Testing for Control In Experiments (was Re: Control of and in imagination)

[rsmarken]

[From Rick Marken (2010.07.21.2120)]

I'm starting this new tread in reply to two posts from Martin in which
he does propose some tests to compare our models of what a subject is
actually doing in an experiment.

�Martin Taylor (2010.07.21.10.33)

MT: OK. We have a disagreement as to what variables are controlled.

That's correct.

I have argued (and occasionally you have agreed) that the subject can perceive
which interval contained the 500 Hz tone without ever intending to produce a
response, and that the subject produces a response ONLY in order that the
experimenter can know which answer the subject decided on.

I don't think I have ever agreed with this. I think the subject
produces a response only in order to control a relationship between S
and R. The subject can also do this in imagination but then he
wouldn't be doing (controlling) what he has been asked to do.

You now seem to
be reverting to your earlier position that the subject can control the
relationship between stimulus and answer only if she emits (or imagines
emitting) a specific kind of response.

I don't see how that contradicts what you said above.

One way of testing the difference between these would be for the
experimenter to ask: "If you hear the tone in the third interval, what would
you say to me", hoping that the subject would respond "three". If the
subject does not say something that includes "three", there would be an
error in the experimenter's control of the perception of the subject's
knowledge of how to relate the possible answers to the overt output, and the
experimenter would act to correct that error.

I don't really see the difference you see and I don't see how this
tests whatever difference you see anyway. If the subject doesn't
relatively consistently do what he's been asked to do then I would
expect the experimenter to intervene just as you say above. What is it
that you predict that's different?

I claim that the relationship
between the interval and the imagined answer is controlled, and that the
output of this control system serves in the normal way as a reference value
for the control systems that eventually output "R".

Yes, that's what I think your model is. The difference between that
model and mine is that I consider a perception of "R" to be a part of
the perception controlled by the subject. The perception controlled in
my model is the relationship between the interval and the perceived
(not imagined) answer, "R".

The control system whose
reference input is the desired answer value is the one that perceives what
overt response was emitted, and is the one responsible for any error
correction that the experimental setup permits ("Mr. Experimenter, I pushed
the wrong button just then -- could I please correct my answer").

Right. So we can't tell the difference between your model and mine by
disturbing the response. Both your model and mine predict that a
disturbance to R will be resisted. I think the difference between our
models is the difference in controlled variables. In your model the
main controlled variable is the relationship between perceived tone
interval, pI, and imagined answer, iA; let's call your controlled
variable pI-iA. In my model the main controlled variable is the
relationship between the perceived tone interval, pI, and the
perceived answer, pA.; let's call my controlled variable pI-pA. So my
model says that pI-pA is controlled; yours says that pI-pA is _not_
controlled (though pA is). So I think we have to set up a force choice
tone detection experiment where we can disturb pI-pA. Your model
predicts that a disturbance to pI-pA will lead to no corrective
action; my model predicts that it will. So how about thinking of a
forced choice experiment that will make this kind of test possible?

But let's see what you come up with in this second post.

Martin Taylor (2010.07.21.11.28)

RM: So how about an experiment to test your theory versus my theory of the
forced choice detection experiment? According to my theory the subject is
controlling a perception which is something like the relationship between the
perceived tone interval and the perceived report of the tone interval. According to
you theory the subject is controlling only in imagination; the report is a controlled
output that is generated after control in imagination occurs.

MT: That's a fair statement of the difference that I hope I understand.

Great!

How about a test such as the thought experiment I suggested a while back?
Here is the skeleton of that proposal.

The subject goes through all the normal tests to ensure that s/he knows what >responses should go with what perceptions of the disturbance (a tone presented >barely audibly in one of four intervals defined by a burst of noise -- or something >equivalent that could be programmed on a computer screen, since I don't have >access to any auditory testing equipment). The subject is asked to listen (or look) >carefully one each trial, and to remember which interval contained the significant >event. After randomly selected trials, the subject is asked to report not which interval >held the significant event on that trial, but which interval it had been on the preceding >trial. The form of the report differs from occasion to occasion, sometimes being a >button press, sometimes a vocalization, sometimes a written numeral.

If the subject can provide a correct response more often than not, the result would >demonstrate two things, that to control the relationship does not require an overt >response, and that to control the relationship does not require the subject to imagine >making any specific kind of response.

No, I don't think this does it. It doesn't test the difference between
our models. I don't think it demonstrates what you say it does either.
But the things that you say it demonstrates are not really in dispute.
No one has argued that an overt response is required to control a
relationship (or any other kind of perception); the relationship
perception, if controlled in imagination, as in your model, requires
no overt response. I don't need to turn the steering wheel (an overt
response) in order to keep my car in its lane in my imagination; but
I do need to turn it in order to control the perception of the
relationship of the car to its lane. In every experiment I know of the
subject is asked to control a perception, not an imagination.

RM So how about proposing a test of your theory versus mine? That would be
better than me doing it, I think.

Does that suffice?

I'm afraid not. To test the difference between our theories of
behavior in the forced choice detection task I really think you have
test to see whether the subject is controlling pI-pA or not.

If not, how could it be improved or replaced by something different?

Again, I think it would be best if you tried to think of a test on
your own. I'm not sure I know a particularly good way to test this
myself. But I think it would be more convincing(to you certainly) if
you could come up with a test that would be acceptable to _both_ of
us.

Best

Rick

···

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

[MartinT]

[Martin Taylor 2010.07.22.00.42]

[From Rick Marken (2010.07.21.2120)]

I have argued (and occasionally you have agreed) that the subject can perceive
which interval contained the 500 Hz tone without ever intending to produce a
response, and that the subject produces a response ONLY in order that the
experimenter can know which answer the subject decided on.

I don't think I have ever agreed with this.

You said [From Rick Marken (2010.07.14.1812)]:
"A person can do the forced choice detection task in imagination

just fine. But if the subject does the experiment entirely in
imagination it will be impossible for the experimenter to
determine which of the intervals the subject selected as the one
containing the tone."

I asked you how I misunderstood this to mean that the subject can do

the task in imagination just fine, but it is only so that the
experimenter knows the chosen answer that the subject makes the
response. You did not tell me that I had misunderstood, so I think
it was wuite fair for me to believe you meant just what you said,
and that we agreed: The subject controls the relationship in
imagination, and the output of this control is what the subject
communicates to the experimenter.

