Testing for Imagination in Experiments (was Changing the foundations of PCT)

[From Rick Marken (2010.08.22.2000)]

Bill Powers (2010.08.22.1640 MDT) to Martin Taylor –

BP: Every time PCT is used to explain an experimental finding,
I want to see it challenged and to find evidence in the experimental
results to meet the challenge. Imagining that someone else is imagining
is not a way to do that. It’s not an unreasonable guess, but it’s not the
only possible guess and no important conclusions should be drawn from it
until we find a way to test it.

I think I do have a way to test Martin’s “control in imagination” model. I’ve now written two pretty simple models which are implementations of the Marken and Martin models of a yes/no tone detection task. The models are written in Visual Basic in Excel (Visual Basic is no longer supported in Excel for the Mac so I’m being backed more and more into a Delphi corner;-). I’m attaching the spreadsheet in case anyone wants to try it out.

The models are still not complete. In particular I don’t now how to model the switch from imagination to response control mode in Martin’s model. Right now, it is assumed that a trial lasts 30 sample periods and that the switch from imagination to response control model happens after 15 samples (there is, of course, not switch in my model). We have to nail down this aspect of Martin’s model so that we can see what we should do to test for a difference between the models.

Right now, what you will see if you run the macros (by pushing the buttons) is that both models produce equivalent behavior in a yes/no detection experiment (in terms of Hits, False Alarms and probability correct (I’ve also included a d’ measure; Martin will know what that is).

Once I find out how the imagination switch is implemented we can work on ironing out other details of the models and I can explain the code and the data display in the spreadsheet in more detail. Then we can start designing an experiment that will distinguish between the two models. We can do this by changing characteristics of the model’s environment and see which changes differentially affect the behavior of the two models.

Whaddaya think?

Best

Rick

PsychExpSim.xls (60 KB)

···


Richard S. Marken PhD
rsmarken@gmail.com
www.mindreadings.com

[From Bill Powers (2010.08.23.0300 MDT)]

Rick Marken (2010.08.22.2000) --

I think I do have a way to test Martin's "control in imagination" model. I've now written two pretty simple models which are implementations of the Marken and Martin models of a yes/no tone detection task. The models are written in Visual Basic in Excel (Visual Basic is no longer supported in Excel for the Mac so I'm being backed more and more into a Delphi corner;-). I'm attaching the spreadsheet in case anyone wants to try it out.

Unfortunately, my Open Office clone of Microsoft Office does not contain the library program needed for the buttons to work. The graph shows up, but I can't interpret it because I don't know what the program is.

The use of continuous variables for the Yes and No answers is also hard to understand. How can a Yes-No response be something between Yes and No? Testing an answer would involve imagining an answer and then seeing if it has the correct relationship to the perceived presentation, tone present or not present. Since the condition being tested is binary, the relationship can be only correct or incorrect.

Is there any provision for the subject to come up with an incorrect answer? The presence of noise can make the tone seem to be present when it's not and not present when it is, so the subject will pick the wrong answer according to the statistics of the noise, with perhaps a recognition threshold. Perhaps when you explain the program more fully I will see how this works.

An aside pertaining to Martin's switch to the memory-reference signal model: It doesn't seem very likely to me that memory would be involved in this sort of control system, except perhaps in the conversion from the relationship error to a specific reference level for a spoken word being varied to correct the error, which wouldn't make much difference in the model. The main reason for introducing memory was that one level of control can't know anything about perceptions of a different level, so there has to be some mechanism by which an error in the higher system, pertaining to one type of variable, can be used to select a reference signal pertaining to a different type of variable in the lower system. If reorganization is the only basis for the selection this doesn't matter, but any systematic kind of process of learning has this problem of translating from one type of variable to another.

All this is connected with a major problem of explaining experience, which is how signals that are all alike -- trains of neural impulses -- can appear to be as different as relationships and words. Could it be that what we experience is the biochemical states of neurons? The problem of "qualia" is a very real one and I don't think we're even close to a solution.

Generally, I think you're on the right track toward a testable model, but the Yes-No problem hasn't yet, to my knowledge, been modeled right. The model should end up saying Yes or No, not emitting a number between 0 and 1. But I may not understand what you're doing yet. In fact I'm sure I don't.

Best,

Bill P.

[David Goldstein (2010.08.21.06:43 EDT)]

[From Rick Marken (2010.08.22.2000)]

Rick,

I ran the models a few times. They seem to be different around 16 seconds. Your model is smooth. Martin’s model has some discontinuities going on.

