Controlled Variables (was Re: Levels of perception)

[From Rick Marken (2009.03.15.1010)]

Re: Martin Taylor (2009.03.14.16.33), Martin Taylor (2009.03.14.17.04) –

In both of these posts, Martin, you suggest that testing for controlled variables is not a particularly important aspect of control theory research. For example, in [Martin Taylor (2009.03.14.16.33)] you say

I have never had any interest in “the
test for the controlled variable” because that was never the question
at issue for me (and still has not been).

The main point of my “Revolution” paper (and several other papers that preceded it) could be stated as follows: You can’t be interested in studying control (the purposeful behavior of closed-loop systems) and not be interested in figuring out what perceptual variables they control (testing for controlled variables). In my “Revolution” paper I refer to people who say they are studying control but don’t test for controlled variables (or just assume they know what people are controlling) as “co-opters” of the control revolution. It is only through the rejection of the central role that testing for controlled variables plays in research on control that one could conclude, as you do (in that same post) that:

I do, however, continue to believe that a lot of “conventional”
research has provided data that is and continues to be useful in a PCT
framework.

I’ve attached a Figure I made for the “Revolution” paper. I eventually removed it in my efforts to play down the “behavioral illusion”; but I like it because I think it shows the position a researcher is in when observing the IV-DV relationships observed in conventional experiments. The researcher in the Figure is implicitly aware of the fact that the observed IV-DV relationship could reflect the open-loop input-output (I-O) characteristics of the system, if the system is open loop (as in the upper part of the Figure) or it could reflect the inverse of the feeback connection between O and I, as in the lower part. In order to properly analyze the situation you have to know whether the system is closed loop and, if so, what variables it is controlling that are relevant to the observed IV-DV relationship. The test for the controlled variable answers both questions. This was Bill Powers’ point in Bill Powers (2009.03.11.0745):

I think that the first step in any such investigation [such as Schouten’s] is to establish
beyond a resonable doubt what it is that you’re investigating. Exactly
what are the control loops involved in the Schouten experiment? The
evidence on that score is pretty sparse, since no attempt was (or could
have been) made to identify the controlled variable, the means of sensing
it, and the means of affecting it. Before you can measure the properties
of a control loop, or convince anyone that that’s what you’re measuring,
you must identify the control loop with whatever tests are appropriate…

In Martin Taylor (2009.03.14.17.04) you again suggest that testing for controlled variables is not very important when you say:

To Rick: As for what the subjects were controlling, I asked them to say
whether a straight line between the dots would go through the disk or
bypass it. I assume they were controlling for telling me what they saw,
but beyond that, I can’t guess. But I don’t think that makes the data
any the less reliable.

It’s not a question of reliability; it’s a question of validity. That’s what my attached Figure shows. The validity of the data (observed IV-DV relationships) obtained in a conventional experiment refers to whether it reflects what conventional psychologists assume it means (causal I-O relationships, as in the top part of the Figure) or what a control theorist thinks it means (inverse of the O-I feedback relationship, as in the bottom part). You can only evaluate the validity of the results (the correct system model, open or closed-loop) by doing some version of the test for the controlled variable.

I have found that the terms “controlled variables” and “test for controlled variables” work on social science researchers in the same way that mirrors work on vampires. Just as vampires recoil when a mirror is placed before them, conventional social science researchers recoil when the test for the controlled variable is placed before them. That’s why Bill’s suggestion that we call testing for controlled variables something else may be a good one – at least, if you want to keep conventional social scientists from recoiling from control research.

Best

Rick

···


Richard S. Marken PhD
rsmarken@gmail.com

[From Fred Nickols (03.15.2009.1036 MST)]
  

From: Richard Marken <rsmarken@GMAIL.COM>

I have found that the terms "controlled variables" and "test for controlled
variables" work on social science researchers in the same way that mirrors
work on vampires. Just as vampires recoil when a mirror is placed before
them, conventional social science researchers recoil when the test for the
controlled variable is placed before them. That's why Bill's suggestion that
we call testing for controlled variables something else may be a good one --
at least, if you want to keep conventional social scientists from recoiling
from control research.

Well, I'm no researcher of any kind, but I can see how the prospect of testing for a controlled variable could be a little overwhelming or intimidating. As I understand it,there are several levels of the hierarchy, at least some of which if not many of which come into play in attempting to consciously, deliberately control any particular variable. Further, if there are several variables being controlled at any point in time, sorting all that out and making some hypotheses about what the person is controlling for and then testing those hypotheses would seem to me to be a daunting task.

Might it be possible for dyed-in-the-wool PCTers such as yourself, Rick, to take some fairly well-known experiments conducted by others and retrofit a scheme for testing for the controlled variable in those experiments, then re-running the experiments for the purpose of determining what the test reveals?

In short, could some "models" be provided?

···

--
Regards,

Fred Nickols
Managing Partner
Distance Consulting, LLC
nickols@att.net
www.nickols.us

"Assistance at A Distance"

[From Rick Marken (2009.03.15.1140)]

Fred Nickols (03.15.2009.1036 MST)–

Well, I’m no researcher of any kind, but I can see how the prospect of testing for a controlled variable could be a little overwhelming or intimidating.

Me too. That’s why it I would love to have some smart people helping me out.

As I understand it,there are several levels of the hierarchy, at least some of which if not many of which come into play in attempting to consciously, deliberately control any particular variable. Further, if there are several variables being controlled at any point in time, sorting all that out and making some hypotheses about what the person is controlling for and then testing those hypotheses would seem to me to be a daunting task.

Actually, those things are less daunting then they may seem; see how effectively the test for the controlled variable can be done, even by a computer, in my “Test for the Controlled Variable” demo at http://www.mindreadings.com/ControlDemo/Mindread.html. And even if it it very daunting to do research aimed at testing for controlled variables, that seems to me to be no reason to avoid doing it since it is really the only way to study and understand living control systems.

Might it be possible for dyed-in-the-wool PCTers such as yourself, Rick, to take some fairly well-known experiments conducted by others and retrofit a scheme for testing for the controlled variable in those experiments, then re-running the experiments for the purpose of determining what the test reveals?

I think that’s the wrong way to go. That’s basically the approach that Martin advocates and I think it is a red herring (diverting attention from the real goal of control based science, which is understanding what perceptual variables organisms control and how they control them).

In short, could some “models” be provided?

I’ve tried but no one seems to have been motivated to follow up on any of them. One model is the baseball research. My paper in JEP:HEPP ( Marken, R. S. (2005) Optical
Trajectories and the Informational Basis of Fly Ball Catching, Journal of Experimental Psychology: Human
Perception & Performance,
31 (3), 630 – 634) describes some things that might be done to test for the controlled variable when catching a ball. My little “Mind reading” demo (referred to above) is another example. I also tried to describe some research examples in the “Revolution” paper. Gary Cziko has provided a number of examples in his books (Gary is the source of the beaver – and the bike stopping – examples in the “Revolution” paper). The “Hierarchy of Perception” demo at my site is another example of the kind of research that might be done to study controlled variables.

I think what it takes to do this research, however, is just the courage to do it. What you have to do (as a researcher) is really turn away from conventional research – not only because it can yield invalid results but, more importantly, I think, it because it asks the wrong questions in the first place (at least from a PCT perspective). Once you have been convinced by the many existing demonstrations control that organisms are, indeed, input control systems (which is what I was convinced of back in 1979 – this is my 30 year PCT anniversary, I guess; yikes!) then I think you just have to venture off (if you are a researcher, as I was) into new research territory. I have done the best I can on my own in this journey, given my limitations (in terms of intelligence and talent). Perhaps that’s why I am so gung ho about trying to get people to join me in this work. Imagine what could be accomplished if really bright people started doing this stuff. Then I think control theory would really become a force in scientific psychology.

Best

Rick

···


Richard S. Marken PhD
rsmarken@gmail.com

[Martin Taylor 2009.03.15.13.22]

[From Rick Marken (2009.03.15.1010)]

Re: Martin Taylor (2009.03.14.16.33), Martin Taylor (2009.03.14.17.04)

In both of these posts, Martin, you suggest that testing for controlled
variables is not a particularly important aspect of control theory
research. For example, in [Martin Taylor (2009.03.14.16.33)] you say

I have never had any interest in “the
test for the controlled variable” because that was never the question
at issue for me (and still has not been).

Do you see the difference between what you said above and what you
quoted from me? I think it’s fairly critical.

The main point of my “Revolution” paper (and several other papers that
preceded it) could be stated as follows: You can’t be interested in
studying control (the purposeful behavior of closed-loop systems) and
not be interested in figuring out what perceptual variables they
control (testing for controlled variables).

That’s a theological statement, not a statement of fact. From my
standpoint, if you can reasonably assume that when you ask someone to
push a button when they see a light come on, they will control for
doing exactly that. No test for the controlled variable is necessary.
And we all agreed, at least up to last week we all agreed, that this
kind of experiment is indeed looking at an open-loop capability of the
observer. If you truly disagree with what you agreed earlier (I guess
because Bill persuaded you, not because I did), please make a
considered criticism of [Martin Taylor 2009.02.17.11.23].

In my “Revolution” paper I refer to people who say they are
studying control but don’t test for controlled variables (or just
assume they know what people are controlling) as “co-opters” of the
control revolution.

I’ve re-read your paper, and I see that you used the term “co-opters”,
but not in this way.

It is only through the rejection of the central role that
testing for controlled variables plays in research on control that one
could conclude, as you do (in that same post) that:

I do, however, continue to believe that a lot of
“conventional”
research has provided data that is and continues to be useful in a PCT
framework.

I can answer this two ways:

(1) There is a lot more to studying control that testing for the
controlled variable.

(2) When there is very little question as to whether a person is
controlling a variable that she is asked to control at a reference
level that is requested, then there seems no point in testing whether
such is the case, nor for testing all of the myriad of other variables
she is controlling at the same time.

You seem obsessed with the idea that a person is always controlling
only one single variable, that this variable is the only thing the
person can perceive, and that the only thing that could possibly be
interesting about the person is what that variable might be. Even if
such a blinkered view of a person’s perceptual and controlling ability
were correct, I think it would still be interesting to figure out
things like the gain of the control loop, transport lags, the form of
the perceptual function, the limitations of the environmental feedback
function, the side-effects of control actions, and the energetic
efficiency of control. You would prohibit all such investigations, or
at least relegate them to the “co-opters” who should be consigned to
the nether regions.

