if the subject chooses to control
one perception, the other is not controlled. In the
picture, area is being controlled. A side-effect of that
control is a disturbance to the perception of perimeter,
which combines with the externally applied disturbance
to the height to form the overall disturbance to the
perimeter.

          Nice picture but I would prefer to put the switch on

the input side, like this (pictured with the perceptual
switch in “control of perimeter” position):

          RM: This seems more consistent with the actual

situation; either perception – area or perimeter –
depending on which perceptual function is switched in, is
controlled using the same output.

                          RM: So variations in w are the outputs

(o) of the system controlling either a or
p;

            MT: Almost right. Note the word "either". But what is

missing are the words “to the external environment” as
in "* So variations in w are the outputs (o) to the
external environment of the system controlling either
a or p* ". Notice that at any moment one or other of
those two systems does not influence the muscle-level
systems.

          RM: That's just because of the way you diagrammed it.

My diagram, which controls either perception also controls
them using the same output.

          In my diagram o corresponds to mouse movement, which

has a proportional effect on w. So either o or w can be
considered the output that affect the controlled variable,
area or perimeter.

MT: That one produces NO output in its own control loop.

          RM: Yes, again only because of the way you modeled it.

I prefer my model because it seems somewhat more
consistent with the neurophysiology.

          It seems highly unlikely that two separate control

systems exist to carry out this task, one to control area
and another to control perimeter, each with separate
efferent connections to the muscles.

          RM: Well I do agree that this is probably confusing to

readers, most of whom probably couldn’t care less. But
this all started because you were explaining to someone
how to measure quality of control (QoC) using the
stability factor and you described it incorrectly. You
said:

            MT:

For a single control unit, Quality of Control can be
measured in a variety of ways, but they all come down to
the same thing: the ratio between some measure of
variation of the controlled perception and the variation
that would be observed if only the disturbance
influenced the perception.

            RM:

I just pointed out that that is not quite correct; the
measure of QoC in terms of stability is the ratio of
observed to expected variation in the cv.

            The

observed variation is just the variation in the variable
that is suspected to be the CV; the expected variation
in the cv is the variation that is expected if there
were no control. In order to determine what the expected
variation in the cv would be sans control you have to
determine how all influences on the controlled
variable would have affected it.

            There are two main influences on any possible controlled

variable: the effect of external disturbances and the
effect of the system’s own outputs.

            So

QoC – the ratio of observed to expected variance of a
hypothetical controlled variable – is var(cv)/[var(d)+var(o)],
not var(cv)/var(d),
as implied by your last statement.

But
truth be told, it really doesn’t make much difference.
If you use var(cv)/var(d)
rather than var(cv)/[var(d)+var(o)]
to measure QoC you are equally able to discriminate
control of of area from control of perimeter.

            So

never mind.

  •  For a single control unit,
    

Quality of Control can be measured in a variety of ways, but they
all come down to the same thing: the ratio between some measure of
variation of the controlled perception and the variation under
open loop conditions when the perception is not being controlled
and only the disturbance influences it. Often the measure of
variation used is the RMS variation of the perception. It could be
the amplitude. It really doesn’t matter in this context*

···

[From Rick Marken (2014.07.17.1040)]

Martin Taylor (2014.07.15.23.12) –

MT: You are saying that th subject simply cannot see the area when

controlling the perimeter, and vice-versa?

RM: No. In my model both perceptions exist as neural perceptual signals; the subject switches one perception or the other in as the input to the control loop. A higher order system would have to do the switching, as would presumably also be true of the output switching in you model.

MT: What a strange thing to say. Normally we expect different

perceptions to be controlled by different control systems.

RM: I agree. I put the switches in to make it more comparable to your output switching model. But I think a better model would be hierarchical, like the one in Marken, R. S., Khatib, Z. and Mansell, W. (2013) Motor Control as the
Control of Perception, Perceptual and
Motor Skills
, 117, 236-247, copied below. Noted that control of each of the different type of perception uses the same output, as in the “Control of Size” demo; all that changes is which perception in the hierarchy is controlled.

RM: It would be interesting to use reaction time methods like those described in the Marken et al paper to see which perception, area or perimeter, is higher in the hierarchy! Once again, Martin, you have suggested a nice, new little research study for me!

MT: That's rather an irrelevant comment, when your objective is to argue

that I was misleading Boris about the meaning of Quality of Control.
I imagine that by now, Boris is rather confused. So I reiterate what
I said initially, adding “open loop” as a clarification not intended
to change the sense of it in any way: “* For a single control unit,
Quality of Control can be measured in a variety of ways, but they
all come down to the same thing: the ratio between some measure of
variation of the controlled perception and the variation under
open loop conditions when the perception is not being controlled
and only the disturbance influences it. Often the measure of
variation used is the RMS variation of the perception. It could be
the amplitude. It really doesn’t matter in this context*.”

RM: Yes, that’s fine. I would only insert the word “hypothetical” before “controlled variable”. I’m sorry I made things confusing. Bill did define the stability factor on pp. 160 and 161 of LCS as 1-sqrt(Vexp/Vobs) where Vexp was desfined as the sum of Var(o) and Var(d). But it turns out that using Var(d) alone as Vexp gives you proportionately the same results, even when there is large Var (o) due to secular variations in the reference specification for the controlled variable. So it’s a difference that really doesn’t make a practical difference. Ergo, my “never mind”.

Best regards

Rick


Richard S. Marken PhD
www.mindreadings.com

          RM: It seems highly unlikely that two separate control

systems exist to carry out this task, one to control area
and another to control perimeter, each with separate
efferent connections to the muscles.

            RM: But

truth be told, it really doesn’t make much difference.
If you use var(cv)/var(d)
rather than var(cv)/[var(d)+var(o)]
to measure QoC you are equally able to discriminate
control of of area from control of perimeter.

            RM: So

never mind.

[Martin Taylor 2014.07.17.15.10]

Why would you expect either to be higher then the other?

But that’s irrelevant. The diagram you show enforces a conflict if
the pathway m( )->o->qi has fewer than three degrees of
freedom. It has only one df in the situation we are discussing. To
avoid the conflict, you say that only one of the perceptions is
being controlled. The diagram doesn’t show this. If you do put switches into the paths either before or after the
comparators (or if you use a higher-level control to turn the output
gain to zero, which is the same thing) you have my diagram, with
zero contribution to qi from the ones not being controlled. If you
don’t, how do you prevent them from attempting to control and
creating an escalating conflict?
Of course, I didn’t show the many layers of control loops that
produce the eventual muscular output, just as you have shown no
output stages at all except m( ), which I presume to mean muscles.
We both presume that the same muscles, and many of the stages before
the muscles, are used to control the same mouse movement, no matter
what is being controlled. So we appear to agree on what is
happening, despite the diagram. What we don’t agree on is the labelling. I argue that the output of
the ECU that controls, say, p1 in your diagram is a disturbance to
p2 and p3, not a consequence of their own ECU output, of which there
is none if they are not controlling. You say the effect of
controlling p1 on the value of uncontrolled p2 or p3 isn’t a
disturbance, because if they did happen next week to be controlling,
the effect of their output would appear at the same place, qi. I
don’t get it.