David

···

----- Original Message -----

From:
Richard Marken

To: CSGNET@LISTSERV.ILLINOIS.EDU

Sent: Sunday, August 22, 2010 11:02 PM

Subject: Testing for Imagination in Experiments (was Changing the foundations of PCT)

[From Rick Marken (2010.08.22.2000)]

Bill Powers (2010.08.22.1640 MDT) to Martin Taylor –

BP: Every time PCT is used to explain an experimental finding, I want to see it challenged and to find evidence in the experimental results to meet the challenge. Imagining that someone else is imagining is not a way to do that. It's not an unreasonable guess, but it's not the only possible guess and no important conclusions should be drawn from it until we find a way to test it.

I think I do have a way to test Martin’s “control in imagination” model. I’ve now written two pretty simple models which are implementations of the Marken and Martin models of a yes/no tone detection task. The models are written in Visual Basic in Excel (Visual Basic is no longer supported in Excel for the Mac so I’m being backed more and more into a Delphi corner;-). I’m attaching the spreadsheet in case anyone wants to try it out.

The models are still not complete. In particular I don’t now how to model the switch from imagination to response control mode in Martin’s model. Right now, it is assumed that a trial lasts 30 sample periods and that the switch from imagination to response control model happens after 15 samples (there is, of course, not switch in my model). We have to nail down this aspect of Martin’s model so that we can see what we should do to test for a difference between the models.

Right now, what you will see if you run the macros (by pushing the buttons) is that both models produce equivalent behavior in a yes/no detection experiment (in terms of Hits, False Alarms and probability correct (I’ve also included a d’ measure; Martin will know what that is).

Once I find out how the imagination switch is implemented we can work on ironing out other details of the models and I can explain the code and the data display in the spreadsheet in more detail. Then we can start designing an experiment that will distinguish between the two models. We can do this by changing characteristics of the model’s environment and see which changes differentially affect the behavior of the two models.

Whaddaya think?

Best

Rick


Richard S. Marken PhD
rsmarken@gmail.com
www.mindreadings.com

[From Rick Marken (2010.08.23.1040)]

David Goldstein (2010.08.21.06:43 EDT)

Rick,
I ran the models a few times. They seem to be different around 16 seconds.
Your model is smooth. Martin's model has some discontinuities going on.

The discontinuity in Martin's model occurs when the reference for
output switches from imagination to output mode. I have no idea why I
posted these models since it is virtually impossible to tell what's
going on unless you read the code. And even then it's probably no all
that easy. I've re-read the code (after reading Bill's questions) and
there were still some mistakes. So I will post a documented version of
a corrected version of the program that implements my model and then
you (and Bill and anyone else interested in modeling) can see if it
makes sense.

Best

Rick

···

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

[From Rick Marken (2010.08.23.1240)]

Here is the program that runs my model of the yes/no detection task.
I've commented it so that it's almost self-explanatory;-) Comment
lines are preceded by an apostrophe.

Basically, the model controls two hierarchically related variables: p2
is the higher level variable, which is the difference between two
categorical (discrete) variables; ps and pr. o is the lower level
variable; this can be thought of as the analog outputs, such as sound
waves, that are eventually perceived at the category level as "yes"
(pr = 1) or "no" (pr = 0).

The time scale is arbitrary; a trial lasts 30 cycles; if a trial is
actually 3 seconds, then the control systems are operating at 10
samples/sec.

Once we iron out this model then we can move to working on Martin's version.

Best

Rick

···

----

Dim YResp(30), NResp(30)

NTrials = Cells(1, 6) 'Number of trials in the experiment

cs = Cells(2, 6) 'Criterion magnitude for categorizing stimulus as
"Tone present" (1)

' Zero accumulators
HR = 0
FA = 0
NH = 0
NFA = 0
PC = 0
NY = 0
NN = 0

' Outer loop running test trials
For i = 1 To NTrials

' Initialize model parameters for the trial
rref = 0
pr =0
o = 0

'Determine whether stimulus on trial will contain tone (Tone = 1) or
not (Tone = 0)
If (Rnd(1) > 0.5) Then Tone = 1 Else Tone = 0 '

' Create stimulus,s, either tone plus noise (if Tone=1) or noise alone
s = (Rnd(1) - 0.5) * 20 + 10 * Tone

'Store stimulus values in spreadsheet
Cells(i + 1, 1) = Tone
Cells(i + 1, 2) = s

' Perceptual function converting stimulus, s, into categorical (1,0)
perception, ps
If (s > cs) Then ps = 1 Else ps = 0