I’ve attached a Figure I made for the “Revolution” paper. I eventually
removed it in my efforts to play down the “behavioral illusion”; but I
like it because I think it shows the position a researcher is in when
observing the IV-DV relationships observed in conventional experiments.
The researcher in the Figure is implicitly aware of the fact that the
observed IV-DV relationship could reflect the open-loop input-output
(I-O) characteristics of the system, if the system is open loop (as in
the upper part of the Figure) or it could reflect the inverse of the
feeback connection between O and I, as in the lower part. In order to
properly analyze the situation you have to know whether the system is
closed loop and, if so, what variables it is controlling that are
relevant to the observed IV-DV relationship.

Yes. In respect of Schouten, we spent weeks making that clear, until
finally Bill said [From Bill Powers (2009.02.20.1946 MST)] in response
to one of your messages:

--------quote----------

Yes, Martin emphasized that the action does not affect the
presentation. That loop is not closed, right? It could be closed if the
action did affect the presentation, which isn’t impossible, but it
doesn’t, so that potential loop is not closed – which means it’s open,
as we use the term. There IS an open-loop relationship between S and R,
isn’t there? Or are you saying that you have to have a closed-loop
relationship and open it before you can call it open-loop? Your rules
for applying these terms are not at all clear.

--------end quote-------

In Martin Taylor (2009.03.14.17.04) you again suggest that testing for
controlled variables is not very important when you say:

To Rick: As for what the subjects were
controlling, I asked them to say
whether a straight line between the dots would go through the disk or
bypass it. I assume they were controlling for telling me what they saw,
but beyond that, I can’t guess. But I don’t think that makes the data
any the less reliable.

It’s not a question of reliability; it’s a question of validity. That’s
what my attached Figure shows. The validity of the data (observed IV-DV
relationships) obtained in a conventional experiment refers to whether
it reflects what conventional psychologists assume it means (causal I-O
relationships, as in the top part of the Figure) or what a control
theorist thinks it means (inverse of the O-I feedback relationship, as
in the bottom part). You can only evaluate the validity of the results
(the correct system model, open or closed-loop) by doing some version
of the test for the controlled variable.

OK. You can make your point if you can show how the subject’s response
“bypasses the circle” influences the subject’s perception of whether
the straight line between the dots does bypass the circle. Otherwise,
we have the same situation as in Bill’s quote: “Yes, Martin
emphasized that the action does not affect the presentation. That loop
is not closed, right? It could be closed if the action did affect the
presentation, which isn’t impossible, but it doesn’t, so that potential
loop is not closed – which means it’s open, as we use the term.

If I had used a tracking study in which the subject moved the circular
disk until the apparent straight line between the dots was just tangent
to it, then the experiment would be closed loop, but I still would not
know the nature of the controlled variable. All I would know is what
environmental configuration gives zero error for that variable – how
far the circle overlaps the physically straight line between the dots
in order for the perceptually straight line to be tangent to the dots.
But that’s exactly what I sought and found in the open-loop situation.
I would know no more by using a control experiment.

You may say that in the control situation you would perform the Test
for the Controlled Variable by influencing different components of the
function, as you do in your area-perimeter demonstration. To do that,
you must have some hypothesis about what those components might be. I
did have such a hypothesis at the time, created before doing the
experiment (one hypothesis, not several, but it made predictions, the
same predictions as it would make in a closed-loop tracking study). I
predicted three different things, which I mentioned in the message to
which you are responding. In both open- and closed-loop cases, the
experiment would tell me just as much about the nature of the
perceptual variable. The difference would be that in the one case it
would be controlled, and in the other it would not.

I find myself frequently using the term “theological” to refer to your
comments. This hasn’t changed in a decade and a half. Before we first
met at CSG-93, I called you a “loose canon”, because you had signed
yourself that way in one of your messages. Might you perhaps influence
a variable you seem to want to control (persuading people to learn and
use PCT) if you were to act less like Torquemada and more like John
Paul I in proseletizing your religion? Or maybe that isn’t truly a
variable you want to control?

I have found that the terms “controlled variables” and “test for
controlled variables” work on social science researchers in the same
way that mirrors work on vampires. Just as vampires recoil when a
mirror is placed before them, conventional social science researchers
recoil when the test for the controlled variable is placed before them.
That’s why Bill’s suggestion that we call testing for controlled
variables something else may be a good one – at least, if you want to
keep conventional social scientists from recoiling from control
research.

I was going to put my following comment before this quote, but it would
need repeating, so I leave it here: PCT theory suggests that when one
mode of action fails to influence a variable you want to control, or
worse, influences it so as to increase error, reorganization – trying
different modes of action – might be useful. (The theory says it will
occur, but you provide a counter-example, if you are actually
controlling for more people to understand and use PCT.)

Martin

[From Rick Marken (2009.03.15.2050)]

Martin Taylor (2009.03.15.13.22)

from my
standpoint, if you can reasonably assume that when you ask someone to
push a button when they see a light come on, they will control for
doing exactly that. No test for the controlled variable is necessary.

You don’t test for controlled variables to figure out whether or not people are following instructions in a conventional experiment. You test for controlled variables in order to understand what people are controlling when they are carrying out particular behaviors of interest, which could include the behavior in psychological experiments that were done with other goals in mind. But if you don’t know what people are controlling then you can’t really study anything else about their behavior.

And we all agreed, at least up to last week we all agreed, that this
kind of experiment is indeed looking at an open-loop capability of the
observer.

I certainly didn’t agree to that. The Schouten experiment is looking at a closed-loop capability; the ability to control something like a perception of a relationship; what exactly is being controlled would have to be determined by test.

If you truly disagree with what you agreed earlier (I guess
because Bill persuaded you, not because I did)

Bill persuaded me only that I was being unclear about which loops I was talking about. To the extent that I was implying that the relationship between IV and DV in a conventional psychology experiment closed-loop, that was a mistake. The relationship between IV and DV in an experiment is open-loop because the subject has no influence over the state of the IV. So in this sense, all experiments, including tests for the controlled variable, are open-loop.It’s the system (subject) in these experiments that is closed-loop with respect to a CV that is influenced by both the IV and DV.

(1) There is a lot more to studying control that testing for the
controlled variable.

You bet. But you have to determine what variables are being controlled before you can start doing these other studies.

(2) When there is very little question as to whether a person is
controlling a variable that she is asked to control at a reference
level that is requested, then there seems no point in testing whether
such is the case, nor for testing all of the myriad of other variables
she is controlling at the same time.

I think there is always quite a bit more than “very little question” about what variables are being controlling when a person is instructed to control something in an experiment. And if you don’t know precisely what perceptual variables are being controlled in a situation there is not much that you can learn about how that controlling occurs.

You seem obsessed with the idea that a person is always controlling
only one single variable, that this variable is the only thing the
person can perceive, and that the only thing that could possibly be
interesting about the person is what that variable might be.

Well, I’m not. I am obsessed with the idea that purposeful behavior is organized around the control of perceptual variables and that the first step in understanding any behavior that seems purposeful – including the behavior in experiments – is determining what variables are under control. I am well aware of the fact that any behavior involves the control of many perceptual variables simultaneously. But I also know that it’s possible to identify one of two controlled perceptions that explain much of the relevant aspects of the behavior of interest. For example, I can explain most of the variance in observed measures of movement while catching a ball, by assuming that two variables are under control, vertical optical velocity and lateral displacement. But, as I show in the JEP:HPP article, more tests are needed to see if these are the best descriptions of the variables controlled in this situation.

Even if
such a blinkered view of a person’s perceptual and controlling ability
were correct, I think it would still be interesting to figure out
things like the gain of the control loop, transport lags, the form of
the perceptual function, the limitations of the environmental feedback
function, the side-effects of control actions, and the energetic
efficiency of control. You would prohibit all such investigations, or
at least relegate them to the “co-opters” who should be consigned to
the nether regions.

Not true at all. I’ve studied all of those things; but you can only study them when you have a very good idea of what variable(s) are under control. If you don’t know what variables are controlled in a particular behavioral situation then it makes no sense to start measuring things like the gain of a control loop because you don’t know what the control loop is (what is being controlled) or even whether there is a control loop involved, for that matter. Imagine trying to study the gain of a thermostat without knowing what variable it is controlling; if it turn out that the “thermostat” is actually controlling ( temperature x water content of the air) you will get some pretty uselessmisleading measures of gain.

Me:

The researcher in the Figure is implicitly aware of the fact that the
observed IV-DV relationship could reflect the open-loop input-output
(I-O) characteristics of the system, if the system is open loop (as in
the upper part of the Figure) or it could reflect the inverse of the
feeback connection between O and I, as in the lower part. In order to
properly analyze the situation you have to know whether the system is
closed loop and, if so, what variables it is controlling that are
relevant to the observed IV-DV relationship.

Yes. In respect of Schouten, we spent weeks making that clear, until
finally Bill said [From Bill Powers (2009.02.20.1946 MST)] in response
to one of your messages:

--------quote----------

Yes, Martin emphasized that the action does not affect the
presentation. That loop is not closed, right? It could be closed if the
action did affect the presentation, which isn’t impossible, but it
doesn’t, so that potential loop is not closed – which means it’s open,
as we use the term. There IS an open-loop relationship between S and R,
isn’t there? Or are you saying that you have to have a closed-loop
relationship and open it before you can call it open-loop? Your rules
for applying these terms are not at all clear.

--------end quote-------

Right. Bill was saying that I was not being clear about what is closed-loop in the experiment. It’s the system (organism)that is closed loop, not the relationship between S (the “presentation” or IV) and R (DV), which is open-loop. As I said earlier, what Bill is pointing out is that the relationship between IV and DV in any experiment is open-loop. The Schouten experiment is no more open-loop in this sense than any experiment in psychology. But the subject in the Schouten experiment is definitely closed-loop and is controlling an input variable that is influenced by the IV and DV simultanously. My little figure shows that the open-loop relationship between IV and DV reflects open loop characteristics of the system under study only if the system itself is open-loop. If the system is not open-loop, then the open-loop relationship between IV and DV reflects the inverse of the feedback function relating output to input.

It’s not a question of reliability; it’s a question of validity. That’s
what my attached Figure shows.

OK. You can make your point if you can show how the subject’s response
“bypasses the circle” influences the subject’s perception of whether
the straight line between the dots does bypass the circle. Otherwise,
we have the same situation as in Bill’s quote: “Yes, Martin
emphasized that the action does not affect the presentation. That loop
is not closed, right? It could be closed if the action did affect the
presentation, which isn’t impossible, but it doesn’t, so that potential
loop is not closed – which means it’s open, as we use the term.