···

[From Rick Marken (2014.07.17.1040)]

            Martin Taylor

(2014.07.15.23.12) –

            MT: You are saying that th subject simply cannot see the

area when controlling the perimeter, and vice-versa?

          RM: No. In my model both perceptions exist as neural

perceptual signals; the subject switches one perception or
the other in as the input to the control loop. A higher
order system would have to do the switching, as would
presumably also be true of the output switching in you
model.

                        RM: It seems highly unlikely that two

separate control systems exist to carry out
this task, one to control area and another
to control perimeter, each with separate
efferent connections to the muscles.

            MT: What a strange thing to say. Normally we expect

different perceptions to be controlled by different
control systems.

          RM: I agree. I put the switches in to make it more

comparable to your output switching model. But I think a
better model would be hierarchical, like t he one in Marken, R.
S., Khatib, Z. and Mansell, W. (2013) Motor Control as
the
Control of Perception, * Perceptual and
Motor Skills* , 117, 236-247, copied below. Noted
that control of each of the different type of perception
uses the same output, as in the “Control of Size” demo;
all that changes is which perception in the hierarchy is
controlled.

            RM:

It would be interesting to use reaction time methods
like those described in the Marken et al paper to see
which perception, area or perimeter, is higher in the
hierarchy! Once again, Martin, you have suggested a
nice, new little research study for me!

                          RM:

But truth be told, it really doesn’t make
much difference. If you use
var(cv)/var(d)
rather than
var(cv)/[var(d)+var(o)]
to measure QoC you are equally able to
discriminate control of of area from
control of perimeter.

                          RM:

So never mind.

            MT: That's rather an irrelevant comment, when your

objective is to argue that I was misleading Boris about
the meaning of Quality of Control. I imagine that by
now, Boris is rather confused. So I reiterate what I
said initially, adding “open loop” as a clarification
not intended to change the sense of it in any way: “* For
a single control unit, Quality of Control can be
measured in a variety of ways, but they all come down
to the same thing: the ratio between some measure of
variation of the controlled perception and the
variation under open loop conditions when the
perception is not being controlled and only the
disturbance influences it. Often the measure of
variation used is the RMS variation of the perception.
It could be the amplitude. It really doesn’t matter in
this context*.”

          RM: Yes, that's fine. I would only insert the word

“hypothetical” before “controlled variable”.

[From Rick Marken (2014.07.19.0940)]

Martin Taylor (2014.07.17.15.10)--

RM: It would be interesting to use reaction time methods like those described in the Marken et al paper to see which perception, area or perimeter, is higher in the hierarchy! Once again, Martin, you have suggested a nice, new little research study for me!

MT: Why would you expect either to be higher then the other?

RM: I expect it based on one of Bill's methods of identifying relative levels of perception, I would expect area to be a higher level perception than perimeter because you have to be able to perceive perimeter (width and height) to perceive area but you don't need to perceive area in order to perceive perimeter. So I predict that perimeter is a lower level perception than area and that, therefore, you can control perimeter at a faster rate of variation than area. This prediction can actually be tested using the "What is size" demo. Can you see how?

MT: The diagram you show enforces a conflict if the pathway m( )->o->qi has fewer than three degrees of freedom. It has only one df in the situation we are discussing. To avoid the conflict, you say that only one of the perceptions is being controlled. The diagram doesn't show this.

RM: That's true. The diagram leaves out the mechanism that shifts control from perceptions at one level versus another. This problem can be solved using switches again. >

MT: If you do put switches into the paths either before or after the comparators (or if you use a higher-level control to turn the output gain to zero, which is the same thing) you have my diagram, with zero contribution to qi from the ones not being controlled. If you don't, how do you prevent them from attempting to control and creating an escalating conflict?

RM: Well, I'm not sure that putting in switches makes my diagram equivalent to yours; mine still has the same output (error signal) path for each perception that is "switched in" while yours has a separate output (error) path for each control system and the switching just affects which output (error) is used to drive the output to the environment. But I'm glad you think the switches will work in my model. >

MT: Of course, I didn't show the many layers of control loops that produce the eventual muscular output, just as you have shown no output stages at all except m( ), which I presume to mean muscles. We both presume that the same muscles, and many of the stages before the muscles, are used to control the same mouse movement, no matter what is being controlled. So we appear to agree on what is happening, despite the diagram.

RM: I think we still differ on how many different output (error) signals drive the muscle output and, more importantly, where the "switching" occurs. In your model it occurs at the output (error) level; the switches determine which system's output (error) drives the muscle force output. In my model it occurs at the perceptual level; the switches determine which perceptual inputs is being controlled (which system is closed-loop with respect to the sensory input from the varying rectangular input, q.i). >

MT: What we don't agree on is the labelling. I argue that the output of the ECU that controls, say, p1 in your diagram is a disturbance to p2 and p3, not a consequence of their own ECU output, of which there is none if they are not controlling. You say the effect of controlling p1 on the value of uncontrolled p2 or p3 isn't a disturbance, because if they did happen next week to be controlling, the effect of their output would appear at the same place, qi. I don't get it.

RM: I'll try to explain this in the context of the "What is Size" demo. When you are controlling area (h*w), for example, your outputs (variations in the width, w, of the rectangle) are also causing variation in perimeter (h+w). So your outputs could be considered a disturbance to perimeter when you are controlling area. But I reserve the word "disturbance" for variations that affect a controlled variable and since perimeter is not controlled when area is being controlled I would not call the output- caused variations in perimeter that result when you are controlling area a "disturbance" to the perception of perimeter. But if you want to call them a disturbance that's fine with me.

MT: That's rather an irrelevant comment, when your objective is to argue that I was misleading Boris about the meaning of Quality of Control. I imagine that by now, Boris is rather confused. So I reiterate what I said initially, adding "open loop" as a clarification not intended to change the sense of it in any way: "For a single control unit, Quality of Control can be measured in a variety of ways, but they all come down to the same thing: the ratio between some measure of variation of the controlled perception and the variation under open loop conditions when the perception is not being controlled and only the disturbance influences it. Often the measure of variation used is the RMS variation of the perception. It could be the amplitude. It really doesn't matter in this context."

RM: Yes, that's fine. I would only insert the word "hypothetical" before "controlled variable".

MT: It's hypothetical for the outside observer, but the context of the question was of the use of QoC as an intrinsic variable that influences the rate of reorganization. In that context, the actual controlled variable is the determining factor. There's no experimenter hypothesising, just an organism controlling well or badly.

RM: In that case I would not even include anything about the disturbance or output (error or environmental output) in the measure of QoC. The system itself knows nothing about var(d), for example. So I would suggest that QoC of an intrinsic variable has to be measured in terms of intrinsic error, the only measure of QoC that the nervous (reorganizing) system itself could possibly have direct access to. So I would suggest something like a running average of intrinsic error as a measure of QoC of an intrinsic variable.
Best regards
Rick

···

>

Martin

--
Richard S. Marken PhD
<http://www.mindreadings.com>www.mindreadings.com

Hi Rick, I think Bill used the root mean squared error of all perceptual control systems as QoC in LCS III and didn’t use intrinsic variables at all.