Now you either should explain how I misunderstood the above quote,

or why you are contradicting yourself when you say

I think the subject
produces a response only in order to control a relationship between S
and R. The subject can also do this in imagination but then he
wouldn't be doing (controlling) what he has been asked to do.

One way of testing the difference between these would be for the
experimenter to ask: "If you hear the tone in the third interval, what would
you say to me", hoping that the subject would respond "three". If the
subject does not say something that includes "three", there would be an
error in the experimenter's control of the perception of the subject's
knowledge of how to relate the possible answers to the overt output, and the
experimenter would act to correct that error.

I don't really see the difference you see and I don't see how this
tests whatever difference you see anyway. If the subject doesn't
relatively consistently do what he's been asked to do then I would
expect the experimenter to intervene just as you say above. What is it
that you predict that's different?

I claim that the relationship
between the interval and the imagined answer is controlled, and that the
output of this control system serves in the normal way as a reference value
for the control systems that eventually output "R".

Yes, that's what I think your model is. The difference between that
model and mine is that I consider a perception of "R" to be a part of
the perception controlled by the subject. The perception controlled in
my model is the relationship between the interval and the perceived
(not imagined) answer, "R".

And if the subject makes no overt response "R", is she then not

controlling any relationship involving the stimulus?

I agree that the perception of "R" is controlled by the subject --

among many other perceptions controlled by the subject.

…I think the difference between our

models is the difference in controlled variables. In your model the
main controlled variable is the relationship between perceived tone
interval, pI, and imagined answer, iA; let's call your controlled
variable pI-iA. In my model the main controlled variable is the
relationship between the perceived tone interval, pI, and the
perceived answer, pA.; let's call my controlled variable pI-pA. So my
model says that pI-pA is controlled; yours says that pI-pA is _not_
controlled (though pA is). So I think we have to set up a force choice
tone detection experiment where we can disturb pI-pA.

How could you do that when pA is a controlled variable? And if you

disturb pA, how would you distinguish the control of pA (my model)
from control of pI-pA (your model)? I think my proposal does achieve
that.

Your model
predicts that a disturbance to pI-pA will lead to no corrective
action; my model predicts that it will.

How could you disturb pI-pA without disturbing pA, since pI is a

given? The logic is not there.

So how about thinking of a
forced choice experiment that will make this kind of test possible?
But let's see what you come up with in this second post.
Martin Taylor (2010.07.21.11.28)

RM: So how about an experiment to test your theory versus my theory of the
forced choice detection experiment? According to my theory the subject is
controlling a perception which is something like the relationship between the
perceived tone interval and the perceived report of the tone interval. According to
you theory the subject is controlling only in imagination; the report is a controlled
output that is generated after control in imagination occurs.

MT: That's a fair statement of the difference that I hope I understand.

Great!

How about a test such as the thought experiment I suggested a while back?
Here is the skeleton of that proposal.

The subject goes through all the normal tests to ensure that s/he knows what >responses should go with what perceptions of the disturbance (a tone presented >barely audibly in one of four intervals defined by a burst of noise -- or something >equivalent that could be programmed on a computer screen, since I don't have >access to any auditory testing equipment). The subject is asked to listen (or look) >carefully one each trial, and to remember which interval contained the significant >event. After randomly selected trials, the subject is asked to report not which interval >held the significant event on that trial, but which interval it had been on the preceding >trial. The form of the report differs from occasion to occasion, sometimes being a >button press, sometimes a vocalization, sometimes a written numeral.

If the subject can provide a correct response more often than not, the result would >demonstrate two things, that to control the relationship does not require an overt >response, and that to control the relationship does not require the subject to imagine >making any specific kind of response.

No, I don't think this does it. It doesn't test the difference between
our models.

Why not? It separates the subject's control of the relationship

between stimulus and answer from the emission of the response, and
tests whether the subject can control that relationship without
knowing either whether a response will later be requires, or in what
form the response will be requested. That seems to me to be a pretty
solid test of whether what is controlled is a relationship between
the stimulus and the actual response of a relationship between the
stimulus and an imagined answer that can optionally be used as a
reference value for an arbitrary later response.

I don't think it demonstrates what you say it does either.
But the things that you say it demonstrates are not really in dispute.
No one has argued that an overt response is required to control a
relationship (or any other kind of perception); the relationship
perception, if controlled in imagination, as in your model, requires
no overt response.

Correct. As you seemed to agree earlier, only the _experimenter_

(your emphasis) needs the response. The subject needs it only if she
is controlling to please the experimenter by emitting meaningful
responses.

I don't need to turn the steering wheel (an overt
response) in order to keep my car in its lane in my imagination; but
I do need to turn it in order to control the perception of the
relationship of the car to its lane.

Correct. The physical car needs the turn of a physical wheel in

order to stay in its lane.

In every experiment I know of the
subject is asked to control a perception, not an imagination.

Yes, here we have it again, the distinction Bill thinks is not a

difference. I do not claim the subject is controlling an
imagination. I am claiming that the subject is controlling the
relationship perception IN imagination. I have been trying to make
clear that distinction – indeed, it was the title of this thread
before you changed it – and here we have the importance of the
distinction made plain. The subject varies the imagined answer until
the controlled perception (the relationship) achieves its reference
value. The output of the relationship control system is the
variation of the imagined answer. The subject is NOT controlling her
output (the imagined answer). The subject is controlling her input
(the state of the relationship perception).

Yes, in any experiment, the experimenter asks the subject to make

overt responses, and if the subject is controlling to perceive the
experimenter to be satisfied, the subject will indeed make
responses, and will try to do the assigned task as best she can in
making the responses.

RM So how about proposing a test of your theory versus mine? That would be
better than me doing it, I think.

Does that suffice?

I'm afraid not. To test the difference between our theories of
behavior in the forced choice detection task I really think you have
test to see whether the subject is controlling pI-pA or not.

I think my suggested experiment does that, by removing the

possibility of controlling pI-pA since pA cannot exist when pI does,
and in the strictest form of the experiment, the subject does not
even know what kind of “A” will be requested if one is requested at
all.

If you disagree, how about saying why you disagree, rather than just

asserting that you do?