Again, Bill is talking about the open-loop relationship between IV (the presentation) and DV, which is open loop in all experiments. It’s the system (organism) that is closed-loop. That’s why (as my figure shows and as Bill has noted in his description of hte "{behavioral illusion) the observed, open-loop relationship between IV (presentation) and DV is not a valid indication of the open-loop I-O characteristics of the system under study (the organism is psychological studies).

Best

Rick

···


Richard S. Marken PhD
rsmarken@gmail.com

[Martin Taylor 2009.03.16.16.58]

[From Rick Marken (2009.03.15.2050)]

What a strange message! At first I decided that there was no point in
responding, and I probably was right. But the other side of my conflict
seems to have won.

Martin Taylor
(2009.03.15.13.22)

From my
standpoint, if you can reasonably assume that when you ask someone to
push a button when they see a light come on, they will control for
doing exactly that. No test for the controlled variable is necessary.

You don’t test for controlled variables to figure out whether or not
people are following instructions in a conventional experiment. You
test for controlled variables in order to understand what people are
controlling when they are carrying out particular behaviors of
interest, which could include the behavior in psychological experiments
that were done with other goals in mind. But if you don’t know what
people are controlling then you can’t really study anything else about
their behavior.

And we all agreed, at least
up to last week we all agreed, that this
kind of experiment is indeed looking at an open-loop capability of the
observer.

I certainly didn’t agree to that. The Schouten experiment is looking at
a closed-loop capability; the ability to control something like a
perception of a relationship; what exactly is being controlled would
have to be determined by test.

The Schouten experiment is USING a closed loop capability; – several,
as we analyzed the situation, but most importantly controlling for a
match between presentation and answer (not button-pushing response), to
look at an open-loop capability, the channel capacity of the “red
arrow” path in the diagrams that were presented. If you hadn’t agreed
to that, your language at the time and for a short time thereafter made
a very good impression that you had. But I guess the elastic band has
snapped back, and you have retreated the position you held before you
gained insight at Bill’s behest.

from here on, you go on with your mantra, with which I think everyone
on this list would agree, that people are control systems, that all
behaviour is the control of perception, that you have to have a good
idea what people are probably controlling before determining the
validity of data, and so forth. You cast it as some kind of accusation
against me, as though I didn’t believe these things, which is why I
called it a “strange message”. But there are other strangenesses. You
seem to argue that even when a subject is asked to push a button if
they see a light, and seems to do so, and says they do so, it is
necessary to do “The Test” to determine whether they are actually
controlling for pushing a button when they see a light. I find that
more than a little weird.

Your messages remind me very much of the people who say you can’t be a
good person unless you believe in Jesus. They carry very little flavour
of the scientist.

Martin

[From Rick Marken (2009.03.16.2340)]

Martin Taylor (2009.03.16.16.58)–

The Schouten experiment is USING a closed loop capability; – several,
as we analyzed the situation, but most importantly controlling for a
match between presentation and answer (not button-pushing response), to
look at an open-loop capability, the channel capacity of the “red
arrow” path in the diagrams that were presented.

I saw the red arrows in your diagram but I didn’t see what data corresponded to the open loop channel capacity. The graph of d’ vs delay seemed to me to show how the timing of the elements of the controlled perception (light/press relationship) affect the ability to control it (as in my hierarchical perception demo).

From here on, you go on with your mantra, with which I think everyone
on this list would agree, that people are control systems, that all
behaviour is the control of perception, that you have to have a good
idea what people are probably controlling before determining the
validity of data, and so forth.

You were doing pretty well up until “you have to have a good
idea what people are probably controlling before determining the
validity of data”. The main point I made (in my thermostat example that you seem to have ignored) is that you have to have a very good (i.e. quantitative) idea of what a person is actually (not probably) controlling in order to be able to determine other properties of the control system, such as it’s gain, transport lag, order (e.g., proportional, integral), etc. I explained the problem in terms of a familiar control system, the thermostat. You can’t learn much about the properties of a thermostat until you know 1) that is a thermostat rather than, say, a humidistat and 2) exactly what the thermostat is controlling – the temperature at the sensor.

You
seem to argue that even when a subject is asked to push a button if
they see a light, and seems to do so, and says they do so, it is
necessary to do “The Test” to determine whether they are actually
controlling for pushing a button when they see a light. I find that
more than a little weird.

I’m confident that you will understand this some day and realize that it is not weird at all. For now, let me just suggest several different perceptions, some or all of which might be under control when we see a person push a button when they see a light: 1) the perception of a particular time interval between light and press 2) the perception of a sequence: light, press 3) the perception of an event: light-press. 4) the perception of a relationship: press follows light 4) the perception of a program contingency: if light then press else no press 5) the control of a transition; change over time from light to press.

There are many more possibilities (for example, it might be the sound rather than the tactile impression of the press that is an element of the controlled perception). In order to understand anything about the controlling involved in pressing when a light comes on – or in any control task, like running to intercept a fly ball – you have to know precisely what the subject is controlling. Once you know precisely (quantitatively) what is being controlled you can start studying those other aspects of control, such as gain and transport lag, that you find so interesting.

Best

Rick

···


Richard S. Marken PhD
rsmarken@gmail.com

[From Bill Powers (2009.03.17.1515 CST)]

From the veranda in Pamela Fox's Villa Nirvana, north of Acapulco, MX:

Rick Marken (2009.03.16.2340) to Martin Taylor:

I'm confident that you will understand this some day and realize
that it is not weird at all. For now, let me just suggest several
different perceptions, some or all of which might be under control
when we see a person push a button when they see a light: 1) the
perception of a particular time interval between light and press 2)
the perception of a sequence: light, press 3) the perception of an
event: light-press. 4) the perception of a relationship: press
_follows_ light 4) the perception of a program contingency: if light
then press else no press 5) the control of a transition; change over
time from light to press.

Well, put Rick. I agree with your position here. The way I see it,
what the experimenter "just knows" is being controlled, or what the
experimenter thinks his instructions conveyed to the subject, carries
no scientific weight at all. This isn't a question about the
experimeter's integrity or intelligence: it's about how you test a
theory without leaving more loopholes than you need to.

It really takes very little effort to fill in the gaps. In the
"TrackAnalyze" experiment, for example, the subject is asked to keep
the cursor aligned with the target, which is equivalent to
maintaining a reference condition for the target-cursor position of
zero. So one of the adjustable variables in the analysis is the
apparent reference setting for the target-cursor distance. It is not
simply assumed that the subject's reference condition corresponds to
the meaning of the instructions (that is, what they meant to the
experimenter who wrote them). The analysis finds the best value for
the reference condition as one parameter among three others. Normally
it comes out to less than a few pixels different from zero, and one
then has to judge whether this is close enough to say that the
subject behaved as directed. Considering that the range of target
movement is around 750 pixels on my laptop screen, a few pixels is
close enough. But even if we reject that idea, we can still analyze
the control behavior using whatever reference condition seems
actually to have been adopted.

This checkup gives the nice feature that if the person decides to
keep the cursor 1 cm above the target during the run, the analysis
program will happily report that fact and tell you what the reference
distance actually was. If I had simply assumed that the subject would
follow the instructions, that would not be possible.

The ideal is for an experiment to be set up so all the most important
assumptions that go into the conclusions are tested, whether they
seem obvious or not. In this way every experiment can stand (or fall)
by itself. There's a practical limit on how thorough this checking
can be, but to omit checks altogether doesn't fit my idea of proper
scientific skepticism -- especially the kind of skepticism that
counts, skepticism about one's own assumptions.

I read that the signoff message "Best" is considered by some to be
the ultimate brushoff, but really, I just mean "Best regards and
wishes to everyone."

Best,

Bill.

No virus found in this outgoing message.
Checked by AVG - www.avg.com
Version: 8.0.238 / Virus Database: 270.11.18/2008 - Release Date: 03/17/09 16:25:00

[Martin Taylor 2009.03.17.16.15 begorrah]

[From Rick Marken (2009.03.16.2340)]

Martin Taylor
(2009.03.16.16.58)–

The Schouten experiment is
USING a closed loop capability; – several,
as we analyzed the situation, but most importantly controlling for a
match between presentation and answer (not button-pushing response), to
look at an open-loop capability, the channel capacity of the “red
arrow” path in the diagrams that were presented.

I saw the red arrows in your diagram but I didn’t see what data
corresponded to the open loop channel capacity.

I’m sorry. I had thought you had understood at one point. My mistake.

The “open loop” part is the path between the physical lights and the
perceptual input function of the “match answer to presentation” control
loop.

The graph of d’ vs delay seemed to me to show how the timing of
the elements of the controlled perception (light/press relationship)
affect the ability to control it (as in my hierarchical perception
demo).

From here on, you go on with
your mantra, with which I think everyone
on this list would agree, that people are control systems, that all
behaviour is the control of perception, that you have to have a good
idea what people are probably controlling before determining the
validity of data, and so forth.

You were doing pretty well up until “you have to have a good
idea what people are probably controlling before determining the
validity of data”. The main point I made (in my thermostat example that
you seem to have ignored) is that you have to have a very good (i.e.
quantitative) idea of what a person is actually (not probably)
controlling in order to be able to determine other properties of the
control system, such as it’s gain, transport lag, order (e.g.,
proportional, integral), etc. I explained the problem in terms of a
familiar control system, the thermostat. You can’t learn much about the
properties of a thermostat until you know 1) that is a thermostat
rather than, say, a humidistat and 2) exactly what the thermostat is
controlling – the temperature at the sensor.

I ignored that because it’s (a) trivial, (b) self-evident, and (c)
irrelevant. Or it is unless you can show that the same kind of results
could be obtained if subjects are NOT controlling for a match between
their perception of the presentation and the available answers. What
else they are controlling when they are behaving in the experiment is
irrelevant to the interpretation of the results.

You
seem to argue that even when a subject is asked to push a button if
they see a light, and seems to do so, and says they do so, it is
necessary to do “The Test” to determine whether they are actually
controlling for pushing a button when they see a light. I find that
more than a little weird.

I’m confident that you will understand this some day and realize that
it is not weird at all. For now, let me just suggest several different
perceptions, some or all of which might be under control when we see a
person push a button when they see a light: 1) the perception of a
particular time interval between light and press 2) the perception of a
sequence: light, press 3) the perception of an event: light-press. 4)
the perception of a relationship: press follows light 4) the
perception of a program contingency: if light then press else no press
5) the control of a transition; change over time from light to press.

Some or all of these were in fact mentioned, along with several others,
in the discussions that led to the point at which I thought you had
understood. The only key point is that all of them except the
availability of input data are unaffected by the timing of the
bip-sequence with respect to the switching on of one of the lights.