Warren

···

[From Rick Marken (2014.07.19.0940)]

Martin Taylor (2014.07.17.15.10)–

MT: Why would you expect either to be higher then the other?

RM: I expect it based on one of Bill’s methods of identifying relative levels of perception, I would expect area to be a higher level perception than perimeter because you have to be able to perceive perimeter (width and height) to perceive area but you don’t need to perceive area in order to perceive perimeter. So I predict that perimeter is a lower level perception than area and that, therefore, you can control perimeter at a faster rate of variation than area. This prediction can actually be tested using the “What is size” demo. Can you see how?

MT:  The diagram you show enforces a conflict if

the pathway m( )->o->qi has fewer than three degrees of
freedom. It has only one df in the situation we are discussing. To
avoid the conflict, you say that only one of the perceptions is
being controlled. The diagram doesn’t show this.

RM: That’s true. The diagram leaves out the mechanism that shifts control from perceptions at one level versus another. This problem can be solved using switches again.

MT: If you do put switches into the paths either before or after the

comparators (or if you use a higher-level control to turn the output
gain to zero, which is the same thing) you have my diagram, with
zero contribution to qi from the ones not being controlled. If you
don’t, how do you prevent them from attempting to control and
creating an escalating conflict?

RM: Well, I’m not sure that putting in switches makes my diagram equivalent to yours; mine still has the same output (error signal) path for each perception that is “switched in” while yours has a separate output (error) path for each control system and the switching just affects which output (error) is used to drive the output to the environment. But I’m glad you think the switches will work in my model.

MT: Of course, I didn't show the many layers of control loops that

produce the eventual muscular output, just as you have shown no
output stages at all except m( ), which I presume to mean muscles.
We both presume that the same muscles, and many of the stages before
the muscles, are used to control the same mouse movement, no matter
what is being controlled. So we appear to agree on what is
happening, despite the diagram.

RM: I think we still differ on how many different output (error) signals drive the muscle output and, more importantly, where the “switching” occurs. In your model it occurs at the output (error) level; the switches determine which system’s output (error) drives the muscle force output. In my model it occurs at the perceptual level; the switches determine which perceptual inputs is being controlled (which system is closed-loop with respect to the sensory input from the varying rectangular input, q.i).

MT: What we don't agree on is the labelling. I argue that the output of

the ECU that controls, say, p1 in your diagram is a disturbance to
p2 and p3, not a consequence of their own ECU output, of which there
is none if they are not controlling. You say the effect of
controlling p1 on the value of uncontrolled p2 or p3 isn’t a
disturbance, because if they did happen next week to be controlling,
the effect of their output would appear at the same place, qi. I
don’t get it.

RM: I’ll try to explain this in the context of the “What is Size” demo. When you are controlling area (h*w), for example, your outputs (variations in the width, w, of the rectangle) are also causing variation in perimeter (h+w). So your outputs could be considered a disturbance to perimeter when you are controlling area. But I reserve the word “disturbance” for variations that affect a controlled variable and since perimeter is not controlled when area is being controlled I would not call the output- caused variations in perimeter that result when you are controlling area a “disturbance” to the perception of perimeter. But if you want to call them a disturbance that’s fine with me.

MT: It's hypothetical for the outside observer, but the context of the

question was of the use of QoC as an intrinsic variable that
influences the rate of reorganization. In that context, the actual
controlled variable is the determining factor. There’s no
experimenter hypothesising, just an organism controlling well or
badly.

RM: In that case I would not even include anything about the disturbance or output (error or environmental output) in the measure of QoC. The system itself knows nothing about var(d), for example. So I would suggest that QoC of an intrinsic variable has to be measured in terms of intrinsic error, the only measure of QoC that the nervous (reorganizing) system itself could possibly have direct access to. So I would suggest something like a running average of intrinsic error as a measure of QoC of an intrinsic variable.

Best regards

Rick

Martin


Richard S. Marken PhD
www.mindreadings.com

            RM:

It would be interesting to use reaction time methods
like those described in the Marken et al paper to see
which perception, area or perimeter, is higher in the
hierarchy! Once again, Martin, you have suggested a
nice, new little research study for me!

            MT: That's rather an irrelevant comment, when your

objective is to argue that I was misleading Boris about
the meaning of Quality of Control. I imagine that by
now, Boris is rather confused. So I reiterate what I
said initially, adding “open loop” as a clarification
not intended to change the sense of it in any way: “* For
a single control unit, Quality of Control can be
measured in a variety of ways, but they all come down
to the same thing: the ratio between some measure of
variation of the controlled perception and the
variation under open loop conditions when the
perception is not being controlled and only the
disturbance influences it. Often the measure of
variation used is the RMS variation of the perception.
It could be the amplitude. It really doesn’t matter in
this context*.”

          RM: Yes, that's fine. I would only insert the word

“hypothetical” before “controlled variable”.

[Martin Taylor 2014.07.19.14.17]

You don't need w+h in order to perceive w*h. To perceive either, you

need w and you need h.
To perceive area, you don’t actually need w or h, and the real
perception of area probably doesn’t do more than estimate how many
retinal pixels are activated after whatever is done to create size
constancy. The same could well be the way you perceive length. So
area could be at the same level as h and w, and therefore two levels
below h+w.
Snarky!
In other words, your system feeds the reference value for every
potentially controlled perception directly into the output
integrators of the relevant control units, leading not only to wild
conflict, but also to indefinitely increasing outputs from the
uncontrolled units into the muscle system. I’m not clear how this
leads to any great advantage over simply not passing output from the
uncontrolled units into the outflow processes that lead to the
muscles.
As you wish. But you are obviously aware that you are basing your
argument on your choosing to use a word differently from the way it
is used in the messages about which you are arguing. Is it worth doing that simply in order to make sure there is a
disagreement?
Thanks for that.
It seems to me reasonable to use the word “disturbance” either way
– to refer only to influences to controlled perception, or to
refer to effects that disturb a perception whether it is being
controlled or not. I don’t control a perception of the positions of
the leaves in the trees outside my window, but I do say that the
wind disturbs them.
This is a valid and important point, because it raises a question
about how the error is determined, and what scales it. The analyst
who looks at and measures the error in the open-loop and closed loop
conditions has no problem because the scaling is the same in both
cases, but internal to the organism, what is the measure of error?
It can’t simply be firing rate in some fibre carrying the error
signal, because those vary all over the lot in the resting
condition. There has to be some kind of internal reference standard,
and you are quite right that nothing internal senses the disturbance
alone. In e-mail, and possibly in print, Bill often talked about large and
increasing error, but if there’s no way to determine size relative
to a standard, possibly we may have to ignore the error magnitude
and consider only the proportional derivative of error, low-pass
filtered to allow the loop time to correct most of the error after a
step change in the disturbance.
This needs thought and further analysis. It would be interesting to
reprogram the Little Man or Arm-2 to use low-pass derivative error
and see whether it would converge at all.
Thanks for that.
Martin

···

[From Rick Marken (2014.07.19.0940)]

            Martin Taylor

(2014.07.17.15.10)–

                          RM:

It would be interesting to use reaction
time methods like those described in the
Marken et al paper to see which
perception, area or perimeter, is higher
in the hierarchy! Once again, Martin, you
have suggested a nice, new little research
study for me!