Martin

[rsmarken]

[From Rick Marken (2010.07.22.1010)]

Martin Taylor (2010.07.22.00.42)--

RM: The perception controlled in my model is the relationship between
the interval and the perceived (not imagined) answer, "R".

MT: And if the subject makes no overt response "R", is she then not c
controlling any relationship involving the stimulus?

I am trying to model what a subject actually _does_ in an experiment.
If the subject just sits there doing nothing (makes no overt response)
then I have no idea what they are doing.

RM: ...I think the difference between our
models is the difference in controlled variables... In my model the main
controlled variable is the relationship between the perceived tone interval,
pI, and the perceived answer, pA.; let's call my controlled variable pI-pA.
So my model says that pI-pA is controlled; yours says that pI-pA is _not_
controlled (though pA is). So I think we have to set up a force choice
tone detection experiment where we can disturb pI-pA.

How could you do that when pA is a controlled variable?

By adding a disturbance to pl-pA: (pl-pA)+ d.

And if you disturb pA, how would you distinguish the control of pA (my model)
from control of pI-pA (your model)?

Again, you have to disturb pI-pA, not just pA.

I think my proposal does achieve that.

Your model achieves what? Your model controls pA but it does not
control pI-pA. Mine controls both but I don't show control of pA
explicitly (kind of like I don't show control of mouse position, m,
explicitly in my tracking models; the tracking model takes control of
m "for granted" and just explicitly controls t - c, the difference
between target and cursor, where c = d + m).

RM: Your model
predicts that a disturbance to pI-pA will lead to no corrective
action; my model predicts that it will.

MT: How could you disturb pI-pA without disturbing pA, since pI is a given?

It doesn't really matter if pA is also disturbed. The important thing
is to disturb pl-pA. If the subject is controlling that variable (as
per my model) then there will be resistance to that disturbance. If
the subject is controlling only pA, as per your model, then there
will be a correction to pA but not to pI-pA.

MT: How about a test such as the thought experiment I suggested a while back?
Here is the skeleton of that proposal...
... The subject is asked to listen (or look) carefully one each trial, and
to remember which interval contained the significant event...

RM: No, I don't think this does it. It doesn't test the difference between
our models.

MT: Why not? It separates the subject's control of the relationship between
stimulus and answer from the emission of the response, and tests whether
the subject can control that relationship without knowing either whether a
response will later be requires, or in what form the response will be requested.

That's not the difference between our models. The difference between
our models is that in my model the subject's behavior in a forced
choice detection experiment involves control of pI-pA; in your model,
the subject's behavior involves control only of pA. My model (with
added memory) would behave the same as yours in your experiment.
That's why your experiment doesn't test the difference between our
models. At least, that's the way I see it by just analyzing it in my
head, without running any computer simulations of the models. If you
still think your experiment distinguishes our models then you can
convince me by writing a computer version of both of our models and
showing me that they behave differently in the experiment.

RM: In every experiment I know of the
subject is asked to control a perception, not an imagination.

MT: Yes, here we have it again, the distinction Bill thinks is not a difference.
I do not claim the subject is controlling an imagination. I am claiming that the
subject is controlling the relationship perception IN imagination.

Yes, I know that and Bill knows it. That's your model. Both Bill and I
agree that there may be some control in imagination on each trial of
an experiment. So we could include an imagination component in our
model as well. Where your model differs from ours is in the "fate" of
the response. In your model, after the subject has matched pI to iA in
imagination, out pops the (controlled) response. In our model, after
the subject match pI to iA in imagination, the (controlled) response
is made in order to control the perception, pI-pA.

MT: Does that suffice?

RM: I'm afraid not. To test the difference between our theories of
behavior in the forced choice detection task I really think you have
test to see whether the subject is controlling pI-pA or not.

MT: I think my suggested experiment does that, by removing the possibility of
controlling pI-pA since pA cannot exist when pI does, and in the strictest form
of the experiment, the subject does not even know what kind of "A" will be
requested if one is requested at all.

If you disagree, how about saying why you disagree, rather than just asserting
that you do?

You can't test to see if pI-pA is controlled by having a person
control a different variable. Your experiment, as I understand it,
asks a person to control the relationship between a remembered value
of pI (mpI) and pA, where the A required can be one of several types
specified on a trial. I presume your model of this experiment would be
analogous to your model of the regular force choice experiment: the
subject is controlling mpI-iA in imagination and then producing a
controlled response, "A". My model would be the same as the simple
forced choice model as well: the subject is controlling mpI-pA instead
of just pI-pA. So your experiment really doesn't seem to test the
fundamental difference between our models, which is that in your model
pI-pA is not controlled and in mine it is. So I suggest that you give
it another try, keeping in mind that we want to disturb pI-pA to see
if it's controlled.

Best

Rick

···

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

[MartinT]

[Martin Taylor 2010.07.22.23.41]

[From Rick Marken (2010.07.22.1010)]
Martin Taylor (2010.07.22.00.42)--

RM: The perception controlled in my model is the relationship between
the interval and the perceived (not imagined) answer, "R".

MT: And if the subject makes no overt response "R", is she then not c
controlling any relationship involving the stimulus?

I am trying to model what a subject actually _does_ in an experiment.
If the subject just sits there doing nothing (makes no overt response)
then I have no idea what they are doing.

RM: ...I think the difference between our
models is the difference in controlled variables... In my model the main
controlled variable is the relationship between the perceived tone interval,
pI, and the perceived answer, pA.; let's call my controlled variable pI-pA.
So my model says that pI-pA is controlled; yours says that pI-pA is _not_
controlled (though pA is). So I think we have to set up a force choice
tone detection experiment where we can disturb pI-pA.

How could you do that when pA is a controlled variable?

By adding a disturbance to pl-pA: (pl-pA)+ d.

You really don't understand, do you?

Let's go through it again.

1. pI is fixed. It's the disturbance to the relationship perception.

2. pA is controlled. It's the perception of the button that is

pushed or the vocalization that is made, and in any model, that
perception is controlled on its own, just as the orientation of the
steering wheel is controlled on its own when you are controlling for
keeping the car in its lane. If control is good, you can’t
effectively disturb pA. You can apply a disturbing influence, but
doing so won’t change pA if the subject is not overwhelmed by the
force of the disturbance.