If what you are studying is, instead, how people go about controlling
for pressing a button corresponding to a light in synchrony with an
auditory signal, then those controls you mention are what you should
study.

There are many more possibilities (for example, it might be the sound
rather than the tactile impression of the press that is an element of
the controlled perception). In order to understand anything about the
controlling involved in pressing when a light comes on – or in any
control task, like running to intercept a fly ball – you have to know
precisely what the subject is controlling. Once you know precisely
(quantitatively) what is being controlled you can start studying those
other aspects of control, such as gain and transport lag, that you
find so interesting.

Maybe you understand now?

The question being asked in detection studies, of which the Schouten
study was one, is of the capability of an open-loop pathway. The
subject’s behaviour in the experiment is affected by all sorts of
control systems and the reference levels for their perceptions, and
those are legitimate objects of study. When the subject cannot
influence the presentation by varying output, then all that matters is
that one of the subject’s control systems is controlling for a
relationship between the perception of the presentation and the answers
in a way corresponding to what the experimenter requested. All the rest
is quite irrelevant, so long as both the subject and the experimenter
have good reason to believe that the subject is indeed trying to do
what was asked. The way the experimenter can determine this is from the
same kinds of consistencies that occur in a tracking study. If that’s
not satisfactory, both kinds of study permit internal consistency
checks.

In a tracking study, one can observe whether the subject’s reference
value appears to be as requested, but if it seems not to be, you have
no idea whether the reason is that the subject’s perception is biased,
or that the subject is not accepting the experimenter’s reference
value. You have only the subject’s word for it that she is trying to do
as asked. At that point it’s a pure assumption on the experimenter’s
part whether the problem is the subject or the subject’s perception. To
look further, one way is to open the loop, and do an experiment with
the same kind of presentation, but in which the subject cannot
manipulate the target. Supposing the tracking task is a compensatory
one in which the subject is asked to keep the target horizontally level
with a marker, one might present the subject with cases in which the
target is lightly above and below the presumed reference level, and see
whether the subject claims to see it as level with the marker when the
deviation is what the reference level bias was in the tracking study or
when it is physically level with the marker.

You can use control studies to test and validate the results of
conventional studies, and, as I just showed, you can do the reverse.

···

As a matter of possible interest: For other purposes, I happened to
glance at an archive from 1997, and I see we were engaged in the same
discussion, using many of the same words, over a decade ago. At that
time, I let it slide, perhaps because I got bored with trying to
achieve a mutual understanding. But it is rather discouraging to think
that we have not moved into a more subtle level of disagreement in all
that time. In fact, the current reversion may even be cruder, since at
the end of one message [Martin Taylor 971231 17:45] I said:

-----------quote-------------

"The question arises as to whether the perceptual input functions
operate the same way when the resulting perception is being controlled
as when it isn’t. This issue is not ordinarily considered within HPCT,
since normally the perceptual input function is taken to be whatever it
is, and only the magnitude of its output is controlled. But it is an
issue, one that might invalidate the uncritical use of the results of
psychophysical studies to assess the elements of related control loops.
"

---------end quote----------

The answer to that question could be quite interesting, but I don’t
know an obvious way to address it. It’s certainly more interesting than
simply answering one message after another that says how naive I am and
that I really don’t understand that humans are control systems.

I want to advance the science of perceptual control, which is something
you berated me for wanting to do in my early days on CSGnet. I think I
understand PCT well enough by now to be permitted to make this attempt,
whereas at the time I thought you might be right and I was being
arrogant in thinking otherwise. If I’m arrogant to think that I might
have something new to say from time to time, so be it. I’m getting too
old to wait until my understanding will pass your elementary exams.

Martin

[From Rick Marken (2009.03.18.2030)]

Martin Taylor 2009.03.17.16.15 begorrah]

Rick Marken (2009.03.16.2340)]

I saw the red arrows in your diagram but I didn’t see what data
corresponded to the open loop channel capacity.

I’m sorry. I had thought you had understood at one point. My mistake.

The “open loop” part is the path between the physical lights and the
perceptual input function of the “match answer to presentation” control
loop.

Yes, that path is open-loop; it’s the path that goes from IV to DV in all conventional experiments. But my question was about data; what data is it that tells you about the open loop channel capacity of the subject. I hope you don’t think it’s the delay vs d’ data. That is the data that shows the relationship between IV and DV. But we know from the analysis of closed loop systems that that relationship doesn’t represent the open loop characteristics of the system. Taking this IV-DV relationship as an indication of an open-loop characteristic of the system is precisely what is meant by the “behavioral illusion” that I describe in the “Revolution” paper.

You were doing pretty well up until “you have to have a good
idea what people are probably controlling before determining the
validity of data”. The main point I made (in my thermostat example that
you seem to have ignored) is that you have to have a very good (i.e.
quantitative) idea of what a person is actually (not probably)
controlling in order to be able to determine other properties of the
control system, such as it’s gain, transport lag, order (e.g.,
proportional, integral), etc.

I ignored that because it’s (a) trivial, (b) self-evident, and (c)
irrelevant. Or it is unless you can show that the same kind of results
could be obtained if subjects are NOT controlling for a match between
their perception of the presentation and the available answers.

I mentioned several other variables that the subjects might be controlling, and when any one of them is controlled you could observe the sa,e kind of results. My point was that describing the controlled variable as “a match between
their perception of the presentation and the available answers” is pretty vague. Many different perceptions, if controlled, could result in the observed behavior. It’s like saying that the fielder in the ball catching studies was controlling for a match between his location and the ball’s location when it comes down. To understand catching you have to know the controlled variables in quantitative detail.

What
else they are controlling when they are behaving in the experiment is
irrelevant to the interpretation of the results.

What else they are controlling may be irrelevant but what they are controlling is very relevant to the interpretation of the results. The results of the Schouten experiment – I’m taking the results to be the observed relationship between delay and d’ – could be quite different depending on what is actually being controlled. Think about it in terms of my size control task; the subject could be controlling area (x * y) or perimeter 2* (x + y) or even sqrt (x^2 +y^2). Suppose you do an experiment to see how variations in x (the IV) relate to variations in y (the DV) when the subject is asked to keep size constant. This experiment is exactly analogous to yours, with x as the “presentation” or IV, y as the “answer” or DV and “constant size” as the “match between light and answer” or CV. The observed relationship between x and y is analogous to the observed relationship betwen delay and d’. Depending on which CV – x * y, 2 *(x + y) or sqrt (x^2 +y^2 – is actually being controlled, the observed result of this experiment – the observed relationship between x (IV) and y (DV) – will be somewhat different. Only by correctly and quantitatively identifying the relevant controlled variable(s) in an experiment is it possible to then go on (using modeling) to study the characteritics of the control system, such as gain and transport lag.

I’m confident that you will understand this some day and realize that
it is not weird at all. For now, let me just suggest several different
perceptions, some or all of which might be under control when we see a
person push a button when they see a light:

Some or all of these were in fact mentioned, along with several others,
in the discussions that led to the point at which I thought you had
understood. The only key point is that all of them except the
availability of input data are unaffected by the timing of the
bip-sequence with respect to the switching on of one of the lights.

The timing of the bip-sequence is very likely a component of the controlled variable. There is no way to tell, however. whether the variable controlled in the experiment is or is not affected by the bip sequence until you have tested to determine what the relevant controlled variable(s) are in this situation. See Bill’s comment on my post in [Bill Powers (2009.03.17.1515 CST)] to see why.

The question being asked in detection studies, of which the Schouten
study was one, is of the capability of an open-loop pathway.

As I note in my “Revolution” paper, that is the question being asked by all conventional experiments. My paper explains why this approach to studying the open-loop characteristics of the system won’t work; it’s because the system itself is closed loop.

When the subject cannot
influence the presentation by varying output, then all that matters is
that one of the subject’s control systems is controlling for a
relationship between the perception of the presentation and the answers
in a way corresponding to what the experimenter requested.

This is exactly the same situation in all experiments. The subject cannot influence the presentation (IV); the IV is a disturbance to a variable the subject is controlling (such as a perceived relationship between IV and DV, as in the detection task). In such experiments observed relationships between IV and DV reflect, now open loop characteristics of the system but the inverse of the feedback connection between DV and CV.

In a tracking study, one can observe whether the subject’s reference
value appears to be as requested, but if it seems not to be, you have
no idea whether the reason is that the subject’s perception is biased,
or that the subject is not accepting the experimenter’s reference
value. You have only the subject’s word for it that she is trying to do
as asked. At that point it’s a pure assumption on the experimenter’s
part whether the problem is the subject or the subject’s perception.

See [ Bill Powers (2009.03.17.1515 CST)]. Good studies of control are based on far more rigorous tests than taking the subject’s word.

I want to advance the science of perceptual control, which is something
you berated me for wanting to do in my early days on CSGnet.

I can’t believe that, but I was so much older then;-) I would like you and anyone else to advance the science of perceptual control. I’m just finding some apparent problems with your descriptions of how to conduct this science.

I think I
understand PCT well enough by now to be permitted to make this attempt,
whereas at the time I thought you might be right and I was being
arrogant in thinking otherwise. If I’m arrogant to think that I might
have something new to say from time to time, so be it. I’m getting too
old to wait until my understanding will pass your elementary exams.

e
I think you understand PCT just fine. But you are saying some things about how to study control systems that don’t seem right to me. It’s hard for me to avoid commenting on this topic – methods of studying control systems – because it is the focus of my interest in PCT, as you can tell from th papers I’ve published on the topic, including the recent “Revolution” paper. So if you don’t want me to comment on your posts it’s probably best to stay away from the topic of methodology.

Best regards

Rick

···

Richard S. Marken PhD
rsmarken@gmail.com

[Martin Taylor 2009.03.19.05.28]

[From Rick Marken (2009.03.18.2030)]

Martin Taylor
2009.03.17.16.15 begorrah]

Rick Marken (2009.03.16.2340)]

I saw the red arrows in your diagram but I didn’t see what data
corresponded to the open loop channel capacity.

I’m sorry. I had thought you had understood at one point. My mistake.

The “open loop” part is the path between the physical lights and the
perceptual input function of the “match answer to presentation” control
loop.

Yes, that path is open-loop; it’s the path that goes from IV to DV in
all conventional experiments.

I guess you haven’t looked at the diagram. The DV is not the input to
the perceptual input function of the “Match answer to presentation”
control loop. Here’s the diagram again, in generic form, without all
the irrelevant control loops.

question-answer.jpg

But my question was about data; what data is it that tells you
about the open loop channel capacity of the subject.