            MT: Why would you expect either to be higher then the

other?

          RM: I expect it based on one of Bill's methods of

identifying relative levels of perception, I would expect
area to be a higher level perception than perimeter
because you have to be able to perceive perimeter (width
and height) to perceive area but you don’t need to
perceive area in order to perceive perimeter.

          So I predict that perimeter is a lower level

perception than area and that, therefore, you can control
perimeter at a faster rate of variation than area. This
prediction can actually be tested using the “What is size”
demo. Can you see how?

            MT:  The diagram you

show enforces a conflict if the pathway m(
)->o->qi has fewer than three degrees of freedom.
It has only one df in the situation we are discussing.
To avoid the conflict, you say that only one of the
perceptions is being controlled. The diagram doesn’t
show this.

          RM: That's true. The diagram leaves out the mechanism

that shifts control from perceptions at one level versus
another. This problem can be solved using switches again.

            MT: If you do put switches into the paths either before

or after the comparators (or if you use a higher-level
control to turn the output gain to zero, which is the
same thing) you have my diagram, with zero contribution
to qi from the ones not being controlled. If you don’t,
how do you prevent them from attempting to control and
creating an escalating conflict?

          RM: Well, I'm not sure that putting in switches makes

my diagram equivalent to yours; mine still has the same
output (error signal) path for each perception that is
“switched in” while yours has a separate output (error)
path for each control system and the switching just
affects which output (error) is used to drive the output
to the environment. But I’m glad you think the switches
will work in my model.

            MT: Of course, I didn't show the many layers of control

loops that produce the eventual muscular output, just as
you have shown no output stages at all except m( ),
which I presume to mean muscles. We both presume that
the same muscles, and many of the stages before the
muscles, are used to control the same mouse movement, no
matter what is being controlled. So we appear to agree
on what is happening, despite the diagram.

          RM: I think we still differ on how many different

output (error) signals drive the muscle output and, more
importantly, where the “switching” occurs. In your model
it occurs at the output (error) level; the switches
determine which system’s output (error) drives the muscle
force output. In my model it occurs at the perceptual
level; the switches determine which perceptual inputs is
being controlled (which system is closed-loop with respect
to the sensory input from the varying rectangular input,
q.i).

            MT: What we don't agree on is the labelling. I argue

that the output of the ECU that controls, say, p1 in
your diagram is a disturbance to p2 and p3, not a
consequence of their own ECU output, of which there is
none if they are not controlling. You say the effect of
controlling p1 on the value of uncontrolled p2 or p3
isn’t a disturbance, because if they did happen next
week to be controlling, the effect of their output would
appear at the same place, qi. I don’t get it.

          RM: I'll try to explain this in the context of the

“What is Size” demo. When you are controlling area (h*w),
for example, your outputs (variations in the width, w, of
the rectangle) are also causing variation in perimeter
(h+w). So your outputs could be considered a disturbance
to perimeter when you are controlling area. But I reserve
the word “disturbance” for variations that affect a
controlled variable

          and since perimeter is not controlled when area is

being controlled I would not call the output- caused
variations in perimeter that result when you are
controlling area a “disturbance” to the perception of
perimeter. But if you want to call them a disturbance
that’s fine with me.

                          MT:

That’s rather an irrelevant comment, when
your objective is to argue that I was
misleading Boris about the meaning of
Quality of Control. I imagine that by now,
Boris is rather confused. So I reiterate
what I said initially, adding “open loop”
as a clarification not intended to change
the sense of it in any way: “* For a
single control unit, Quality of Control
can be measured in a variety of ways,
but they all come down to the same
thing: the ratio between some measure of
variation of the controlled perception
and the variation under open loop
conditions when the perception is not
being controlled and only the
disturbance influences it. Often the
measure of variation used is the RMS
variation of the perception. It could be
the amplitude. It really doesn’t matter
in this context*.”

                        RM: Yes, that's fine. I would only insert

the word “hypothetical” before “controlled
variable”.

            MT: It's hypothetical for the outside observer, but the

context of the question was of the use of QoC as an
intrinsic variable that influences the rate of
reorganization. In that context, the actual controlled
variable is the determining factor. There’s no
experimenter hypothesising, just an organism controlling
well or badly.

          RM: In that case I would not even include anything

about the disturbance or output (error or environmental
output) in the measure of QoC. The system itself knows
nothing about var(d), for example. So I would suggest that
QoC of an intrinsic variable has to be measured in terms
of intrinsic error, the only measure of QoC that the
nervous (reorganizing) system itself could possibly have
direct access to. So I would suggest something like a
running average of intrinsic error as a measure of QoC of
an intrinsic variable.

[From Rick Marken (2014.07.19.1415)]

···

On Sat, Jul 19, 2014 at 11:04 AM, Warren Mansell wmansell@gmail.com wrote:

WM: Hi Rick, I think Bill used the root mean squared error of all perceptual control systems as QoC in LCS III and didn’t use intrinsic variables at all.

RM: Yes, RMS error is the controlled variable in his model of reorganization described in Ch. 7 and referred to specifically on p. 120-121. So in this case QoC is a time average of the error signal in a control system. There are three control systems in the demo and the rate of change in RMS error in each system is the perception controlled by the reorganization system.

RM: I think this discussion got a bit confused because QoC can refer to 1) an objective measure of how well someone is controlling (such as the “stability factor” S = 1 - sqrt(Var.obs/Var.exp)) or 2) a measure of control that can be used in a model of reorganization (such as rate of changein RMS error). I assumed we were talking QoC meaning 1) since the measure that Martin suggested was basically the “stability factor”, S, which can’t be used as the basis of reorganization since the reorganization system can’t compute Var.exp.

RM: As far as intrinsic versus non-intrinsic variables, the model Bill describes is all about intrinsic variables; the rate of change in RMS error in the three systems described in that model are considered intrinsic perceptions. For example, on p. 124 ofLCS III Bill says: “In our example of reorganization here, the intrinsic variables are the rates of change of the RMS values of error signals in the three behavioral systems”. The variables controlled by the reorganization system – including the perceived error in the “behavioral control systems” – are called “intrinsic” variables. They are intrinsic in the sense that they are not controlled by the behavioral control systems but by reorganization of the parameters of those systems.

Best

Rick


Richard S. Marken PhD
www.mindreadings.com

[Martin Taylor 2014.07.19.17.48]

Yes, but the key words here are "In our example of reorganization".