3. If you can't change pA and you can't change pI, how do you

disturb pI-pA?

And if you disturb pA, how would you distinguish the control of pA (my model)
from control of pI-pA (your model)?

Again, you have to disturb pI-pA, not just pA.

Now it becomes a question of logic. pI is fixed. There's nothing you

can do about that. You want to disturb pI-pA, not just pA, and you
want to do this without changing pI. How, logically, can that be
possible?

I think my proposal does achieve that.

Your model achieves what? Your model controls pA but it does not
control pI-pA.

That is indeed the difference between our conceptions. I certainly

do NOT control pI-pA. To do so would make no sense, so far as I can
see. Mine controls pI-iA in your terminology, the resulting value of
iA being fed as rA (reference value for pA) to the system that
controls pA. Your model has to have a system that controls pA, too,
and it needs to be fed a reference value. Where does that reference
value come from?

To me, your model sounds like a model for the car-in-lane control

structure in which the perceived steering wheel orientation is an
input to the perception of where the car is in the lane. In most
models of controlling the car in its lane, the perception of the
relationship between the car position and the lane boundaries is
purely visual, and the output of the relationship control system
provides a reference value for the control system that controls its
perception of the steering wheel orientation. That’s the structure I
use in my model for the psychophysical experiment.

Mine controls both but I don't show control of pA
explicitly (kind of like I don't show control of mouse position, m,
explicitly in my tracking models; the tracking model takes control of
m "for granted" and just explicitly controls t - c, the difference
between target and cursor, where c = d + m).

Let's consider this example, because it is pertinent. Yes, the mouse

position is controlled. It is a location on the physical desk, not a
location on the screen. The cursor is a location on the screen, so
the expression c - d + m is a category error. Control of the mouse
position has a side-effect, which is that the position of the cursor
on the screen changes. But note that it is a side-effect, in that
the controller of mouse position knows nothing of the cursor. You
could disturb the mouse position control system by pushing the mouse
physically. You can’t do that with the cursor. If you want to
disturb the cursor position, you have to do that in the electronics.

![pursuit_tracking.jpg|428x333](upload://nGhfmlanYhHqvuXQl8j4nH2u7vt.jpeg)

RM: Your model
predicts that a disturbance to pI-pA will lead to no corrective
action; my model predicts that it will.

MT: How could you disturb pI-pA without disturbing pA, since pI is a given?

It doesn't really matter if pA is also disturbed. The important thing
is to disturb pl-pA. If the subject is controlling that variable (as
per my model) then there will be resistance to that disturbance. If
the subject is controlling only pA, as per your model, then there
will be a correction to pA but not to pI-pA.

Earlier, you recognized that my model controls pI-iA. Today you say

my model says the subject is controlling only pA. Why?

MT: How about a test such as the thought experiment I suggested a while back?
Here is the skeleton of that proposal...
... The subject is asked to listen (or look) carefully one each trial, and
to remember which interval contained the significant event...

RM: No, I don't think this does it. It doesn't test the difference between
our models.

MT: Why not? It separates the subject's control of the relationship between
stimulus and answer from the emission of the response, and tests whether
the subject can control that relationship without knowing either whether a
response will later be requires, or in what form the response will be requested.

That's not the difference between our models. The difference between
our models is that in my model the subject's behavior in a forced
choice detection experiment involves control of pI-pA; in your model,
the subject's behavior involves control only of pA.

I'd prefer it if you went back to the message in which you said

(correctly) that in my model the subject’s behaviour involves
control of pI-iA as well as of pA.

My model (with
added memory) would behave the same as yours in your experiment.

How is that? The experiment separates the perception of which

interval held the signal from the occurrence of the response. Are
you saying that in your model with memory, the subject holds the
waveforms of the four intervals in mind for one trial while
listening to the next, so that the remembered sound sequence can be
compared on demand with a response if one is demanded? That would
seem to demand an awful lot of the memory, to generate some level of
perception that just might be asked to enter into a controlled
relationship, while other similar sequences are being perceived on
the same path.

Your model has suddenly become awfully complicated!! It seems a very

desparate rearguard action to preserve some vestige of the original
model. I don’t buy it. But I suppose anything is possible, if you
imagine hard enough.

That's why your experiment doesn't test the difference between our
models. At least, that's the way I see it by just analyzing it in my
head, without running any computer simulations of the models. If you
still think your experiment distinguishes our models then you can
convince me by writing a computer version of both of our models and
showing me that they behave differently in the experiment.

If one were to do that, you would have to specify how and where the

memory of the sound sequence would be stored before being supplied
on demand to the relationship control system that determines which
interval label matches the interval with the tone, and you would
have to specify what control systems control what perceptions to
retrieve the sound sequence, and the mechanism whereby those
memory-switching systems work.

RM: In every experiment I know of the
subject is asked to control a perception, not an imagination.

MT: Yes, here we have it again, the distinction Bill thinks is not a difference.
I do not claim the subject is controlling an imagination. I am claiming that the
subject is controlling the relationship perception IN imagination.

Yes, I know that and Bill knows it.

OK, so why do you claim I don't know it?

…

You can't test to see if pI-pA is controlled by having a person
control a different variable. Your experiment, as I understand it,
asks a person to control the relationship between a remembered value
of pI (mpI) and pA,

No. my model doesn't do that. It required the person to control pA

to a reference value that is a remembered value of the final output
of the pI-iA relationship control system. In other words, to control
pA with a reference value that is equivalent to a number 1…4. It’s
your model that requires remembering pI. And that’s what I find
makes your model so complex, storing pI and retrieving it after
another value of pI has been processed, but only on request.
Frankly, it’s totally implausible – reminds me of Ptolemy’s
epicycles upon epicycles. They work pretty well, but it takes a
strange kind of background conceptual environment to make them
plausible.

Martin

···

On 2010/07/22 1:08 PM, Richard Marken wrote:

[MartinT]

[Martin Taylor 2010.07.23.10.57]

Here's another variant on the experiment to distinguish between control of the relationship pI-iA and pI-pA (relationship between perceived stimulus and imagined answer as opposed to perceived stimulus and perceived actual answer).

In this experiment, the subject is again asked to identify which of four intervals contained a "signal". A set of four intervals constitutes one trial.