'Count occurrence of tone/no tone trials for calculating hit, false alarm rate
If Tone = 1 Then
    NY = NY + 1
Else
    NN = NN + 1
End If

' 30 cycles of operation of two level control system controlling for
' correct response on trial
For j = 1 To 30

' Higher level system controls p2, the difference between categorical
perception of
' stimulus, ps, and categorical perception of response, pr
p2 = ps - pr

'Output of higher level system is reference for lower level output
control system
'implicit reference for p2 is 0, which is true when ps=pr
rref = rref + 0.01 * (40 * (p2) - rref)

'Output of lower level system is analog movement
o = o + 0.05 * (20 * (rref - 0) - o)

'pr is the categorically perceived output, either 1 (yes) or 0 (no)
pr = Round(o, 1)

'Accumulating output over 30 samples to be averaged over trials for display
If ps = 1 Then
    YResp(j) = YResp(j) + o
Else
    NResp(j) = NResp(j) + o
End If

Next j

'Store answer, as perceived by subject and experimenter
Cells(i + 1, 3) = pr

'Read what actually happened on trial (Tone or no tone) into variable a
a = Cells(i + 1, 1)

'Compute Hit Rate (HR), False Alarm Rate (FA) and Proportion Correct (PC)

If (a = 1) Then NH = NH + 1

If (a = 1 And pr = 1) Then
    HR = HR + 1
    PC = PC + 1
End If

If (a = 0 And pr = 1) Then FA = FA + 1

If (a = 0 And pr = 0) Then PC = PC + 1

Next i

For i = 1 To 30
    Cells(i + 1, 4) = YResp(i) / NY
    Cells(i + 1, 5) = NResp(i) / NN
Next i

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

[From Rick Marken (2010.08.23.1256)]

Correction:

'Output of lower level system is analog movement
o = o + 0.05 * (20 * (rref - 0) - o)

Should be:

'Output of lower level system is analog movement
o = o + 0.05 * (20 * (rref - o) - o)

Best

Rick

···

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

Greetings,

I have a question that relates to Bill's comments below regarding the influence of noise on recognition thresholds: Is stochastic resonance factored into PCT with respect to finding the optimal level of noise to improve the detection of desired signals?

Source: http://en.wikipedia.org/wiki/Stochastic_resonance_(sensory_neurobiology)

Cheers,
Chad

Chad Green, PMP
Program Analyst
Loudoun County Public Schools
21000 Education Court
Ashburn, VA 20148
Voice: 571-252-1486
Fax: 571-252-1633

Bill Powers <powers_w@FRONTIER.NET> 8/23/2010 5:36 AM >>>

Is there any provision for the subject to come up with an incorrect
answer? The presence of noise can make the tone seem to be present
when it's not and not present when it is, so the subject will pick
the wrong answer according to the statistics of the noise, with
perhaps a recognition threshold.

[From Rick Marken (2010.08.23.1900)]

Greetings,

I have a question that relates to Bill's comments below regarding the influence of noise on
recognition thresholds: Is stochastic resonance factored into PCT with respect to finding
the optimal level of noise to improve the detection of desired signals?

PCT really doesn't get into threshold level perception stuff very
much. It could, I suppose, and we could put stochastic resonance into
the model if it were determined to be a real phenomenon (I studied
signal detection in graduate school and never ran across this
concept).

I really can't think of many control situations where one is
controlling a perception near it's threshold. Control near the
perceptual threshold is pretty poor so I think people tend to avoid
situations where they are required to control near threshold.

not in order to do better threshold measurement but, rather, to see
whether the behavior in this simple behavioral experiment is open loop
(as it appears to be and as it is modeled in conventional psychology)
or closed-loop (as per PCT).

Best

Rick

···

On Mon, Aug 23, 2010 at 2:01 PM, Chad Green <Chad.Green@lcps.org> wrote:
from my point of view, we are modeling the tone detection experiment,
--
Richard S. Marken PhD
rsmarken@gmail.com
www.mindreadings.com

[Martin Taylor 2010.08.24.11.04]

[From Rick Marken (2010.08.23.1900)]

> From my point of view, we are modeling the tone detection experiment,
not in order to do better threshold measurement but, rather, to see
whether the behavior in this simple behavioral experiment is open loop
(as it appears to be and as it is modeled in conventional psychology)
or closed-loop (as per PCT).