In the preceding paragraphs he says: “I proposed (1960) that there
are certain internal variables, ‘intrinsic variables,’ monitored by
a set of reorganizing control systems, one control system per
variable, and that for each variable there is a built-in, inherited,
reference level …” … “It is this set of intrinsic variables and
their reference levels that provide the measure of error required if
e-coli reorganization is to work.” He then, in your quote, refers to
the example, saying that for the purpose of the example (a Skinner
pigeon training exercise) he will use QoC for each of three control
systems as three intrinsic variables.
This is the closest I have come to finding a reference for treating
QoC as an intrinsic variable, and here it is just a demonstrative
convenience, not a suggestion that it may actually be true for
living organisms. Indeed, he goes on in the next paragraph to
speculate about what might be the actual intrinsic variables for the
pigeon in the example.
I’m beginning to think that there is no published reference to a
place where Bill suggested that QoC might be a real live intrinsic
variable, though it is clear from his e-mail discussions that he
entertained the possibility.
Martin

···

[From Rick Marken (2014.07.19.1415)]

        On Sat, Jul 19, 2014 at 11:04 AM,

Warren Mansell wmansell@gmail.com wrote:

              WM: Hi Rick, I think Bill used the root mean

squared error of all perceptual control systems as QoC
in LCS III and didn’t use intrinsic variables at all.

          RM: As far as intrinsic versus non-intrinsic variables,

the model Bill describes is all about intrinsic variables;
the rate of change in RMS error in the three systems
described in that model are considered intrinsic
perceptions. For example, on p. 124 ofLCS III Bill says:
“In our example of reorganization here, the ** intrinsic
variables** are the rates of change of the RMS values
of error signals in the three behavioral systems”. The
variables controlled by the reorganization system –
including the perceived error in the “behavioral control
systems” – are called “intrinsic” variables. They are
intrinsic in the sense that they are not controlled by the
behavioral control systems but by reorganization of the
parameters of those systems.

Hi both, I agree that the RMS error across perceptual control systems doesn’t seem to be what Bill conceived of as an intrinsic variable. However thinking about it, if the key quality of an intrinsic variable is that it confers survival and reproductive advantages, then maybe optimising the control of all systems currently operating is one of those because it would entail greater control overall and therefore better skill at various activities that improve survival and reproduction, whatever they may be.

I want to send you a quote from Klaus Grawe who used PCT in an integrative model of psychopathology. He talks about a basic ‘need for control’. I need to check the book to see where that idea comes from.

All the best,

Warren

···

[From Rick Marken (2014.07.19.1415)]

        On Sat, Jul 19, 2014 at 11:04 AM,

Warren Mansell wmansell@gmail.com wrote:

              WM: Hi Rick, I think Bill used the root mean

squared error of all perceptual control systems as QoC
in LCS III and didn’t use intrinsic variables at all.

          RM: As far as intrinsic versus non-intrinsic variables,

the model Bill describes is all about intrinsic variables;
the rate of change in RMS error in the three systems
described in that model are considered intrinsic
perceptions. For example, on p. 124 ofLCS III Bill says:
“In our example of reorganization here, the ** intrinsic
variables** are the rates of change of the RMS values
of error signals in the three behavioral systems”. The
variables controlled by the reorganization system –
including the perceived error in the “behavioral control
systems” – are called “intrinsic” variables. They are
intrinsic in the sense that they are not controlled by the
behavioral control systems but by reorganization of the
parameters of those systems.

[Martin Taylor 2014.07.20.07.33]

Warren, you are going the same direction as I started, but there's

more to it than that. I asked for the reference to Bill having said
this or anything related to it so that I wouldn’t get criticised in
my chapter for appropriating one of his ideas. I’m sure he knew it,
but I don’t know if he said it publicly.
Beyond what you say, it is important to recognize that, because
intrinsic variables are by definition not accessible to the
perceptual hierarchy as perceptions to be controlled, all the
influences on them due to the actions of the hierarchy are
side-effects. Side-effects are consistent effects when the
mechanisms used for control are consistent and the environment is
consistent. Control, however, depends on the ability to use
different outputs to counter different values of disturbances. The
implication is that the side-effects of control will vary. If they
do, they will do so independently of the current states of any
intrinsic variables they influence. That presents a conundrum, since
it suggests that nothing reorganization could do would enable
control of the not-perceivable intrinsic variables.
The answer is probably that the higher-level perceptions are less
subject to the “any means to an end” problem, insofar as for any
particular state of reorganization they can rely on the more stable
inputs provided by the lower-level controllers. In other words,
though their environment is no stable, it is more stable, and
moreover, it is unstable in ways that have been determined by
reorganization. Accordingly, even in an environment in which the
low-level outputs vary all over the lot, they do so in ways that are
patterned by the requirements of the higher-level controls. Those
requirements patterns have been set by reorganization according to
how well the side-effects combine to serve to control the intrinsic
variables.
The implication of this is that although the values of the intrinsic
variables cannot be directly perceived, nevertheless they must
influence something about perception. One might get testy when
hungry, a perception related to stomach contraction, but ultimately
probably an effect of low blood sugar. If nothing is immediately
available to eat, one goes after it, changing the reference levels
for lots of actually perceivable things. None of this could work if
control were unstable, or if useless perceptions such as the product
of temperature and the distance to the nearest lamppost were being
controlled.
So we come to the conclusion that reorganization must not only use
QoC as a criterion, but also that the actual internal intrinsic
variables determine how reorganization influences what perceptions
should be controlled, an area Bill explicitly did not address. The
nearest he came was to point out that although we could perceive any
old function of sensory variables, we don’t. The selection of what
kinds of perceptual functions we have is relatively consistent
within a species, though different across species if only because of
the different sensory systems and life environments between, say, a
butterfly and a tree.
Yesterday I was actually trying to reword the reorganization section
of my chapter to describe these ideas more fluently, so your comment
is quite timely.
I’ll look forward to the Grawe quote.
Martin

···

On 2014/07/20 3:35 AM, Warren Mansell
wrote:

    Hi both, I agree that the RMS error across perceptual control

systems doesn’t seem to be what Bill conceived of as an
intrinsic variable. However thinking about it, if the key
quality of an intrinsic variable is that it confers survival and
reproductive advantages, then maybe optimising the control of
all systems currently operating is one of those because it would
entail greater control overall and therefore better skill at
various activities that improve survival and reproduction,
whatever they may be.

    I want to send you a quote from Klaus Grawe who used PCT in

an integrative model of psychopathology. He talks about a basic
‘need for control’. I need to check the book to see where that
idea comes from.

All the best,

Warren

[From Rick Marken (2014.07.20.0850)]

[Martin Taylor 2014.07.19.14.17

[From Rick Marken (2014.07.19.0940)]

Martin Taylor (2014.07.17.15.10)--

RM: It would be interesting to use reaction time methods like those described in the Marken et al paper to see which perception, area or perimeter, is higher in the hierarchy! Once again, Martin, you have suggested a nice, new little research study for me!

MT: Why would you expect either to be higher then the other?