On each trial except the first, the subject makes a response. There are two buttons, labelled respectively "Same" and "Different". The subject is asked to press "Same" if the interval containing the signal was the same on this trial as it was on the last, and "Different" if it was not. Using Rick's notation, the only possible values of p(A) are "Same" and "Different", whereas the possible values for the interval to be detected are {1, 2, 3, 4}.

In this variant, there are three important relationship perceptions,
(1) a controlled perception of the relationship (Ra = pI-iA) between the stimulus and the answer for a single trial,
(2) an uncontrolled perception of the relation between the intervals chosen in the last two trials (pb = IA1 XOR IA2, where IA is the value of iA when the first relationship control has come to its settled value and been switched into the following circuitry), and
(3) a controlled perception of the relation between pb and the response answer (imagined or physically executed).

If the subject in the conventional experiment is controlling pI-pA, then in this experiment the subject would be controlling pb-pA, but could never be controlling be pI-(any overt response). (Of course, I would expect the subject to be controlling pb-iA rather than pb-pA, but this experiment doesn't test for that).

The point is that the overt response has no relation with the interval in which the subject detects the signal, and cannot even be mapped onto the set of possible intervals, so that control of the critical first relationship (Ra) that is the object of the study cannot be pI-pA. If the subject can perform the task at all, the perception controlled in Ra must be pI-iA.

The presentation and interval-determination requirement of the primary relationship control are the same in this experiment as in the classic "tell me which interval contained the signal" experiment, so it would be a reasonable assumption that the relationship control has the same input variables in the two cases.

Since you didn't like my first cut, and presented a workable but extraordinary baroque alternative explanation of the results, maybe you will agree this version would discriminate whether subjects determine the interval containing the signal by controlling pI-iA or pi-pA. -- but I'm not holding my breath :-).

Or maybe you have an even more baroque alternative explanation to offer?

Martin

[rsmarken]

[From Rick Marken (2010.07.23.0900)]

Martin Taylor (2010.07.22.23.41)

Rick Marken (2010.07.22.1010)-- 

RM: ...So I think we have to set up a force choice
tone detection experiment where we can disturb pI-pA.

MT: How could you do that when pA is a controlled variable?

RM: By adding a disturbance to pl-pA: (pl-pA)+ d.

1. pI is fixed. It's the disturbance to the relationship perception.

Actually, the disturbance is I, the interval in which the tone occurs. It is the independent variable in the experiment to it can be manipulated. So pI is not necessarily fixed; it can presumably be varied by varying I.

2. pA is controlled.

Yes.

3. If you can't change pA and you can't change pI, how do you

disturb pI-pA?

By changing I (the interval containing the tone); that will presumably change pI and thus pI-pA.

Now it becomes a question of logic. pI is fixed. There's nothing you

can do about that. You want to disturb pI-pA, not just pA, and you
want to do this without changing pI. How, logically, can that be
possible?

Your premise is incorrect, I believe. We can disturb the hypothetical controlled variable in my model (pI-pA) by manipulating I. One way to manipulate I is to have the interval containing the tone (I) vary during a trial. I would suggest changing I when A begins. If pI-pA is not controlled, then such a change should have no effect on the answer (A) that is being generated in response to control of pI-iA, where iA is the imagined answer to the initial value of I during the trial.

MT:. I certainly

do NOT control pI-pA. To do so would make no sense, so far as I can
see.

It doesn’t matter whether it makes sense to you or not. It’s a model of how things work and I’m pretty sure I could implement it as a computer program that would imitate the behavior in a forced choice experiment. I’m sure a computational version of your model, which controls pI-iA as well as pA, would also produce behavior that matches that seen in a forced choice experiment. What I’m looking for is an experiment that will discriminate between these two models. I think the one I describe above – where I is varied during a trial – would do the job. Don’t you think?

Mine controls pI-iA in your terminology, the resulting value of

iA being fed as rA (reference value for pA) to the system that
controls pA. Your model has to have a system that controls pA, too,
and it needs to be fed a reference value. Where does that reference
value come from?

From the system controlling pI-pA, same as yours. The error in the system controlling pI-pA is the reference for the system controlling pA. It’s just the same as yours, really, except that your model only controls the relationship between the perceived values of I and A in imagination.

Let's consider this example, because it is pertinent. Yes, the mouse

position is controlled. It is a location on the physical desk, not a
location on the screen. The cursor is a location on the screen, so
the expression c - d + m is a category error. Control of the mouse
position has a side-effect, which is that the position of the cursor
on the screen changes. But note that it is a side-effect, in that
the controller of mouse position knows nothing of the cursor. You
could disturb the mouse position control system by pushing the mouse
physically. You can’t do that with the cursor. If you want to
disturb the cursor position, you have to do that in the electronics.

Yes, I understand this. That’s why our models work so well. Sometimes we explicitly put in a lower level system that controls m. Usually we just assume that m is a controlled output, without specifying how it is controlled (via muscle forces); then m corresponds to the controlled mouse position read by the computer and c = d + m is not a category error.

MT: Earlier, you recognized that my model controls pI-iA. Today you say

my model says the subject is controlling only pA. Why?

Sorry, I meant that pA is the only variable in your model that can be seen to be controlled by an observer. But I know that your model, as implemented computationally, would control both pI-iA and pA.

RM: The difference between
our models is that in my model the subject’s behavior in a forced
choice detection experiment involves control of pI-pA; in your model,
the subject’s behavior involves control only of pA.

MT: I'd prefer it if you went back to the message in which you said

(correctly) that in my model the subject’s behaviour involves
control of pI-iA as well as of pA.

Again, I should have said that this is the only testable difference between our models. My model predicts that disturbances to pI-pA and pA will be resisted; both are controlled variables in my model. Your model predicts that disturbance to pI-iA and pA will be resisted but there is no way to monitor pI-iA to see if it is actually controlled. So I think the only way to compare our models is to see whether pI-pA is controlled. If it isn’t, then your model wins since pI-pA is not controlled in your model.

RM: My model (with
added memory) would behave the same as yours in your experiment.