Do you have an open-loop model to serve as a check against your model? Personally, I find it hard to think of how an open-loop model might look. You would have to have some intentional behaviour that was not influencing any controlled perception, and somewhere you would have to invent a way to trigger this behaviour. I think that such a model would have to be quite complicated, and I can't put one together in my mind. But to do the comparison you suggest, you would have to have some kind of model that would be accepted as such by some open-loop proponent, if you could find one who would be interested in making the comparison.

Going back to more realistic control models, are you going to comment on the third test I proposed to discriminate between your model and mine? I mean the test of a human's ability to say "Yes" and push button 1 simultaneously as the "response" to a detected signal, and to say "No" and push button 2 simultaneously as a response to an interval not containing a detected signal? I don't believe your model could deal with multiple simjultaneous correlated responses because it depends on a comparison in the relationship control unit's PIF between the detection event and the actual (perceived) response output, whereas I believe a human could do it quite easily, as could the version of my model that incorporates Bill's associative memory reference input function.

A test would not be hard to program, I think, since it doesn't depend on the discrimination perception itself. It could be done using a clearly discriminable "stimulus" such as a big displayed number for which the subject would be asked to say "Even" or "Odd" while pushing the "E" or "O" key on the keyboard. I don't know how to get the input from a microphone into a voice-recognition program, but it really doesn't matter since you can be your own subject to determine whether you can do the task at all. Only if the test on yourself seemed to be sufficiently difficult to lead to lots of errors would it be worthwhile to extend it to a real experiment by recording the timing and accuracy of the voice response and matching it with the key response.

Martin

[From Bill Powers (2010.08.14.1030 MDT)]

Martin Taylor 2010.08.24.11.04]

(re: Rick Marken (2010.08.23.1900))

MMT: Do you have an open-loop model to serve as a check against your model? Personally, I find it hard to think of how an open-loop model might look.

Here's one:

              other inputs
                   >
                   v
Stimulus -->[some function] --> ref signal --> Control system --> response

I think this represents your diagram. "Some function" may contain internal feedback loops but there is no feedback from the observed response to "some function" or to the observed "stimulus". Overall, this is an S-R system.

Best,

Bill P.

[From Rick Marken (2010.08.24.1510)]

Bill Powers (2010.08.14.1030 MDT)

Martin Taylor 2010.08.24.11.04]

(re: Rick Marken (2010.08.23.1900))

MMT: Do you have an open-loop model to serve as a check against your
model? Personally, I find it hard to think of how an open-loop model might
look.

BP: Here's one:

� � � � � � other inputs
� � � � � � � � �|
� � � � � � � � �v
Stimulus -->[some function] --> ref signal --> Control system --> response

I think this represents your diagram. "Some function" may contain internal
feedback loops but there is no feedback from the observed response to "some
function" or to the observed "stimulus". Overall, this is an S-R system.

I've been trying to avoid saying this because it seems to upset
Martin. But the diagram above is exactly how I modeled Martin's
system. Martin's comment does, however, suggest to me that I should
add the conventional open-loop model -- the general linear model; the
one I used in my thesis -- to the set of models to be tested. So I
will add the following model to the spreadsheet:

             Noise
                  >
                  v
Tone -->[some function] --> linear transfer function --> response

I'll post the code for Martin's model soon. I was hoping to get sign
off from him (and you) on the model I just posted before I went to far
down the wrong modeling path. Any comments or suggestions? One
assumption implicit in the model that may be difficult to notice is
that the stimulus ends before control of p2 begins. So the model
assumes that ps, the categorical perception of the stimulus "Tone" (1)
or "No Tone"(0), which is a component of p2 (p2=ps-pr) remains in
memory, as a 1 or 0, throughout the trial. I think that's a fair
assumption since, once you have perceived the stimulus as being in the
category "Tone' or No Tone" you are unlikely to forget which category
was perceived, even if the trial lasts several seconds. Of course,
your memory of the sensory value of the stimulus, the variable s, may
change though the trial so the category perception of s, ps, might
also change. I would prefer to have the stimulus present throughout a
trial. That would make modeling and testing the model easier. I think
so, anyway.

Best

Rick

···

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

[From Bill Powers (2010.08.25.1044 MDT)]

Rick Marken (2010.08.24.1510) –

I’ll post the code for Martin’s
model soon. I was hoping to get sign

off from him (and you) on the model I just posted before I went to
far

down the wrong modeling path. Any comments or suggestions? One

assumption implicit in the model that may be difficult to notice is

that the stimulus ends before control of p2 begins.