RM: I expect it based on one of Bill's methods of identifying relative levels of perception, I would expect area to be a higher level perception than perimeter because you have to be able to perceive perimeter (width and height) to perceive area but you don't need to perceive area in order to perceive perimeter.

MT: You don't need w+h in order to perceive w*h. To perceive either, you need w and you need h.

MT: To perceive area, you don't actually need w or h, and the real perception of area probably doesn't do more than estimate how many retinal pixels are activated after whatever is done to create size constancy. The same could well be the way you perceive length. So area could be at the same level as h and w, and therefore two levels below h+w.

RM: Excellent proposal. So now we have three alternatives: perception of area is 1) at a higher level than perception of perimeter (mine), 2) the same level (yours) and 3) at a lower level (yours). Now let's test it.

RM: So I predict that perimeter is a lower level perception than area and that, therefore, you can control perimeter at a faster rate of variation than area. This prediction can actually be tested using the "What is size" demo. Can you see how?

MT: Snarky!

RM: I'm sorry. I meant it as an honest question. I certainly didn't mean to be snarky. I'm sorry it came off that way to you. Anyway, I believe the "What is Size" demo can provide at least a preliminary test the predictions about the relative levels of area and perimeter perception by simply comparing the stability value you get for each. Since the disturbance frequency is the same for both variables, the variable with the lower stability value (poorer control at that disturbance frequency) can be considered the one that is higher level. If the stability values are the same then the two variables can be considered to be at the same level. I've done this test and my stability value for area is consistently lower than it is for perimeter. The difference isn't as big as I would like it to be: the average stability for area is ~.9 while for perimeter it's ~.95. So my preliminary conclusion is that area is a higher level perception than perimeter. But that's just preliminary; it's still possible that they are actually at the same level and the slightly (but consistently) higher stability of perimeter control is an artifact the way control was tested (using a sine wave disturbance, for example, that makes it possible to get a high stability measure by just moving the mouse back and forth at the same frequency but 180 degrees out of phase with the disturbance; I'll have to develop a version using a narrow band noise disturbance).
But in the meantime, why don't you (and anyone else out there who is willing) try doing the "Control of Size" demo, controlling area a couple times and perimeter a couple times (in random order, of course) and let me know what the area and perimeter stability measures were on the trials when you controlled each. I may have been unconsciously biasing my results.

MT: In other words, your system feeds the reference value for every potentially controlled perception directly into the output integrators of the relevant control units, leading not only to wild conflict,

RM: I don't think so; but I will build a model of this once I get the tests set up. This will be my next research project.
Best
Rick

···

--
Richard S. Marken PhD
<http://www.mindreadings.com>www.mindreadings.com

Hi Martin, it sounds very interesting and I can see why there are reasons to expand on the workings of intrinsic variables in PCT because they were relatively broadly described by Bill, when in fact, for any one species of organism, and any one individual within that species, they will be quite specific. I can also sense the relevance of a perceivable feature that is associated with intrinsic variables deviating from their reference values and I think that Bill did consider the role of arousal and emotion in this context unless I am mistaken, in B:CP and beyond?
I am really looking forward to seeing the logical steps and implications spelled out in the chapter!
Warren

···

On 20 Jul 2014, at 13:06, Martin Taylor <mmt-csg@mmtaylor.net> wrote:

[Martin Taylor 2014.07.20.07.33]

On 2014/07/20 3:35 AM, Warren Mansell wrote:
Hi both, I agree that the RMS error across perceptual control systems doesn't seem to be what Bill conceived of as an intrinsic variable. However thinking about it, if the key quality of an intrinsic variable is that it confers survival and reproductive advantages, then maybe optimising the control of all systems currently operating is one of those because it would entail greater control overall and therefore better skill at various activities that improve survival and reproduction, whatever they may be.

Warren, you are going the same direction as I started, but there's more to it than that. I asked for the reference to Bill having said this or anything related to it so that I wouldn't get criticised in my chapter for appropriating one of his ideas. I'm sure he knew it, but I don't know if he said it publicly.

Beyond what you say, it is important to recognize that, because intrinsic variables are by definition not accessible to the perceptual hierarchy as perceptions to be controlled, all the influences on them due to the actions of the hierarchy are side-effects. Side-effects are consistent effects when the mechanisms used for control are consistent and the environment is consistent. Control, however, depends on the ability to use different outputs to counter different values of disturbances. The implication is that the side-effects of control will vary. If they do, they will do so independently of the current states of any intrinsic variables they influence. That presents a conundrum, since it suggests that nothing reorganization could do would enable control of the not-perceivable intrinsic variables.

The answer is probably that the higher-level perceptions are less subject to the "any means to an end" problem, insofar as for any particular state of reorganization they can rely on the more stable inputs provided by the lower-level controllers. In other words, though their environment is no stable, it is more stable, and moreover, it is unstable in ways that have been determined by reorganization. Accordingly, even in an environment in which the low-level outputs vary all over the lot, they do so in ways that are patterned by the requirements of the higher-level controls. Those requirements patterns have been set by reorganization according to how well the side-effects combine to serve to control the intrinsic variables.

The implication of this is that although the values of the intrinsic variables cannot be directly perceived, nevertheless they must influence something about perception. One might get testy when hungry, a perception related to stomach contraction, but ultimately probably an effect of low blood sugar. If nothing is immediately available to eat, one goes after it, changing the reference levels for lots of actually perceivable things. None of this could work if control were unstable, or if useless perceptions such as the product of temperature and the distance to the nearest lamppost were being controlled.

So we come to the conclusion that reorganization must not only use QoC as a criterion, but also that the actual internal intrinsic variables determine how reorganization influences what perceptions should be controlled, an area Bill explicitly did not address. The nearest he came was to point out that although we could perceive any old function of sensory variables, we don't. The selection of what kinds of perceptual functions we have is relatively consistent within a species, though different across species if only because of the different sensory systems and life environments between, say, a butterfly and a tree.

Yesterday I was actually trying to reword the reorganization section of my chapter to describe these ideas more fluently, so your comment is quite timely.

I want to send you a quote from Klaus Grawe who used PCT in an integrative model of psychopathology. He talks about a basic 'need for control'. I need to check the book to see where that idea comes from.
All the best,
Warren

I'll look forward to the Grawe quote.

Martin

[From Rick Marken (2014.07.20.0945)]

···

Martin Taylor (2014.07.19.17.48)–

MT: Yes, but the key words here are “In our example of reorganization”…

MT: This is the closest I have come to finding a reference for treating

QoC as an intrinsic variable…

MT: I'm beginning to think that there is no published reference to a

place where Bill suggested that QoC might be a real live intrinsic
variable, though it is clear from his e-mail discussions that he
entertained the possibility.

RM: I suspect that the best place to look to see whether or not QoC is a “real live intrinsic variable” is in real live organisms, not in Bill’s writings.

Best

          RM:  The

variables controlled by the reorganization system –
including the perceived error in the “behavioral control
systems” – are called “intrinsic” variables. They are
intrinsic in the sense that they are not controlled by the
behavioral control systems but by reorganization of the
parameters of those systems.