How is that? The experiment separates the perception of which

interval held the signal from the occurrence of the response. Are
you saying that in your model with memory, the subject holds the
waveforms of the four intervals in mind for one trial while
listening to the next, so that the remembered sound sequence can be
compared on demand with a response if one is demanded? That would
seem to demand an awful lot of the memory, to generate some level of
perception that just might be asked to enter into a controlled
relationship, while other similar sequences are being perceived on
the same path.

Your model has suddenly become awfully complicated!! It seems a very

desparate rearguard action to preserve some vestige of the original
model. I don’t buy it. But I suppose anything is possible, if you
imagine hard enough.

I don’t really know how our models would work in the experiment you described. Your experiment seems very complicated and not to the point, which from my point of view is see whether or not subjects are controlling pI-pA or something like it (as per my model) or not (as per your model). Since the difference between our models is a difference in controlled variables, it seems to me that research to test the models should be focused on testing to see what subjects actually are controlling in these situations.

Best

Rick

···

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

[rsmarken]

[From Rick Marken (2010.07.23.1010)]

Martin Taylor (2010.07.23.10.57)--

Here's another variant on the experiment to distinguish between control of
the relationship pI-iA and pI-pA (relationship between perceived stimulus
and imagined answer as opposed to perceived stimulus and perceived actual
answer).

In this experiment, the subject is again asked to identify which of four
intervals contained a "signal". A set of four intervals constitutes one
trial.

On each trial except the first, the subject makes a response. There are two
buttons, labelled respectively "Same" and "Different". The subject is asked
to press "Same" if the interval containing the signal was the same on this
trial as it was on the last, and "Different" if it was not. Using Rick's
notation, the only possible values of p(A) are "Same" and "Different",
whereas the possible values for the interval to be detected are {1, 2, 3,
4}.

I just don't see how this test distinguish between control of pI-iA vs pI-pA.

If the subject in the conventional experiment is controlling pI-pA, then in
this experiment the subject would be controlling pb-pA, but could never be
controlling be pI-(any overt response).

Of course they could. But they are being asked to control something
else, which is the relationship between their Same/Different response
and a perception of whether the tone intervals on successive trials
were the same or different.

(Of course, I would expect the
subject to be controlling pb-iA rather than pb-pA, but this experiment
doesn't test for that).

But we are talking about models that control either pI-pA (mine) or
not (yours). Why not develop an experiment to test that?

Since you didn't like my first cut, and presented a workable but
extraordinary baroque alternative explanation of the results, maybe you will
agree this version would discriminate whether subjects determine the
interval containing the signal by controlling pI-iA or pi-pA. -- but I'm not
holding my breath :-).

Good. I don't think this experiment does any better than the first.

Or maybe you have an even more baroque alternative explanation to offer?

I think it makes more sense to design experiments to test a model than
to develop experiments to see if we can think of models that would
explain the results. I think we are agreed that the difference between
our models is a difference in what variables we think the subject
controls. My model says that the subject controls a perception of the
relationship between perceived tone interval and reported tone
interval; your model says that the subject controls this perception
only in imagination. So, according to your model, if we disturb the
relationship between perceived and reported tone interval, there
should be no resistance to this disturbance. According to my model
there should be resistance. So let's try to think of an experiment
that tests this difference between the models. I suggested one
possibility. If that's not acceptable to you let me know and maybe we
can try something else.

Best

Rick

···

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

[MartinT]

[Martin Taylor 2010.07.27.17.19]

I waited a few days before responding to this message [From Rick

Marken (2010.07.23.0900)] and to the immediately following one [From
Rick Marken (2010.07.23.1010)], in the hope that you would see for
yourself some of the problems with your comments on my two suggested
experiments. Since you haven’t yet sent the message I was hoping
for, I suppose I must state the obvious myself.

[From Rick Marken (2010.07.23.0900)]

Referring to the experiment in which the subject provides an overt

response as to which of four intervals contained a signal tone only
if the experimenter requests it, using a modality the experimenter
changes with each request, and in which the experimenter’s requests
come after the trial following the one for which the response is
requested.

        Martin Taylor

(2010.07.22.23.41)

Rick Marken (2010.07.22.1010)--

RM: ...So I think we have to set up a force choice
tone detection experiment where we can disturb pI-pA.

MT: How could you do that when pA is a controlled variable?

RM: By adding a disturbance to pl-pA: (pl-pA)+ d.

        1. pI is fixed. It's

the disturbance to the relationship perception.

      Actually, the disturbance is I, the interval in which the tone

occurs. It is the independent variable in the experiment to it
can be manipulated. So pI is not necessarily fixed; it can
presumably be varied by varying I.

That happens on the next trial. We are discussing what happens

within a single trial. I is fixed, and so is pI, although the memory
mpI may differ from pI by the time the following trial has been
presented.

2. pA is controlled.

      Yes.

        3. If you can't change

pA and you can’t change pI, how do you disturb pI-pA?

      By changing I (the interval containing the tone); that will

presumably change pI and thus pI-pA.

Next trial, you can change I, but for this trial I is fixed. WITHIN

a trial, how do you disturb pI-pA independently of disturbing pA?

        Now it becomes a

question of logic. pI is fixed. There’s nothing you can do
about that. You want to disturb pI-pA, not just pA, and you
want to do this without changing pI. How, logically, can
that be possible?

      Your premise is incorrect,, I believe.  We can disturb the

hypothetical controlled variable in my model (pI-pA) by
manipulating I. One way to manipulate I is to have the
interval containing the tone (I) vary during a trial.

The tone occurs in exactly one interval of the four. How do you vary

it within a trial? Either it is in interval 1 and none of the
others, or in interval 2 and none of the others or … What you
suggest is weirder than quantum entanglement, though it has much the
same flavour of things being in two states at once.

I would suggest changing I when A begins.

When A begins, the fourth interval has long since been completed. In

fact, in the experiment as described, the following trial has also
been completed – four more presentation intervals.

      If pI-pA is not controlled, then such a change should have

no effect on the answer (A) that is being generated in
response to control of pI-iA, where iA is the imagined answer
to the initial value of I during the trial.

          MT:. I certainly do NOT control pI-pA. To do

so would make no sense, so far as I can see.

      It doesn't matter whether it makes sense to you or not. It's a

model of how things work and I’m pretty sure I could implement
it as a computer program that would imitate the behavior in a
forced choice experiment.

This, I gotta see! Time reversal and quantum entanglement in a

single simulation on a classical computer! Go for it!