You keep changing the topic, talking about “the open loop channel
capacity of the subject”. You are the only one in this discussion to
consider this to be a possible topic of conversation, since nobody else
(so far as I know) considers the concept to have any meaning.

What the data tells you about is the channel capacity of the “red
arrow” pathway.

I hope you don’t think it’s the delay vs d’ data. That is the
data that shows the relationship between IV and DV.

This is true. However, the rest of the control apparatus is invariant.
All that changes is the relation of the onset time of the lights to all
the controlling events that go into the subject pushing the button in
synchrony with the bips. That is to say, the timing of the growth of
the perceptual signal on the left of the “Perceptual Function” box in
the figure.

But we know from the analysis of closed loop systems that that
relationship doesn’t represent the open loop characteristics of the
system. Taking this IV-DV relationship as an indication of an open-loop
characteristic of the system is precisely what is meant by the
“behavioral illusion” that I describe in the “Revolution” paper.

You deal with a structurally different situation. It’s not comparable.
The relation between the subject’s output and the presentation input is
not closed-loop. There is no influence of the subject’s output and
whether the next input is “left light” or “right light”.

You were doing pretty well up until “you
have to have a good
idea what people are probably controlling before determining the
validity of data”. The main point I made (in my thermostat example that
you seem to have ignored) is that you have to have a very good (i.e.
quantitative) idea of what a person is actually (not probably)
controlling in order to be able to determine other properties of the
control system, such as it’s gain, transport lag, order (e.g.,
proportional, integral), etc.

I ignored that because it’s (a) trivial, (b) self-evident, and
(c)
irrelevant. Or it is unless you can show that the same kind of results
could be obtained if subjects are NOT controlling for a match between
their perception of the presentation and the available answers.

I mentioned several other variables that the subjects might be
controlling, and when any one of them is controlled you could observe
the sa,e kind of results.

You had better explain this. You must describe how the same kind of
results would be explained if the subject is NOT controlling for a
match between the perception of which light is lit and the answer
(which sets a reference for which button to push). I don’t mind what
you describe, so long as it does not include such a matching control
loop or its equivalent.

My point was that describing the controlled variable as “a match
between
their perception of the presentation and the available answers” is
pretty vague. Many different perceptions, if controlled, could result
in the observed behavior. It’s like saying that the fielder in the ball
catching studies was controlling for a match between his location and
the ball’s location when it comes down. To understand catching you have
to know the controlled variables in quantitative detail.

Indeed. You have a closed loop situation there. The catcher’s actions
influence his perception of the ball’s location relative to himself and
the rest of the environment.

What
else they are controlling when they are behaving in the experiment is
irrelevant to the interpretation of the results.

What else they are controlling may be irrelevant but what they are
controlling is very relevant to the interpretation of the results. The
results of the Schouten experiment – I’m taking the results to be the
observed relationship between delay and d’ – could be quite different
depending on what is actually being controlled.

Make a suggestion, as I’ve asked above and in previous messages, and
then we can analyze whether there is any way of choosing between it and
the obvious one (the subject intends to match the answer “left” or
“right” to the perception of “left light” or “right light”). After (or
perhaps before) we do that, we can determine whether it makes any
difference to the interpretation of the data as determining a lower
bound on the channel capacity of the “red arrow” pathway, whern your
suggestion is substituted for the match control system.

Think about it in terms of my size control task; the subject
could be controlling area (x * y) or perimeter 2* (x + y) or even sqrt
(x^2 +y^2). Suppose you do an experiment to see how variations in x
(the IV) relate to variations in y (the DV) when the subject is asked
to keep size constant. This experiment is exactly analogous to yours,
with x as the “presentation” or IV, y as the “answer” or DV and
“constant size” as the “match between light and answer” or CV. The
observed relationship between x and y is analogous to the observed
relationship betwen delay and d’. Depending on which CV – x * y, 2 *(x

  • y) or sqrt (x^2 +y^2 – is actually being controlled, the observed
    result of this experiment – the observed relationship between x (IV)
    and y (DV) – will be somewhat different.

Of course they will. You had a bunch of plausible possibilities – an
infinite set, actually. You chose to model three of them, but you have
no guarantee that any of them is what a subject actually uses, or even
whether the same function applies as the configuration changes from
long and thin to squarish. All you know is that of the ones you
compared, one of them models the results better than do the others.
Nothing about it is precise.

Only by correctly and quantitatively identifying the relevant
controlled variable(s) in an experiment is it possible to then go on
(using modeling) to study the characteritics of the control system,
such as gain and transport lag.

I’m confident that you will understand
this some day and realize that
it is not weird at all. For now, let me just suggest several different
perceptions, some or all of which might be under control when we see a
person push a button when they see a light:

Some or all of these were in fact mentioned, along with several others,
in the discussions that led to the point at which I thought you had
understood. The only key point is that all of them except the
availability of input data are unaffected by the timing of the
bip-sequence with respect to the switching on of one of the lights.

The timing of the bip-sequence is very likely a component of the
controlled variable. There is no way to tell, however. whether the
variable controlled in the experiment is or is not affected by the bip
sequence until you have tested to determine what the relevant
controlled variable(s) are in this situation. See Bill’s comment on my
post in [Bill Powers (2009.03.17.1515 CST)] to see why.

Yu would have a point if it were not for the fact that whatever
variable is controlled in this experiment, it is assumed NOT to be
affected by the light timing. What is affected is the information
available to that control system, and that influences the accuracy of
the reference signal it provides to all the control loops involved with
button pushing; they are the ones affected by the bip timing, not the
one that provides the answer reference signal. At least that’s the way
I conceived it. If you have a different proposal, it would be nice, as
I have several times said, to know what it might be.

The question being asked in detection studies, of which the Schouten
study was one, is of the capability of an open-loop pathway.

As I note in my “Revolution” paper, that is the question being asked by
all conventional experiments.

And in some it is valid, in some it is not. To lump all “conventional”
experiments together is unscientific.

My paper explains why this approach to studying the open-loop
characteristics of the system won’t work; it’s because the system
itself is closed loop.

All closed loops contain open-loop pathways.

When the subject cannot
influence the presentation by varying output, then all that matters is
that one of the subject’s control systems is controlling for a
relationship between the perception of the presentation and the answers
in a way corresponding to what the experimenter requested.

This is exactly the same situation in all experiments. The subject
cannot influence the presentation (IV); the IV is a disturbance to a
variable the subject is controlling (such as a perceived relationship
between IV and DV, as in the detection task). In such experiments
observed relationships between IV and DV reflect, now open loop
characteristics of the system but the inverse of the feedback
connection between DV and CV.

In a tracking study, one can
observe whether the subject’s reference
value appears to be as requested, but if it seems not to be, you have
no idea whether the reason is that the subject’s perception is biased,
or that the subject is not accepting the experimenter’s reference
value. You have only the subject’s word for it that she is trying to do
as asked. At that point it’s a pure assumption on the experimenter’s
part whether the problem is the subject or the subject’s perception.

See [ Bill Powers (2009.03.17.1515 CST)].

I wrote that as an answer to Bill’s comment about being able to
determine reference levels in a tracking study. I should have
referenced his message, but I thought you would get the point. I’m
saying that the PCT researcher has to make the same kind of assumptions
and is able to check them using the same kind of methods as one does in
any “conventional” detection study. Bill was making the assumption that
the subject’s perception was unbiased and that therefore a measurable
average deviation of the track from the target indicates a failure to
set the reference level correctly. That may be correct, but it’s an
assumption of the same kind as assuming that when the subject is asked
to push a button corresponding to which light is lit, the subject is
actually controlling a perception of doing just that.

Good studies of control are based on far more rigorous tests
than taking the subject’s word.

Good studies of anything are based on far more rigorous tests than
taking the subject’s word.

I think you understand PCT just fine. But you are saying some things
about how to study control systems that don’t seem right to me.

That much is apparent. I keep trying to show you why they are in fact
right, and am quite prepared for you to show me how they are in fact
wrong. But you never do. You make general pontifications of the kind
“Of course the stars and planets are set in crystal spheres; crystal is
the perfect material and spheres are the perfect form, so how could one
possibly move in an imperfect ellipse?” I want to see analyses of my
errors, not false analogies and mystical necessities. I know that
humans are control systems, and that every output is the result of
actions to control some perceptions (or at least I should not say I
“know” it, because I only strongly believe it). I “know” that
perception has many levels of abstraction (and have believed that since
at least the age of 10, when I read about the work of Donders), and
that therefore behaviour has to be seen at the same levels of
abstraction. I “know” that the behaving organism is physical, and is
subject to the laws of thermodynamics – which is the primary reason
that I am so convinced of the rightness of PCT. There’s a lot else
about it that you don’t have to keep reminding me of, as if it was in
some way inconsistent with things I say (unless it is in fact
inconsistent, but so far, you haven’t done that).

It’s hard for me to avoid commenting on this topic – methods of
studying control systems – because it is the focus of my interest in
PCT, as you can tell from th papers I’ve published on the topic,
including the recent “Revolution” paper. So if you don’t want me to
comment on your posts it’s probably best to stay away from the topic of
methodology.

I don’t want you to comment on my posts; I want you to comment on the
CONTENT of my posts, the ideas, the details, what’s right and what’s
wrong, why what’s right is right, and why what’s wrong is wrong, or
whether there are debatable points about which we should not be sure
either way. You hardly ever do, and I wish you would, so we wouldn’t
need to be having the same discussion 12 years after we should have
arrived at a concensus.

Martin

[From Bill Powers (2009.03.20.0818 MDT)]

[Martin Taylor 2009.03.19.05.28]

[From Rick Marken
(2009.03.18.2030)]

The “open loop” part
is the path between the physical lights and the perceptual input function
of the “match answer to presentation” control
loop.

Yes, that path is open-loop;
it’s the path that goes from IV to DV in all conventional experiments.

I guess you haven’t looked at the diagram. The DV is not the input to the
perceptual input function of the “Match answer to presentation”
control loop. Here’s the diagram again, in generic form, without all the
irrelevant control loops.

7f942.jpg

As Martin points out, Rick has the “answer” output above
confused with the S1 input to the perceptual function. But Martin has
omitted the input from the actually produced “Answers” to the
perceptual function: organisms do not simply compute (in imagination) the
reference signal they should emit to achieve a particular result and then
emit it; they also monitor and adjust the action to make sure that the
result of sending this reference signal is in fact the perception – that
is, the spoken or written answer – they have selected. The lower systems
that convert the reference signal as imagined into the answer as spoken
or written are not in the diagram, but I’m sure we can all imagine them
correctly. And I’m sure that we have all had the experience of producing
written or spoken answers that, when we hear or see ourselves saying it,
turns out not to convey the answer we had intended. Or am I the only one
here who reads posts again, and edits them, before sending them?