Why? I had been assuming, at first, that pressing the button turned off
the light. But rereading the description I realized that the light must
stay on until after a button is pressed and then after a pause goes off
to indicate that the trial is over, staying off until the start of the
next trial. If the button, and only the correct button, turned off
the light immediately, that would give the subject information about the
correctness of the response. But rereading the descriptions, I realized
that nothing was said about the correct button turning the light
off.
I was probably thinking of that button-and-lights experiment I did at the
VA hospital in the '50s and that Dick Robertson adapted to a computer.
The task then was “Press the button that turns the light off.”
The trick that subjects finally learned was to press the right button
before the next light turned on (always in a fixed sequence),
which gave a negative response-time and ran the machine’s score,
eventually, back down to zero so the machine didn’t win. A positive
response time increased the machine’s score. Might be fun to set that one
up again on a modern computer. Don’t spread the solution around – we
might want to do this experiment again with modern equipment.

Back to Martin: If the light stays on, the stimulus remains present until
the end of the trial whether the button-press is right or wrong and then
turns off after enough of a pause that the button doesn’t seem to cause
the light to go out. That’s how I interpret Martin’s description now.
Communication problems again.

I still can’t run your program because I don’t have the real Excel, and
the buttons just cause an error message.

Best,

Bill P.

[From Rick Marken (2010.08.25.1550)]

Bill Powers (2010.08.25.1044 MDT)--

Rick Marken (2010.08.24.1510) --

RM: One assumption implicit in the model that may be difficult to notice is
that the stimulus ends before control of p2 begins.

BP: Why? I had been assuming, at first, that pressing the button turned
off the light.

This is getting confusing because we (Martin and I, mainly) keep
hopping around to different experiments. This whole thing started long
ago with the Schouten reaction time experiment and I know that I
worked briefly with you on a reaction time experiment of my own. But
the discussion moved to forced choice detection, for some reason and I
think I finally got Martin to agree that what we're going to model is
a yes/no detection task. I like the yes/no task better than the forced
choice task because it seems easier to model. The other nice thing
about it is that I did a yes/no detection experiment for my thesis so
I know exactly how such an experiment is done.

So the models that I am working on are models of behavior in the
following very simple tone detection experiment: On each trial there
is a brief (say 1/2 sec) burst of noise which either contains a 500 Hz
tone or not. The trials are spaced 5 seconds apart and after each
noise burst the subject is to write either "Y" or "N"; "Y" if there
seemed to be a tone in the burst and the "N" if not. The only time
requirement is for "Y" or "N" to be written before the next burst is
presented but there is plenty of time for that. And that's all there
is to it.

There are other ways to do this simple experiment. Some people have
the noise on continuously and turn a light on for a brief time to
indicate the interval when a tone might have been added to the noise.
The response could be different too; you could have the subject press
buttons to indicate "Yes" or "No". But I'm going with the noise burst
followed by writing "Y" or "N" because that's basically how I did it
in my thesis. This might be a chance for me to show that the open loop
analysis I used in my PhD thesis was wrong: present Marken rejecting
past Marken.

Best

Rick

···

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

[From Bill Powers (2010.08.25.1710 MDT)]

Rick Marken (2010.08.25.1550) --

This is getting confusing because we (Martin and I, mainly) keep
hopping around to different experiments.

  ...

So the models that I am working on are models of behavior in the
following very simple tone detection experiment: On each trial there
is a brief (say 1/2 sec) burst of noise which either contains a 500 Hz
tone or not. The trials are spaced 5 seconds apart and after each
noise burst the subject is to write either "Y" or "N"; "Y" if there
seemed to be a tone in the burst and the "N" if not. The only time
requirement is for "Y" or "N" to be written before the next burst is
presented but there is plenty of time for that. And that's all there
is to it.

OK. So if a tone seems to have occurred in the middle of a noise, the instructions are to write the letter "N" and if it didn't, the letter "Y". I wonder what N and Y stand for.

Did you have to go back and read what you wrote to make sure you didn't reverse N and Y or Tone and No Tone? Could you equally well have been told to write T for tone and noise and N for noise alone? Or 1 and 0? T for True and F for False?

My point is that a very complex set of control systems and perceptual functions is required to carry out this task. It's not simple. How does a reference signal get sent to the system that writes one letter/numeral instead of the one that writes a different letter/numeral? How does the person know that when the tone exists, the right thing to do is to write one thing, otherwise not to write that but to write something else?