      I'm sure a computational version of your model, which

controls pI-iA as well as pA, would also produce behavior that
matches that seen in a forced choice experiment. What I’m
looking for is an experiment that will discriminate between
these two models. I think the one I describe above – where I
is varied during a trial – would do the job. Don’t you think?

No. I think it is procedural fantasy to require time reversal in a

real-world experiment, or even in a thought-experiment.

        Let's consider this

example, because it is pertinent. Yes, the mouse position is
controlled. It is a location on the physical desk, not a
location on the screen. The cursor is a location on the
screen, so the expression c - d + m is a category error.
Control of the mouse position has a side-effect, which is
that the position of the cursor on the screen changes. But
note that it is a side-effect, in that the controller of
mouse position knows nothing of the cursor. You could
disturb the mouse position control system by pushing the
mouse physically. You can’t do that with the cursor. If you
want to disturb the cursor position, you have to do that in
the electronics.

      Yes, I understand this. That's why our models work so well.

Sometimes we explicitly put in a lower level system that
controls m. Usually we just assume that m is a controlled
output, without specifying how it is controlled (via muscle
forces); then m corresponds to the controlled mouse position
read by the computer and c = d + m is not a category error.

Your "m" corresponds to my "A". It's a controlled output in the

sense that it is an output that produces directly a controlled
perception that is brought to a defined reference value. But to call
it a controlled output is to confuse people who do believe that we
can control our outputs. We can create outputs that bring
perceptions (perhaps of those outputs) to their reference values,
but we only control inputs, and it would be better if we tried to
not use casual language that might lead some people to think
otherwise.

[From Rick Marken (2010.07.23.1010)] referring to an experimental variant in which the relationship control produces an answer for the interval containing the signal on each trial, but the subject is never asked to make a response that identifies which interval that was. Instead, the subject is asked only to compare the answers produced by the relationship control system on trial N with the answer it produced on trial N-1. I argued that since the subject never makes a response that corresponds to the signal interval, the relationship control system that decides the appropriate answer on each trial cannot be using pA (the perceived actual response) as one of its inputs. It must be using an imagined answer which is remembered for comparison with the answer on the next interval.
[RM] I just don't see how this test distinguish between control of pI-iA vs pI-pA.
[MT] Because pI is of the form "it sounded louder in interval 3 than the other three intervals", whereas pA has only two possible values "Same" and "Different". I don't see how it can be any clearer, unless you are Lord Nelson, putting your cognitive telescope to your cognitive blind eye.
[RM] But they are being asked to control something
else, which is the relationship between their Same/Different response
and a perception of whether the tone intervals on successive trials
were the same or different.
[MT] Of course they are being asked to control that relationship, but in order to generate the "I" input to that relationship, they have to have controlled the relationship that is the point of the experiment, the relationship between the perception of the interval and the perception of the (real or imagined) answer. I'm not concerned as to whether the "Same-Different" relationship control system has a perceived real response input or an imagined response as its other input. That's irrelevant to the demonstration that the interval relationship control does not involve the perception of any physical response.
The only way you can get around this is by invoking a memory system even more baroque than the one you produced to squirm out of the issue addressed in the first experiment (discussed in the first part of this message). Not only do you have to find a way for the subject to remember the phsyical sound patterns of two sets of intervals, you then have to have the subject match not the remembered sounds but the interpretation of those sounds in terms of an interval number which at that point they have not generated by controlling any kind of relationship. I think baroque isn't the word. Rococo might be better -- pleasing in its imagery of swirling interwoven curlicues, but totally non-functional.
[RM] I think we are agreed that the difference between
our models is a difference in what variables we think the subject
controls. My model says that the subject controls a perception of the
relationship between perceived tone interval and reported tone
interval; your model says that the subject controls this perception
only in imagination. So, according to your model, if we disturb the
relationship between perceived and reported tone interval, there
should be no resistance to this disturbance.
[MT] I gather you still haven't figured out the difference between controlling imagination and controlling IN imagination. Yes, there would be resistance. If you could disturb I without getting into time reversal and quantum entanglement, then the imagined Answer would change to compensate, as would the overt response that corresponds to it because of the reference value supplied to the output hierarchy. There would be no difference in that respect between the models.

<details class='elided'>
<summary title='Show trimmed content'>&#183;&#183;&#183;</summary>

-----------------
Howsoever that may be, are you still in agreement with yourself that both models involve a relationship control system that has inputs pI and (real or imagined) pA (i.e. is controlling either pI-pA (yours) or pi-iA (mine)), and that both will reliably produce an overt response that corresponds closely to the interval in which the subject perceived the signal to have occurred? Your agreements with yourself don't have much staying power, so despite that you say it again in the paragraph quoted immediately above, I think it is necessary to ask.
Martin

[rsmarken]

[From Rick Marken (2010.07.26.2300)]

Martin Taylor (2010.07.27.17.19)--

I waited a few days before responding to this message [From Rick Marken
(2010.07.23.0900)] and to the immediately following one [From Rick Marken
(2010.07.23.1010)]

Nice to have you back. But I think you and I may be the only two who
are still interested in this. Perhaps Bill will comment when he gets
back. But I really would like to get this turned into a piece of
research so any comments would be welcome.

Rick Marken (2010.07.22.1010)--

RM: Actually, the disturbance is I, the interval in which the tone occurs.

MT: That happens on the next trial. We are discussing what happens within a
single trial.

The value of I (the interval containing the tone on each trial) is the
_value_ of the IV on a trial, and, of course, that value varies over
trials. You can also vary I within a trial if you like; but the
structure of a trial in a tone detection task would have to be changed
to make such variation experimentally useful. One way to do this would
be to change the sequence of intervals that might contain a tone into
parallel intervals. For example, on a trial the tone could either be
presented to the left, right or both ears (centered). Then within a
trial the tone could shift from left to center, center to right, etc.

MT: 3. If you can't change pA and you can't change pI, how do you disturb
pI-pA?

RM: By changing I (the interval containing the tone); that will presumably
change pI and thus pI-pA.

MT:Next trial, you can change I, but for this trial I is fixed. WITHIN a trial,
how do you disturb pI-pA independently of disturbing pA?

In the auditory task it would take some ingenuity, but that's the art
of experimentation.