Therefore the higher system shown here is a closed-loop control system
which controls the relationship between the perceived answer and the
(observed) presentation. The presentation is not altered by the answer.
Changing the presentation alone disturbs the relationship; the answer is
adjusted to cancel this effect of the presentation and keep the
perception of the relationship matching the reference condition. Since we
have no information about just what this relationship is for the person
in question, the analysis stops there. The reference signal is not an
“interpretation” because an interpretation is a kind of
perception; it is a specification for the state of the perceptual
signal.

But my question was about data;
what data is it that tells you about the open loop channel capacity of
the subject.

You keep changing the topic, talking about “the open loop channel
capacity of the subject”. You are the only one in this discussion to
consider this to be a possible topic of conversation, since nobody else
(so far as I know) considers the concept to have any
meaning.

You two guys are clearly irritated with each other and letting it show,
and you aren’t trying very sincerely to understand each other or explain
things to each other. If you have no interest in changing anything you
think or explaining things so the other person can be persuaded to
change, why bother to have a conversation? Simply insisting that
you are right is a mode of interaction unworthy of your IQs.

The Schouten experiment, in my opinion, is clever and interesting but
leaves too many assumptions unjustified (as it stands) to serve as the
basis for any important conclusions. Martin may know, but the rest of us
don’t, just how the time of actual pressing of the button relates to the
timing of the bips, or in English, beeps. I don’t think anyone knows how
long before the button is actually pressed enough to close a contact the
decision to press it is made, or how long it takes after the light comes
on for that decision process to commence, or how long it takes, once it
commences, to try different answers and find the one that generates the
right relationship. We don’t know how the perceptual signal changes with
time after a light comes on, or for that matter how long it takes for the
filament to warm enough after the circuit is closed to generate visible
light. As in many psychological experiments, the mere physics of what is
going on (as well as the neurology) is oversimplified if it is noticed at
all.

In my view, this reduces the Schouten experiment from a finished product
to a pilot experiment – interesting, but incomplete. Too much of the raw
material for reasoning is provided by the analyst’s imagination, and not
enough by formal data-taking. I would say, yes, the Schouten experiment
looks interesting. So why not actually do it in such a way that others
will be convinced by it?

I’m prejudiced against information theory, because so far I haven’t seen
anything you can discover by using it that you can’t discover some other
way, except of course for the final conversion, through the log to the
base two, into units of bits. The old-timers in electrical engineering
liked to speak of amplification and signal loss in units of decibels, but
I never found that very helpful either. I wouldn’t really mind changing
my opinion about information theory, but just being told that I’m
ignorant doesn’t provide the kind of motivation that gets me going. I
think I can find the channel capacity of an electronic circuit
sufficiently well to estimate the limits on signals that can be
transmitted faithfully from input to output. If someone can tell me why
knowing that capacity measured in bits per second is more informative
than simply knowing the frequency response of the channel, I will
consider that possibility, when I need more precision.

As I recall, this whole subject arose either because of or after I made
some comments about people bringing old ideas and accomplishments with
them into PCT. Martin cited the Schouten experiment as an example of
valid information obtained from traditional (or at least non-PCT)
approaches. Apparently Martin hopes that something of value can be
retained out of the pre-PCT history of science. Rick argues against
Martin not because he understands what Martin says and sees what it wrong
with it, but because he wants Martin to be wrong about the value of
pre-PCT science. Of course I’m only reporting one bystander’s subjective
impressions here, but it doesn’t seem to me that anyone is really
interested in FINDING OUT what the truth of the matter is. They already
know the answer they prefer and are just looking for a way to make it
come out right.

I suppose I’m in a peculiar position here because I have no prior
accomplishments in psychology from long ago to defend and I never
believed much of what I learned in psychology courses. So maybe I can’t
properly sympathize with those who had inspiring advisors and who labored
long and hard to earn the approval of a thesis committee. But at the
moment I can’t just go along to get along.

Best,

Bill P.

[Martin Taylor 2009.03.20.14.32]

[From Bill Powers (2009.03.20.0818 MDT)]

[Martin Taylor 2009.03.19.05.28]

[From Rick Marken (2009.03.18.2030)]
The "open loop" part is the path between the physical lights and the perceptual input function of the "match answer to presentation" control loop.
Yes, that path is open-loop; it's the path that goes from IV to DV in all conventional experiments.

I guess you haven't looked at the diagram. The DV is not the input to the perceptual input function of the "Match answer to presentation" control loop. Here's the diagram again, in generic form, without all the irrelevant control loops.

As Martin points out, Rick has the "answer" output above confused with the S1 input to the perceptual function. But Martin has omitted the input from the actually produced "Answers" to the perceptual function: organisms do not simply compute (in imagination) the reference signal they should emit to achieve a particular result and then emit it; they also monitor and adjust the action to make sure that the result of sending this reference signal is in fact the perception -- that is, the spoken or written answer -- they have selected.

I can interpret this in at least three ways:
(1) That to determine the appropriate answer is a control process, varying the possible answers until one matches the category perception of the presentation.
(2) That there are control processes involved in producing output appropriate to the answer that matches the category perception of the presentation,
(3) That the perception that is controlled in the physical method of answering (button) is part of the perceptual input function to the matching process.

Since I don't know which, if any, of those you intend, here's my comment on all of them.
(1) Assumed to be true.
(2) Assumed to be true
(3) Unlikely but conceivable

I do have a problem with: "organisms do not simply compute (in imagination) the reference signal they should emit to achieve a particular result and then emit it; they also monitor and adjust the action to make sure that the result of sending this reference signal is in fact the perception -- that is, the spoken or written answer -- they have selected." This way of putting things seems to incorporate into a single control loop a variety of disparate functions and to make the controlled perception into a complex, rather than the scalar variable that we usually prefer to consider. There's nothing wrong with thinking about control of a complex variable (I don't mean one with "real" and "imaginary" parts in either the mathematical or the PCT sense of "real" and "imaginary"), and maybe we should, but so long as scalars work, Occam's Razor suggests we should continue to use them by preference.

In this situation, the response is more "fire and forget" than "continuous control". Yes, there is control, in that the subject can see the "fall of shot" (correctness of the button push) and use it to influence the choice of which button corresponds to which answer, or which answer to which perception. That seems to me to be a separate control system from the one that does the actual matching of the answer category to the category perception of the presentation.

Another problem I have is with the assertative mode, which suggests that you have indisputable knowledge of what organisms do or don't do: "Organisms do not ..." might be better stated as: "As I choose to model them, organisms do not..." Nothing in PCT argues that organisms do not compute in imagination the reference signal they should emit. How else can "fire and forget" systems, which alter their performance according to the results of earlier instances, determine where to aim to hit a particular target? "Last shot hit 20m left of where I aimed, so next time I have a target I will aim 20m further to its right". Is that not computing in imagination the reference signal that should be emitted to the gun layer for the next shot? What is the difference when it's not a gun but a button push that corresponds to an answer determined by control in imagination?

Therefore the higher system shown here is a closed-loop control system which controls the relationship between the perceived answer and the (observed) presentation. The presentation is not altered by the answer. Changing the presentation alone disturbs the relationship; the answer is adjusted to cancel this effect of the presentation and keep the perception of the relationship matching the reference condition.

Yes. That all goes on before the button-push action is initiated. The output should be the reference value for the button-push control system.

Since we have no information about just what this relationship is for the person in question, the analysis stops there.

I think we have reason to believe that the relationship is what the experimenter has requested that it be. In effect, the experimenter has performed "The Test for the Controlled Variable", even if he had no idea such a Test exists. He has determined that the subject can perceive the disturbing variable, created various disturbances (left-light, right-light) and shown that when the signal delay is sufficient, there is an almost perfect match between the button-push and the disturbance (output tracking the cursor disturbance in a tracking task). The Test says that somewhere inside the subject, there is a control system controlling for a match between the presentation and the answer. The only assumption necessary is that the subject is trying to control in the same way when the available information is insufficient as when it is sufficient. The reason it is sensible to believe this to be the case is that the data vary smoothly and consistently as the bip-delay changes, and more specifically that the rate of information gain is very tightly linear. If there were a break in the information gain rate, one might be led to suspect a change in the subject's control parameters or systems, but there isn't.

The reference signal is not an "interpretation" because an interpretation is a kind of perception; it is a specification for the state of the perceptual signal.

Yes, I see the "answer" as a reference signal for the system that controls the perception of which button is being pushed.

But my question was about data; what data is it that tells you about the open loop channel capacity of the subject.

You keep changing the topic, talking about "the open loop channel capacity of the subject". You are the only one in this discussion to consider this to be a possible topic of conversation, since nobody else (so far as I know) considers the concept to have any meaning.

You two guys are clearly irritated with each other and letting it show, and you aren't trying very sincerely to understand each other or explain things to each other.

I dispute that. I was frustrated by my inability to get Rick to do one of three things: (1) Understand what I am trying to say; (2) Explain to me what it is he doesn't understand; or (3) Explain to me what is wrong with something I have said. But I'm not frustrated any more, since Rick said it was now clear, and I choose to believe he was serious when he said that. For myself, I had been almost desperately trying to understand what he is trying to get at, beyond that humans are control systems. I've tried many different ways to ask, but with zero success.

The Schouten experiment, in my opinion, is clever and interesting but leaves too many assumptions unjustified (as it stands) to serve as the basis for any important conclusions. Martin may know, but the rest of us don't, just how the time of actual pressing of the button relates to the timing of the bips, or in English, beeps.

I take a "bip" to be a short "beep". Since I made the effort to find Schouten's paper, I do now know how the time of the button press relates to the bips. The distribution is what one might call "by eye Gaussian" with a standard deviation averaged over all subjects that depends on the actual bip time (by the way, the bip sequence was 3 20 msec bips separated by 75 msec). Scouten mentions that the changes in the histograms of button-press times as a function of bip time was different for different subjects, so one should not take the averaged data as any more than suggestive. For what it's worth, the by-eye SD in the mid-range of bip delays combining the data from all subjects was about 20 msec, but wider for very short and very long delays, so for any individual subject, the variation must be less than that. Schouten mentions that only one subject was precise at very short delays, and only a few were precise at the long delays, so your "grain of salt" in considering the SD I mentioned should be rather large.