At the very least, a control system that emits an output has to monitor some sensory effect of that output. "No organism ever produces an action except for the purpose of controlling a perception." That can be made more specific by saying "No system in the brain ever emits a reference signal except for the purpose of altering the perception in the emitting control system." That could be called a basic principle of PCT.

This means that if a system is controlling the relationship between a reference signal and a stimulus of some sort, that relationship must be perceived. Both the stimulus and a perception affected by any response, real or imagined, must be perceived in order to know if the right response was emitted to establish the intended relationship between stimulus and response. The brain never just assumes that the right relationship was established.

Of course that principle might be incorrect. It may well allow for exceptions, although if there is an exception it has to be a case in which actually emitting the correct response is not considered very important, like the case of a fed-up student who marks the ovals in a questionnaire at random, or in some alternating or other systematic pattern, just to get it over with.

If the participant is trying to be sure that the actual response, when made, matches the stimulus condition correctly, there should be a way of finding out whether this is true. It could be arranged, for example, to use two adjacent keyboard keys like T and F to indicate true and false, and then occasionally print "F" when "T" is struck and "T" when "F" is struck. If the participant is monitoring the actual result of the keystroke, those would be error trials. Of course the correction has to be allowed; if the participant immediately presses a second key on error trials, we can guess that the keystroke result is being monitored (especially if the error is corrected), rather than the keystroke being emitted blindly by kinesthetic control systems.

So we don't need to assume that the so-called basic principle is true; we can find out when it is and isn't true. That's why we do experiments: to test the premises.

Best,

Bill P.

[From Rick Marken (2010.08.25.2045)]

Bill Powers (2010.08.25.1710 MDT)--

Rick Marken (2010.08.25.1550) --

RM: The trials are spaced 5 seconds apart and after each
noise burst the subject is to write either "Y" or "N"; �"Y" if there
seemed to be a tone in the burst and the "N" if not.

BP: OK. So if a tone seems to have occurred in the middle of a noise, the
instructions are to write the letter "N" and if it didn't, the letter "Y".

No, the opposite.

BP: My point is that a very complex set of control systems and perceptual
functions is required to carry out this task. It's not simple.

Neither is tracking behavior; all those hand muscles and positions to
control. But I think it's possible to model the main aspects of the
task (as per our diagrams). I sent you my computer model of the task.
I'm attaching my computer implementation of Martin's at the end of
this post, unannotated I'm afraid. The annotated one didn't get much
attention so I'm not going to waste time on it for this one, which
will probably get the same amount of attention. I guess I'm on my own
here.

How does a
reference signal get sent to the system that writes one letter/numeral
instead of the one that writes a different letter/numeral?

See lower level control system in the diagram of both models which
gets it's reference from the system controlling the relationship
between ps (perceived stimulus) and pr (perceived response) in my
model or between ps and pri (the imagined response) in Martin's.

How does the
person know that when the tone exists, the right thing to do is to write one
thing, otherwise not to write that but to write something else?

They are told what to do in the instructions, same as in a tracking
task where they are told to move the mouse up or down (or right/left
in my versions) to affect the cursor position.

At the very least, a control system that emits an output has to monitor some
sensory effect of that output.

And it does in both my model and Martin's

This means that if a system is controlling the relationship between a
reference signal and a stimulus of some sort, that relationship must be
perceived.

Not in Martin's model. Martin's model just controls the response pr,
and it works fine to account for the results of the detection task;
mine controls both the response, pr, and the relationship between pr
and ps.

Both the stimulus and a perception affected by any response, real
or imagined, must be perceived in order to know if the right response was
emitted to establish the intended relationship between stimulus and
response. The brain never just assumes that the right relationship was
established.

That's what we are testing by developing the two models, one that does
what you say above and one that doesn't.

If the participant is trying to be sure that the actual response, when made,
matches the stimulus condition correctly, there should be a way of finding
out whether this is true. It could be arranged, for example, to use two
adjacent keyboard keys like T and F to indicate true and false, and then
occasionally print "F" when "T" is struck and "T" when "F" is struck.

Both Martin's model and mine would see this as an error and correct
it. This disturbance will not test for a difference between Martin's
and my model.

So we don't need to assume that the so-called basic principle is true; we
can find out when it is and isn't true. That's why we do experiments: to
test the premises.

If the "basic principle" is the one you mention above -- that the
subject must be controlling a relationship between ps and pr (as per
my model) -- then that's exactly what I'm trying to do; test that
basic principle. In fact, Martin's model does not behave according to
that basic principle (see code below) and yet it behaves exactly as my
model does in the yes/no detection task. So we have to develop an
experiment -- a variant of the yes/no detection task -- where these
models make different predictions. I have an idea of the kinds of
experiment that will distinguish between the models but I was hoping
to get some agreement about the models (in the form of programs)
before I proposed it. I'll wait.