Rick


Richard S. Marken PhD
www.mindreadings.com

[Martin Taylor 2014.07.20.14.18]

That's true of everything Bill said, but when one is writing for

publication and one wants to pursue an idea, one wants to give
credit for the idea where credit is due. Or at least I do,
especially in a Festschrift for the person who probably did have the
idea.
Martin

···

[From Rick Marken (2014.07.20.0945)]

            Martin Taylor

(2014.07.19.17.48)–

            MT: I'm beginning to think that there is no published

reference to a place where Bill suggested that QoC might
be a real live intrinsic variable, though it is clear
from his e-mail discussions that he entertained the
possibility.

          RM: I suspect that the best place to look to see

whether or not QoC is a “real live intrinsic variable” is
in real live organisms, not in Bill’s writings.

[From Erling Jorgensen (2014.07.31 1130EDT)]

[Martin Taylor 2014.07.20.14.18]

[From Rick Marken (2014.07.20.0945)]

Martin Taylor (2014.07.19.17.48)--

MT: I'm beginning to think that there is no published reference to a place

where Bill suggested that QoC might be a real live intrinsic variable, though
it is clear from his e-mail discussions that he entertained the possibility.

RM: I suspect that the best place to look to see whether or not QoC is a

"real live intrinsic variable" is in real live organisms, not in Bill's
writings.

MT: That's true of everything Bill said, but when one is writing for

publication and one wants to pursue an idea, one wants to give credit for the
idea where credit is due. Or at least I do, especially in a Festschrift for
the person who probably did have the idea.

EJ: Hi Martin,
I'm not sure if you ever found the citation you were looking for. But I think
one possibility that I just noticed is in the second Powers, Clark, &
McFarland paper back in 1960, "A General Feedback Theory of Human Behavior:
Part II," from _Perceptual and Motor Skills_, 1960, 11, 309-323.

On page 316, in the section headed "N-System," they raise as a possible
intrinsic reference signal that of "mean error rate in the hierarchy." That
may be close to what you were seeking. Check out that paragraph, to see if it
fits your needs.

The link to the article, that Warren just sent me, is:
<http://pctweb.org/Powers%20et%20al%201960%20part2.pdf&gt;

All the best,
Erling

Thanks Erling, can’t believe I didn’t see that!
Warren

···

On Thu, Jul 31, 2014 at 4:47 PM, Erling Jorgensen ejorgensen@riverbendcmhc.org wrote:

[From Erling Jorgensen (2014.07.31 1130EDT)]

[Martin Taylor 2014.07.20.14.18]

[From Rick Marken (2014.07.20.0945)]

Martin Taylor (2014.07.19.17.48)–

MT: I’m beginning to think that there is no published reference to a place

where Bill suggested that QoC might be a real live intrinsic variable, though

it is clear from his e-mail discussions that he entertained the possibility.

RM: I suspect that the best place to look to see whether or not QoC is a

“real live intrinsic variable” is in real live organisms, not in Bill’s

writings.

MT: That’s true of everything Bill said, but when one is writing for

publication and one wants to pursue an idea, one wants to give credit for the

idea where credit is due. Or at least I do, especially in a Festschrift for

the person who probably did have the idea.

EJ: Hi Martin,

I’m not sure if you ever found the citation you were looking for. But I think

one possibility that I just noticed is in the second Powers, Clark, &

McFarland paper back in 1960, "A General Feedback Theory of Human Behavior:

Part II," from Perceptual and Motor Skills, 1960, 11, 309-323.

On page 316, in the section headed “N-System,” they raise as a possible

intrinsic reference signal that of “mean error rate in the hierarchy.” That

may be close to what you were seeking. Check out that paragraph, to see if it

fits your needs.

The link to the article, that Warren just sent me, is:

<http://pctweb.org/Powers%20et%20al%201960%20part2.pdf>

All the best,

Erling


Dr Warren Mansell
Reader in Psychology
Cognitive Behavioural Therapist & Chartered Clinical Psychologist
School of Psychological Sciences

Coupland I
University of Manchester
Oxford Road
Manchester M13 9PL
Email: warren.mansell@manchester.ac.uk

Tel: +44 (0) 161 275 8589

Website: http://www.psych-sci.manchester.ac.uk/staff/131406

See teamstrial.net for further information on our trial of CBT for Bipolar Disorders in NW England

The highly acclaimed therapy manual on A Transdiagnostic Approach to CBT using Method of Levels is available now.

Check www.pctweb.org for further information on Perceptual Control Theory

[Martin Taylor 2014.07.15.16.39]

No, I didn't miss the point at all. I used that fact in my message.

In fact, it was the central point of my message: If you are
controlling the area with the mouse, you can’t be controlling the
perimeter, and vice-versa. If you try to do both, you will be in a
conflict situation, since there is only one degree of freedom in the
environmental feedback path.
Here’s a picture of the situation. The dashed line means the two
switches are connected so that if the subject chooses to control one
perception, the other is not controlled. In the picture, area is
being controlled. A side-effect of that control is a disturbance to
the perception of perimeter, which combines with the externally
applied disturbance to the height to form the overall disturbance to
the perimeter.
If these two control systems were in different people, we would have
a classical escalating conflict if both were trying to control their
perceptual variable at the same time.
Almost right. Note the word “either”. But what is missing are the
words “to the external environment” as in “”. Notice that at any moment one or
other of those two systems does not influence the muscle-level
systems. That one produces NO output in its own control loop. Unless
the user tries (necessarily without success) to control both a and p
simultaneously, only one of them influences the muscular reference
values.
Of course, if you mean output to the external environment,
specifically to the “width” variable. Why put that in bold, as
though there were some question about it that you are trying to
correct?
Of course. That’s what the demo is about, isn’t it? At least, so I
assumed when I wrote the message with which you are trying to
disagree.
You confuse two things: (1) the output of a control unit, and (2)
the combined output to the world of all the control units that are
currently exercising control by acting in the external environment.
Your var(o) is the latter, and has no relevance to the former, which
is what we are discussing.
Yes. And many perceptual systems use some of these aspects in
creating their perceptual values. If those aspects are being used as
part of the environmental path by other control systems, the cross
influences are side-effects creating disturbances, as control of a
does for the perception of p.
All true. It’s a nice demo.
Hopefully you can now see that your criticism is incorrect. As well
as probably being confusing to readers who might be less able than
you and I to see the difference between the output of a control
system at one level and the eventual effect on the world of the
outputs of all the control systems currently active.
Martin

(Attachment HeightWidth.jpg is missing)

···
        [From

Rick Marken (2014.07.15.1050)]

              Martin Taylor

(2014.07.14.14.40)–

              MT:

True, but since in the absence of control, var(o) = 0,
cv = d.

              RM: I think you must know that this is not the case.

But if not, you can demonstrate to yourself that it’s
not true using, once again, the “Control of Size” demo
(http://www.mindreadings.com/ControlDemo/Size.html).