I don't think anyone knows how long before the button is actually pressed enough to close a contact the decision to press it is made, or how long it takes after the light comes on for that decision process to commence, or how long it takes, once it commences, to try different answers and find the one that generates the right relationship.

Right. But unless there is reason to believe these things change as a function of bip time, they don't influence the interpretation of the results in respect of the behaviour or capacity of the "red pathway".

We don't know how the perceptual signal changes with time after a light comes on, or for that matter how long it takes for the filament to warm enough after the circuit is closed to generate visible light. As in many psychological experiments, the mere physics of what is going on (as well as the neurology) is oversimplified if it is noticed at all.

The only one of these that matters is the first. The experiment was intended to test how the perceptual signal changes after the light comes on. Schouten's original interpretation was the same as what Bill proposed in an earlier message, an exponential rise to some asymptote. What I showed Schouten (and what I published) was to show that his data implied that the subject was gaining information about the presentation at a steady uniform rate for at least the first 100-150 msec.

I'm prejudiced against information theory, because so far I haven't seen anything you can discover by using it that you can't discover some other way, except of course for the final conversion, through the log to the base two, into units of bits.

As a carpenter with a saw, hammer, and chisel might be prejudiced against screwdrivers, because one can certainly turn a screw with a chisel, and the carpenter could probably make nice furniture without a screwdriver. When I was in high school in Toronto, one of the exam question was supposed to be answered by complex trigonometric calculations. I did it by calculus, which I had learned in Scotland, but I was told that was unfair, and got no marks for the question. You certainly can solve a lot of problems with the tools you know, but if better tools for a particular job are available you don't need to. Unfortunately, it's very hard to know when a particular tool is likely to be better until you have some familiarity with it. Screwdrivers don't make very good chisels, though they have sometimes been used for that purpose!

The old-timers in electrical engineering liked to speak of amplification and signal loss in units of decibels, but I never found that very helpful either. I wouldn't really mind changing my opinion about information theory, but just being told that I'm ignorant doesn't provide the kind of motivation that gets me going.

Is that all you have gathered from my attempts at explanation? I must be an awful lot worse at communicating than I had imagined. Let's just try one comment, which applies to both decibels and information: both can be useful because they make certain operations into simple additions. Put a 10db gain amplifier in series with a 3 db loss, and you have a 7 db gain. Certainly you can get the result by multiplication, but quite often addition is easier. For information, if you get n bits from observing 1 second of a signal, and the next second is independent of the first, you get 2n bits by observing 2 seconds. That's the situation in the Schouten study.

As I recall, this whole subject arose either because of or after I made some comments about people bringing old ideas and accomplishments with them into PCT. Martin cited the Schouten experiment as an example of valid information obtained from traditional (or at least non-PCT) approaches. Apparently Martin hopes that something of value can be retained out of the pre-PCT history of science.

Hopes? No, I neither hope that it does or hope that it does not. There's no vested interest, just an understanding that everything I know about PCT argues that there is a lot of value to be gathered from the last couple of centuries of experiments in psychology. Of course there's a lot of junk. The question is to determine what is and what is not junk. I think that in this thread I have demonstrated that at least pretty well all detection and discrimination experiments done in the conventional way have produced results that can be used in PCT-science.

Rick argues against Martin not because he understands what Martin says and sees what it wrong with it, but because he wants Martin to be wrong about the value of pre-PCT science. Of course I'm only reporting one bystander's subjective impressions here, but it doesn't seem to me that anyone is really interested in FINDING OUT what the truth of the matter is.

Well, let's just ask the lurkers.

IS anyone really interested in finding out what the truth of the matter is? I made a start in [Martin Taylor 2009.02.17.11.23] at suggesting a catalogue of conditions under which non-PCT psychological results might be valid and useful. Nobody has commented on it, though I know it got sent out to the list. The "generic" diagram on which Bill and Rick were commenting comes from that message, which had the subject line "Measurement of functional components".

Martin

[From Rick Marken (2009.03.20.1430)]

Bill Powers (2009.03.20.0818 MDT)–

As Martin points out, Rick has the “answer” output above
confused with the S1 input to the perceptual function. But Martin has
omitted the input from the actually produced “Answers” to the
perceptual function

You suggest that I am being confrontational with Martin and not trying to understand him and here you are being confrontational by accusing him of making a mistake which, if you stopped being confrontational and tried to understand him, you would see is no mistake at all. Martin is simply saying that what is controlled is a perception that is based in part on observation (S1) and imagination (S2). Martin’s model of the situation is actually more consistent with your view of PCT than with mine. Recall a conversation we had some time ago when I argued that imagination was not a significant part of control (based on nothing more than my intuition, of course) and you and Dag argued the alternative. Well, Martin’s model is a perfectly legitimate representation of control of a perception that is based largely on imagination. He is saying that in the Schouten experiment, subjects imagine which button they press (that’s the imagined “Answer”) while also producing an actual press open loop .

I would not have modeled the behavior in the experiment this way because my inclination is to think of controlled variables as being based on physical variables in the environment. So I would have had no imagination loop in my model and had the “Answers”, which are the button presses, enter the perceptual function as inputs from the environment. So in my model, S2 would be the sensory representation of the environmental “Answers”. The controlled variable in my model is the relationship between the sensory observation (S1) and the sensory representation of the actual answers (S2).

The difference between Martin’s model and mine is just in what variable is controlled. So we’re back to the importance of testing for controlled variables in order to understand what went on in an experiment. According to Martin, what is controlled in the Schouten experiment is the relationship between light onset (Observation) and imagined Answer. According to my model, what is controlled is the relationship between light onset and actual Answer (the button actually pressed). Again, it seems to me that Martin’s model is perfectly consistent with yours and Dag’s ideas about control. Martin is just hypothesizing a different controlled variable than I would hypothesize. I can think of several possible ways to test whether Martin’s or my hypothesis about the controlled variable is more correct. But that’s what we need: constructive posts about how to do the appropriate experiments to test these ideas instead of just finding fault with people’s proposals.

Of course, if Martin’s model is correct and what people control in this experiment is the relationship between observed “presentations” and imagined “answers” then the relationship between IV and DV in this experiment gives a perfectly good representation of the open-loop characteristics of the system under system (assuming that the closed loop imagination connection doesn’t cause any problems). And, of course, this could apply to all experiments in conventional psychology. To the extent that it does then the points made in my “Revolution” experiment would apply only to those (possibly very few) experiments where the controlled variable(s) do not involve control of perceptions that include an imagined representation of what the subject is doing in the experiment. In other words, Martin has made a very good case for my “Revolution” paper being a huge over-reaction to what may be a very minor problem with conventional research; a problem that only arises when people are not controlling perceptual consequences of their own outputs in their imagination.

Therefore the higher system shown here is a closed-loop control system
which controls the relationship between the perceived answer and the
(observed) presentation.

But this hypothesis about the controlled variable (which is the same as mine) carries no more weight than Martin’s. Indeed, based on your ideas about how much imagination is involved in perception it seems less plausible than Martin’s. Either way, these hypotheses about what is controlled have to be tested, something that Martin, for some reason, thinks is unimportant.

You two guys are clearly irritated with each other and letting it show,
and you aren’t trying very sincerely to understand each other or explain
things to each other.

I’m certainly irritated about Martin’s belief that testing for controlled variables is not an essential component of research but I think I’m doing a better job at understanding Martin than you, as you try to conceal your irritation. I think you should just work on your own irritation before trying to referee ours.

As I recall, this whole subject arose either because of or after I made
some comments about people bringing old ideas and accomplishments with
them into PCT. Martin cited the Schouten experiment as an example of
valid information obtained from traditional (or at least non-PCT)
approaches. Apparently Martin hopes that something of value can be
retained out of the pre-PCT history of science. Rick argues against
Martin not because he understands what Martin says and sees what it wrong
with it, but because he wants Martin to be wrong about the value of
pre-PCT science.

I disagree with your analysis. I think I understand Martin as well as you do – better, perhaps, based on your comments about his diagram – and I think I have clearly and constructively explained the basis for my main disagreement with him, which is about the importance of testing for controlled variables in behavioral research.

Of course I’m only reporting one bystander’s subjective
impressions here, but it doesn’t seem to me that anyone is really
interested in FINDING OUT what the truth of the matter is. They already
know the answer they prefer and are just looking for a way to make it
come out right.

I think I am open to being convinced of the unimportance of determining what variables are being controlled in an experiment. But Martin’s diagram (which I can’t seem to include in my posts the way you did) certainly seems to argue my point for me. Martin says that the variable controlled in the Schouten experiment is a relationship between observed inputs and imagined outputs. I think the controlled variable in the experiment is more like the relationship between observed inputs and observed outputs. I think the only way to settle the matter is to do tests to determine what variable is controlled. So Martin’s diagram implicitly shows why it is important to test for controlled variables in order to know what’s going on in an experiment.

Best regards

Rick

···


Richard S. Marken PhD
rsmarken@gmail.com

[From Fred Nickols (2009.03.20.1430 PDT)]

If I understand Rick's paragraph below, isn't there something terribly important in there; namely, the controlled variable being an imagined response vs an observed one?

···

-------------- Original message ----------------------
From: Richard Marken <rsmarken@GMAIL.COM>

The difference between Martin's model and mine is just in what variable is
controlled. So we're back to the importance of testing for controlled
variables in order to understand what went on in an experiment. According to
Martin, what is controlled in the Schouten experiment is the relationship
between light onset (Observation) and imagined Answer. According to my
model, what is controlled is the relationship between light onset and actual
Answer (the button actually pressed). Again, it seems to me that Martin's
model is perfectly consistent with yours and Dag's ideas about control.
Martin is just hypothesizing a different controlled variable than I would
hypothesize. I can think of several possible ways to test whether Martin's
or my hypothesis about the controlled variable is more correct. But that's
what we need: constructive posts about how to do the appropriate experiments
to test these ideas instead of just finding fault with people's proposals.

Of course, if Martin's model is correct and what people control in this
experiment is the relationship between observed "presentations" and imagined
"answers" then the relationship between IV and DV in this experiment gives a
perfectly good representation of the open-loop characteristics of the system
under system (assuming that the closed loop imagination connection doesn't
cause any problems). And, of course, this could apply to all experiments in
conventional psychology. To the extent that it does then the points made in
my "Revolution" experiment would apply only to those (possibly very few)
experiments where the controlled variable(s) do not involve control of
perceptions that include an imagined representation of what the subject is
doing in the experiment. In other words, Martin has made a very good case
for my "Revolution" paper being a huge over-reaction to what may be a very
minor problem with conventional research; a problem that only arises when
people are not controlling perceptual consequences of their own outputs in
their imagination.