Best

Rick

Dim YResp(30), NResp(30)

Cells(4, 7) = "Martin Model"

NTrials = Cells(1, 6)
cs = Cells(2, 6)

HR = 0
FA = 0
NH = 0
NFA = 0
PC = 0
NY = 0
NN = 0

For i = 1 To NTrials

rref = 0#
pri = 0
pr = 0
o = 0.5

If (Rnd(1) > 0.5) Then Tone = 1 Else Tone = 0

s = (Rnd(1) - 0.5) * 20 + 10 * Tone

Cells(i + 1, 1) = Tone
Cells(i + 1, 2) = s

If (s > cs) Then ps = 1 Else ps = 0

If Tone = 1 Then
    NY = NY + 1
Else
    NN = NN + 1
End If

'Higher level loop

For j = 1 To 15

'Imagination loop

p2 = ps - pri

rref = rref + 0.01 * (80 * (p2) - rref)

pri = Round(rref, 1)

If ps = 1 Then
    YResp(j) = YResp(j) + rref
Else
    NResp(j) = NResp(j) + rref
End If

Next j

For k = 16 To 30

o = o + 0.01 * (40 * (rref - o) - o)

pr = Round(o, 1)

If ps = 1 Then
    YResp(k) = YResp(k) + o
Else
    NResp(k) = NResp(k) + o
End If

Next k

'pr = Round(pr, 1)

Cells(i + 1, 3) = pr

a = Cells(i + 1, 1)

If (a = 1) Then NH = NH + 1

If (a = 1 And pr = 1) Then
    HR = HR + 1
    PC = PC + 1
End If

If (a = 0 And pr = 1) Then FA = FA + 1

If (a = 0 And pr = 0) Then PC = PC + 1

Next i

For i = 1 To 30
    Cells(i + 1, 4) = YResp(i) / NY
    Cells(i + 1, 5) = NResp(i) / NN
Next i

Cells(5, 6) = HR / NH
Cells(6, 6) = FA / (NTrials - NH)
Cells(7, 6) = PC / NTrials

···

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

[From Bill Powers (2010.08.26.0815 MDT)]

Rick Marken (2010.08.25.2045) --

> BP earlier: OK. So if a tone seems to have occurred in the middle of a noise, the instructions are to write the letter "N" and if it didn't, the letter "Y".

RM: No, the opposite.

I can't seem to learn not to pull these subtle tricks on people. They pass unnoticed. Or maybe they're not subtle enough and are ignored on purpose. The next paragraph I wrote was

BP earlier: Did you have to go back and read what you wrote to make sure you didn't reverse N and Y or Tone and No Tone? Could you equally well have been told to write T for tone and noise and N for noise alone? Or 1 and 0? T for True and F for False?

BP: My proposal is that some system inside you always checks the relationship between the response and the stimulus even when you're not aware of doing it. I disturbed (stimulus) a relationship perception without in the least disturbing what you actually said; you automatically corrected it (response) without even noticing it as a disturbance, and certainly not as a deliberate one (at a still higher level). So the next paragraph made no sense to you and you just skipped past it. Is that what happened?

The question remains, how did you know that what I wrote was wrong?

...

RM: If the "basic principle" is the one you mention above -- that the
subject must be controlling a relationship between ps and pr (as per
my model) -- then that's exactly what I'm trying to do; test that
basic principle. In fact, Martin's model does not behave according to
that basic principle (see code below) and yet it behaves exactly as my
model does in the yes/no detection task. So we have to develop an
experiment -- a variant of the yes/no detection task -- where these
models make different predictions.

Yes, that's the problem. But we have progress anyway. You say there are two possible models, one sensing the real relationship and one not doing that. Yet they both behave the same way. This means that if anyone ever did, it's no longer possible to accept that Martin's model as the only possible one, and we have to put his conclusions -- and ours -- on hold until we find a distinguishing test.

I have an idea of the kinds of
experiment that will distinguish between the models but I was hoping
to get some agreement about the models (in the form of programs)
before I proposed it. I'll wait.

It's going to take me a while to work out what your program does -- the references to cells don't help since I can't run your program, and now you're even leaving out the comments, which makes it harder. I wonder how many other people are having trouble understanding your model (those who don't know programming will have problems, for sure). I might have to try a Delphi version just to see it running.