              MT: As usual, we are talking at cross-purposes. I am

comparing open-loop with closed-loop conditions,
whereas you are introducing an added disturbance that
is a side-effect of controlling something else. It’s
not surprising we come to different conclusions.

              MT:

I think I didn’t word that very well. Let’s try again.

              MT:

A side-effect is an effect on something that is not in
the control loop of the perception being
controlled…when in Rick’s demo you control the area,
a side-effect is that the perception of perimeter
length may be influenced.

            RM: I think you miss the point of the demo, which is

that there are two perceptual aspects of the same
physical situation that can be controlled. The physical
situation is the rectangle on the screen which has
height (h) and width (w). The computer varies h and
mouse movement varies w. The two aspects of this
situation that can be controlled are area (a) and
perimeter (p):

            So a and p are two different aspects of the _same_

physical variables, w and h. In order to control either
a or p the subject must move the mouse to vary w in
order to appropriately compensate for changes in the
computer generated variations in h. So variations in w
are the outputs (o) of the system controlling either a
or p;

  •  So variations in w are
    

the outputs (o) to the external environment of the system
controlling either a or p*

          Which perceptual aspect of the

display the subject controls is selected by the subject.
But regardless of which aspect of the display the subject
chooses to control, a or p, there output will be varied as
the means of controlling it. ** So o will be varying
whether the subject is controlling a or p;**

          ...regardless of whether a or p is

being controlled, both a and p will be varying, although a
will be varying much less than p if a is under control
and p will be varying less than a if p is controlled.

            The demonstration shows that you can determine which

perceptual aspect of the same physical situation is
being controlled by the subject be computing the
stability factor for both possible controlled
perceptions, a and p.

          So it's not true that var(o) is

zero when a perception is not under control.

MT
2014.07.14.14.40: Since, by the definition of the problem, the
perception of perimeter is not being controlled, the output of
the perimeter control system is zero, and any changes in the
perimeter perception are due to the joint influences of the
disturbance to, and any side-effects of, the area control
system.

          Outputs are always affecting many

different perceptual aspects of the world simultaneously
(as in the demo where variations in w are affecting both a
and p at the same time).

          The goal of testing for controlled

perceptions is to determine which perceptual aspects of a
situation are being controlled by those outputs. The
“Control of Size” demo is another approach to showing what
is meant by control of perception: that we control various
aspects of physical reality and that we can control
different aspects of the same physical reality. Because
of the latter fact, it is often difficult to determine
what an organism is doing (what perceptual aspect of its
environment it is controlling).

              MT:

Since, by the definition of the problem, the
perception of perimeter is not being controlled, the
output of the perimeter control system is zero, and
any changes in the perimeter perception are due to the
joint influences of the disturbance to, and any
side-effects of, the area control system.

            RM: Hopefully you can see now that this statement is

incorrect.

(Attachment HeightWidth1.jpg is missing)

image17.png

···

[From Rick Marken (2014.07.15.2000)]

Martin Taylor (2014.07.15.16.39)–

MT: Here's a picture of the situation. The dashed line means the two

switches are connected so that if the subject chooses to control one
perception, the other is not controlled. In the picture, area is
being controlled. A side-effect of that control is a disturbance to
the perception of perimeter, which combines with the externally
applied disturbance to the height to form the overall disturbance to
the perimeter.

Nice picture but I would prefer to put the switch on the input side, like this (pictured with the perceptual switch in “control of perimeter” position):

RM: This seems more consistent with the actual situation; either perception – area or perimeter – depending on which perceptual function is switched in, is controlled using the same output.

MT: Almost right. Note the word "either". But what is missing are the

words “to the external environment” as in "* So variations in w are
the outputs (o) to the external environment of the system
controlling either a or p* ". Notice that at any moment one or
other of those two systems does not influence the muscle-level
systems.

RM: That’s just because of the way you diagrammed it. My diagram, which controls either perception also controls them using the same output. In my diagram o corresponds to mouse movement, which has a proportional effect on w. So either o or w can be considered the output that affect the controlled variable, area or perimeter.

MT: That one produces NO output in its own control loop.

RM: Yes, again only because of the way you modeled it. I prefer my model because it seems somewhat more consistent with the neurophysiology. It seems highly unlikely that two separate control systems exist to carry out this task, one to control area and another to control perimeter, each with separate efferent connections to the muscles. But I can imagine that there are different perceptual functions that can be used by the a control system that uses the one connection to the muscles that move the mouse to vary w in order to control area or perimeter, depending on which perceptual function is switched in.

            RM: So variations in w

are the outputs (o) of the system controlling either a
or p;

RM: ** So o will be varying
whether the subject is controlling a or p;**

MT: Of course, if you mean output to the external environment,

specifically to the “width” variable. Why put that in bold, as
though there were some question about it that you are trying to
correct?

MT: I put it in bold because I assumed that when you said var(o) = 0 when there was no control you were talking about what you call the “output to the external environment” because that is the output that is measured when we calculate var (o) for the stability factor. The output you are talking about is a theoretical variable that can’t be (or at least isn’t) measured.

MT: You confuse two things: (1) the output of a control unit, and (2)

the combined output to the world of all the control units that are
currently exercising control by acting in the external environment.
Your var(o) is the latter, and has no relevance to the former, which
is what we are discussing.

RM: It’s impossible that we were discussing you meaning (1) of var(o); the output of a control unit (per meaning 1) cannot be measured and, thus, cannot be used in the calculation of the stability factor.

MT: Hopefully you can now see that your criticism is incorrect. As well

as probably being confusing to readers who might be less able than
you and I to see the difference between the output of a control
system at one level and the eventual effect on the world of the
outputs of all the control systems currently active.

RM: Well I do agree that this is probably confusing to readers, most of whom probably couldn’t care less. But this all started because you were explaining to someone how to measure quality of control (QoC) using the stability factor and you described it incorrectly. You said:

MT: For a single control unit, Quality of Control can be measured in a variety of ways, but they all come down to the same thing: the ratio between some measure of variation of the controlled perception and the variation that would be observed if only the disturbance influenced the perception.

RM: I just pointed out that that is not quite correct; the measure of QoC in terms of stability is the ratio of observed to expected variation in the cv. The observed variation is just the variation in the variable that is suspected to be the CV; the expected variation in the cv is the variation that is expected if there were no control. In order to determine what the expected variation in the cv would be sans control you have to determine how all influences on the controlled variable would have affected it. There are two main influences on any possible controlled variable: the effect of external disturbances and the effect of the system’s own outputs. So QoC – the ratio of observed to expected variance of a hypothetical controlled variable – is var(cv)/[var(d)+var(o)], not var(cv)/var(d), as implied by your last statement. But truth be told, it really doesn’t make much difference. If you use var(cv)/var(d) rather than var(cv)/[var(d)+var(o)] to measure QoC you are equally able to discriminate control of of area from control of perimeter.

So never mind.

Best

Rick


Richard S. Marken PhD
www.mindreadings.com

          RM: So it's not true that var(o) is

zero when a perception is not under control.