--
Regards,

Fred Nickols
Managing Partner
Distance Consulting, LLC
nickols@att.net
www.nickols.us

"Assistance at A Distance"

[Martin Taylor 2009.03.20.17.52]


[From Fred Nickols (2009.03.20.1430 PDT)]
If I understand Rick's paragraph below, isn't there something terribly important in there; namely, the controlled variable being an imagined response vs an observed one?

Yes. It’s a constructive criticism and a suggestion of a viable
alternative. I’m not at all clear how Rick’s concept would work (I’ll
address that in a response to Rick’s message to Bill), but it is
definitely a step toward coming to an understanding of where our
differences probably lie.

I do thank you very much for commenting. I wish more people would,
especially when two-person or three-person discussions seem to be going
nowhere!

Martin

···

rsmarken@GMAIL.COMnickols@att.netwww.nickols.us

[From Fred Nickols (2009.03.20.1504 PDT)]

Well, I asked because I think the conversation ties to something
I’ve been toying with lately. The notion of proximal vs distal
controlled variables. The proximal controlled variables are those we can
observe on a direct and immediate basis; they are close by in space and
time. Distal CVs are far removed in space and time. In a more
concrete vein, executives take actions now that they “hope” will
have the desired effects later. They are trying to control distal CVs but
owing to the distance between their actions and any effects on those distal
CVs, they have to imagine the effects (not to mention envision or imagine the
linkages between their direct, immediate actions and the effects on those
distal CVs at some later time and place). I’m poking around in this
because if I can’t tie PCT to anything except proximal CVs it won’t
be of much interest to anyone in the workplace.

I sure wish someone smarter than me was working on this.

Regards,

Fred Nickols

Managing Partner

Distance Consulting LLC

nickols@att.net

www.nickols.us

image00217.jpg

“Assistance at a Distance”SM

[Martin Taylor 2009.03.20.17.52]

[From Fred Nickols (2009.03.20.1430 PDT)]

If I understand Rick's paragraph below, isn't there something terribly important in there; namely, the controlled variable being an imagined response vs an observed one?

Yes. It’s a constructive criticism and a suggestion of a viable alternative.
I’m not at all clear how Rick’s concept would work (I’ll address that in a
response to Rick’s message to Bill), but it is definitely a step toward coming
to an understanding of where our differences probably lie.

I do thank you very much for commenting. I wish more people would, especially
when two-person or three-person discussions seem to be going nowhere!

Martin


[From Rick Marken (2009.03.20.1820)]

Fred Nickols (2009.03.20.1430 PDT)--

If I understand Rick's paragraph below, isn't there something terribly important
in there; namely, the controlled variable being an imagined response vs an
observed one?

I don't think there is anything terribly important in there;-) But one
thing that is not in there (important or not) is that the controlled
variable is an imagined response. Martin's model is that the
controlled variable is the perception of the state of a _relationship_
between the light (S1), which is above either the right or left
button, and the _imagined_ response (S2), which is either pressing the
right or left button. This model says that the subject doesn't care
which button is actually pressed, just which button is imagined to
have been pressed.

Best

Rick

···

--
Richard S. Marken PhD
rsmarken@gmail.com

[Martin Taylor 2009.03.20.23.30]

[From Rick Marken (2009.03.20.1820)]

Fred Nickols (2009.03.20.1430 PDT)--
If I understand Rick's paragraph below, isn't there something terribly important
in there; namely, the controlled variable being an imagined response vs an
observed one?
I don't think there is anything terribly important in there;-) But one
thing that is not in there (important or not) is that the controlled
variable is an imagined response. Martin's model is that the
controlled variable is the perception of the state of a _relationship_
between the light (S1), which is above either the right or left
button, and the _imagined_ response (S2), which is either pressing the
right or left button. This model says that the subject doesn't care
which button is actually pressed, just which button is imagined to
have been pressed.

No it doesn’t. I thought you had understood what the model says, but it
isn’t what you describe, so I guess you didn’t. I’ll quote you up to
the point where your description diverges from mine.

"Martin's model is that the controlled variable is the perception of the state of a _relationship_ between the light (S1), which is above either the right or left button, and the _imagined_" answer (S2), which is whether it was the left or the right light that was lit [NOT: *response (S2), which is either pressing the right or left button*
]. The value of S2 at the required time of the response (third bip) is the reference value for the control system that controls a perception of which button is pushed. This model says that the subject controls which button is pushed, to correspond with the answer that the experimenter has requested when either the left or the right light was lit. Control in imagination is required because the concept of a "correct" answer exists only in the subject's memory of the experimenter's instructions. Without those instructions, the subject could have a category perception of the "left" versus "right" for the lights (S1). The experimenter has provided a category repertoire for acceptable answers ("left" and "right" rather than, say "red" "green" or "blue"). One of those is S2, and the subject can vary which one to compare with the category perception of the light. That comparison is one thing the subject controls. A separate thing the subject controls is the perception of producing the overt response, whether it be by voice, semaphore, eye-blink, or button push. I call that a "response" to distinguish it from the "answer". The answer is in the subject's mind, whereas the response is intended to be perceptible to the experimenter. To make the distinction clearer, imagine that instead of "left" and "right" as the discrimination, it was "red light" or "blue light". That might make it easier to see the distinction between controlling for (a) a match between seeing a red light and the answer "red", and (b) controlling for pushing the left button if the answer is "red" and the right button if it is any other colour. I hope this may help make it clearer.
Martin

[Martin Taylor 2009.03.20.17.56]

[From Rick Marken (2009.03.20.1430)]

Bill Powers (2009.03.20.0818 MDT)–

As Martin points out, Rick has the “answer” output above
confused with the S1 input to the perceptual function. But Martin has
omitted the input from the actually produced “Answers” to the
perceptual function

Martin’s model is a perfectly legitimate representation of control of a
perception that is based largely on imagination. He is saying that in
the Schouten experiment, subjects imagine which button they press
(that’s the imagined “Answer”) while also producing an actual press
open loop .

I would not have modeled the behavior in the experiment this way
because my inclination is to think of controlled variables as being
based on physical variables in the environment. So I would have had no
imagination loop in my model and had the “Answers”, which are the
button presses, enter the perceptual function as inputs from the
environment. So in my model, S2 would be the sensory representation of
the environmental “Answers”. The controlled variable in my model is the
relationship between the sensory observation (S1) and the sensory
representation of the actual answers (S2).

Thanks, really, for making this clear. I had no idea until now that
this was what you were thinking.

Let’s examine the differences between the two concepts, which might
allow us to determine whether either is plausible, and if both are, to
identify a test of which is more plausible.

In a control loop, the essential ability is to alter the output so as
to influence the perception. Only by influencing the perception can the
error signal be reduced. I think we agree on that. How do the two
proposals compare in this respect? Here are diagrams of the two
suggestions. Correct me if I got yours wrong:

ButtonMatch_2versions.jpg

The difference between Martin’s model and mine is just in what variable
is controlled. So we’re back to the importance of testing for
controlled variables in order to understand what went on in an
experiment. According to Martin, what is controlled in the Schouten
experiment is the relationship between light onset (Observation) and
imagined Answer. According to my model, what is controlled is the
relationship between light onset and actual Answer (the button actually
pressed).

I don’t understand how your model can work without imagination. The
essence of control is the ability of the controller to vary its action
so as to influence the perception, which in this case is the match
between the category perception of the light and the perception of
which button is (being?) pushed. One button is pushed, and there is
then (and without imagination, only then) a perception of
correspondence (no error) or mismatch (error) in the upper control
system. What can the controller do if there is error? The button push
has been committed. If the subject were instead to be controlling the
button-push in imagination and then committing the actual push, I can
see how that might work, though I think it’s a less probable
organization than separating the answer from its over manifestation as
a button push. But I can’t see how you can make it work with no control
in imagination at all.

Again, it seems to me that Martin’s model is perfectly
consistent with yours and Dag’s ideas about control. Martin is just
hypothesizing a different controlled variable than I would hypothesize.
I can think of several possible ways to test whether Martin’s or my
hypothesis about the controlled variable is more correct. But that’s
what we need: constructive posts about how to do the appropriate
experiments to test these ideas instead of just finding fault with
people’s proposals.

Could you propose such an experiment? Simply to make the proposal might
make your version more easy for me to understand, since my failure to
see how it could work probably means I don’t actually understand what
you are really proposing…

In other words, Martin has made a very good case for my “Revolution”
paper being a huge over-reaction to what may be a very minor problem
with conventional research; a problem that only arises when people are
not controlling perceptual consequences of their own outputs in their
imagination.

I don’t understand this comment either. I thought, and still think,
that you made a good case in your Revolution paper; when I first read
it, I was pleased to see that what you wrote (as opposed, apparently,
to what you intended to write) had actually set out conditions under
which conventional experiments would produce valid results, these being
precisely the conditions of most detection and discrimination
experiments.

Either way, these hypotheses about what is controlled have to
be tested, something that Martin, for some reason, thinks is
unimportant.

Again, you misunderstand. I think testing for the controlled variable
is not always important. When it is important, then it is important :slight_smile:
I’m sorry if you get irritated by my belief that testing for the
controlled variable isn’t the only appropriate topic of PCT research
(which is what I believe, whereas what you said – “Martin’s belief
that testing for controlled variables is not an essential component of
research” – is not what I believe, though perhaps it is if you read
“essential component” to be equivalent to “only appropriate topic”).

I think I am open to being convinced of the unimportance of
determining what variables are being controlled in an experiment. But
Martin’s diagram (which I can’t seem to include in my posts the way you
did) certainly seems to argue my point for me. Martin says that the
variable controlled in the Schouten experiment is a relationship
between observed inputs and imagined outputs.

I suppose that if looking at the diagrams I have presented in a variety
of levels of detail don’t disabuse you of that notion, then I guess my
saying you misunderstand won’t help. In very brief: “answer” !=
“output”.

I think the controlled variable in the experiment is more like
the relationship between observed inputs and observed outputs. I think
the only way to settle the matter is to do tests to determine what
variable is controlled. So Martin’s diagram implicitly shows why it is
important to test for controlled variables in order to know what’s
going on in an experiment.

I agree that it’s important when there are plausible alternatives, and
to choose one or the other would affect the interpretation of the
experimental results. Failing the latter criterion, it’s just
intellectual curiosity. In the present case, it seems to me that both
proposals lead to exactly the same interpretation of the experimental
results. That leads me to think that yes, it would be nice to discover
which is closer to the truth (I’m sure neither is totally true), but in
this particular case it probably isn’t worth the bother.

Martin