It would help if, instead of using just cell references, you could give the cells and the other variables meaningful names, which would allow translating into another programming language much easier. Does your version of Basic limit you to just a few letters as variable names?

Best,

Bill P.

[From Rick Marken (2010.08.26.1450)]

Bill Powers (2010.08.26.0815 MDT)--

Rick Marken (2010.08.25.2045) --

> BP earlier: OK. So if a tone seems to have occurred in the middle of a
> noise, the instructions are to write the letter "N" and if it didn't, the
> letter "Y".

RM: No, the opposite.

I can't seem to learn not to pull these subtle tricks on people. They pass
unnoticed. Or maybe they're not subtle enough and are ignored on purpose.
The next paragraph I wrote was

I got it after I read over the reply. Good trick but I didn't notice
at first, I think, because it sounded like you were purposefully
misunderstanding me (which you were, kinda) but I thought you were
doing it to criticize my model in some obtuse way. The disturbance was
at a level much higher than Y N. In any case, I solved the problem by
ignoring it;-)

RM:... So we have to develop an experiment -- a variant of the yes/no
detection task -- where these models make different predictions.

BP: Yes, that's the problem. But we have progress anyway. You say there
are two possible models, one sensing the real relationship and one not doing
that. Yet they both behave the same way. This means that if anyone ever
did, it's no longer possible to accept that Martin's model as the only possible
one, and we have to put his conclusions -- and ours -- on hold until we find a
distinguishing test.

Coming soon to a listserv near you.

It's going to take me a while to work out what your program does -- the
references to cells don't help since I can't run your program, and now
you're even leaving out the comments, which makes it harder. I wonder how
many other people are having trouble understanding your model (those who
don't know programming will have problems, for sure). I might have to try a
Delphi version just to see it running.

I thought the first one was pretty well documented. But I will try again.

It would help if, instead of using just cell references, you could give the
cells and the other variables meaningful names, which would allow
translating into another programming language much easier. Does your version
of Basic limit you to just a few letters as variable names?

No, I can use meaningful names. But I'll try commenting mode
completely first. I'll start with Martin's model this time.

Best

Rick

···

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

[From Rick Marken (2010.08.28.1200)]

Rick Marken (2010.08.26.1450)

BP: It would help if, instead of using just cell references, you could give the
cells and the other variables meaningful names, which would allow
translating into another programming language much easier. Does your version
of Basic limit you to just a few letters as variable names?

RM: No, I can use meaningful names. But I'll try commenting mode
completely first. I'll start with Martin's model this time.

On second thought I'll just finish up the spreadsheet versions of the
2 (possibly 3) models and once I've got everything working I'll post
it. I've already gotten versions of the Marken and Martin models
working with disturbances applied to the responses (as you suggested
in an earlier post) and they both act the same. And best of all they
both appear to be controlling the relationship between S and R (since
that variable is kept constant by both models). I will soon have a
version of the program that does a test that will distinguish between
the models. I think it's coming along rather nicely though I don't
have much time to work on it.

By the way, is it Open Office that doesn't run the Visual Basic
Macros? I'll check.

Best

Rick

···

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

[From Rick Marken (2010.08.28.1420)]

Well, that's a drag. OpenOffice does have VisualBasic but you can't
just read in my spreadsheet and run the macros. There is some kind of
incompatibility that I don't have the time or inclination to try to
overcome, at least not at the moment. So I'll just storm ahead with
Excel 2003 and those who have that version of Excel will be able to
follow this. If Martin has that version of Excel then I think we're in
business. Can you run Excel 2003 Martin?

Best

Rick

···

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

[Martin Taylor 2010.08.28. 23.37]

···

On 2010/08/24 12:36 PM, Bill Powers wrote:

[From Bill Powers (2010.08.14.1030 MDT)]

Martin Taylor 2010.08.24.11.04]

(re: Rick Marken (2010.08.23.1900))

MMT: Do you have an open-loop model to serve as a check against your model? Personally, I find it hard to think of how an open-loop model might look.

Here's one:

             other inputs
                  >
                  v
Stimulus -->[some function] --> ref signal --> Control system --> response

I think this represents your diagram. "Some function" may contain internal feedback loops but there is no feedback from the observed response to "some function" or to the observed "stimulus". Overall, this is an S-R system.

I don't see much relationship between this diagram and my model. Could you explain where in it the loop representing the dialogue between subject and experimenter is shown?

In your diagram, what are "other inputs" and "ref signal"?

Martin