Now it becomes a question of logic. pI is fixed. There's nothing you can
do about that. You want to disturb pI-pA, not just pA, and you want to do
this without changing pI. How, logically, can that be possible?

It's not a question of logic. It's a question of designing an
experiment to test whether the subject is controlling for a match
between interval (I) and answer (A) only in imagination (as per your
model) or in reality (perception thereof, anyway). I think an
experiment, where the location of the tone (left, right or center)
changes during some trials just as the subject starts to answer would
test the difference between our models. If your model is correct the
subject will continue to give the answer appropriate to the interval
where the tone first occurred; if my model is correct, the subject
will will start changing the answer immediately after the tone occurs
in the new interval.

RM: The tone occurs in exactly one interval of the four. How do you vary it
within a trial?

By changing the interval (spatial location) containing the tone during
the trial. If the trial starts with the tone centered then, after say
100 msec, change the location (on half the trials, so the subject
won't know whether it's going to change or not) to the left or right.

RM: I would suggest changing I when A begins.

When A begins, the fourth interval has long since been completed.

Good point. As I said, we would have to use some ingenuity so that
this could be done. Rather than using tones I think it would be easier
to do a visual version of this forced choice task. There could be 4
boxes on the screen (the 4 intervals). The subject is to move a cursor
to the box containing the target (an easily visible dot would be nice)
as soon as the target appears. When the mouse starts moving the box
containing the target will change (on half the trials) or not. The
cursor would be influenced by a disturbance, of course. According to
your model (as I understand it), as soon as the target comes on in one
of the boxes the subject imagines the response and then a few msecs
later sends a reference to the response system, which will actually
move the cursor to what had been the imagined answer box, and do this
in a controlled way. So when the target changes to another box while
the response is in progress, the response would still continue to the
"old" target for some time. In my model, the response would start
changing instantly and proceed to the new target.

MT:. I certainly do NOT control pI-pA. To do so would make no sense, so
far as I can see.

RM: It doesn't matter whether it makes sense to you or not. It's a model of how
things work and I'm pretty sure I could implement it as a computer program
that would imitate the behavior in a forced choice experiment.

MT: This, I gotta see! Time reversal and quantum entanglement in a single
simulation on a classical computer! Go for it!

I don't understand. What time reversal and quantum entanglement is
involved? If it's the sequential nature of the auditory forced choice
task then, as I said, we will have to devise the experiment so that
that is not a factor. There is also no time travel in my model, which
controls the present time perception of the relationship between the
perceived interval containing the target (pI) and the perceived answer
(pA); the perception that I call pI-pA.

MT: [From Rick Marken (2010.07.23.1010)] referring to an experimental
variant in which the relationship control produces an answer for the interval
containing the signal on each trial, but the subject is never asked to make
a response that identifies which interval that was...

[RM] I just don't see how this test distinguish between control of pI-iA vs
pI-pA.

[MT] Because pI is of the form "it sounded louder in interval 3 than the
other three intervals", whereas pA has only two possible values "Same" and
"Different".

What I don't see is how our models make different predictions in this
experiment. I presume your model would predict that the subject would
say "Same" when the tone falls into the same interval in the prior and
current trials and "Different" when it doesn't. My model would do the
same. The experiment doesn't seem to discriminate between the models
as my proposed experiment (described above) does.

[MT] Of course they are being asked to control that relationship, but in
order to generate the "I" input to that relationship, they have to have
controlled the relationship that is the point of the experiment, the
relationship between the perception of the interval and the perception of
the (real or imagined) answer. I'm not concerned as to whether the
"Same-Different" relationship control system has a perceived real response
input or an imagined response as its other input. That's irrelevant to the
demonstration that the interval relationship control does not involve the
perception of any physical response.

I hope you can see now that that problem with the experiment (from my
perspective) is that it doesn't seem to discriminate between your
control in imagination model and my control of perception model. If
the models do make different predictions about what will happen in
this experiment then please let me know; I can't see it.

[RM] I think we are agreed that the difference between
our models is a difference in what variables we think the subject
controls. My model says that the subject controls a perception of the
relationship between perceived tone interval and reported tone
interval; your model says that the subject controls this perception
only in imagination. So, according to your model, if we disturb the
relationship between perceived and reported tone interval, there
should be no resistance to this disturbance.

[MT] I gather you still haven't figured out the difference between
controlling imagination and controlling IN imagination. Yes, there would be
resistance. If you could disturb I without getting into time reversal and
quantum entanglement, then the imagined Answer would change to compensate,
as would the overt response that corresponds to it because of the reference
value supplied to the output hierarchy. There would be no difference in that
respect between the models.

So then your model and mine are the same; they are both controlling a
perception. But yours is also controlling an imagination in parallel.
So in that case there is probably no experiment that can discriminate
between the two and I would just argue that mine is to be preferred
because it is more parsimonious.

Howsoever that may be, are you still in agreement with yourself that both
models involve a relationship control system that has inputs pI and (real or
imagined) pA (i.e. is controlling either pI-pA (yours) or pi-iA (mine)), and
that both will reliably produce an overt response that corresponds closely
to the interval in which the subject perceived the signal to have occurred?

Yes. But as I understand your model, the overt response is produced
after the imagined version of this response has been produced in
imagination. So if I is changed once the switch from imagination to
overt response mode has occurred, there should be no change in the
observed response until the system returns to imagination mode; and
then there should be another delay until the new imagined response has
been produced and a switch back to overt response is made. This kind
of switching doesn't occur in my model. So if I is changed when the
subject starts to respond there should be an immediate change in the
response. That's how I think we can discriminate between the two
models.

Best

Rick

···

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

[Bill_Powers1]

[From Bill Powers (2010.07.26.1601 GMT)]

Martin Taylor 2010.07.27.17.19 –

MMT: I waited a few days before
responding to this message [From Rick Marken (2010.07.23.0900)] and to
the immediately following one [From Rick Marken (2010.07.23.1010)], in
the hope that you would see for yourself some of the problems with your
comments on my two suggested experiments. Since you haven’t yet sent the
message I was hoping for, I suppose I must state the obvious
myself.

BP: The clarity of this exchange isn’t improving much, and the impatience
level is increasing. “Stating the obvious” is always a good
thing to do since it seldom turns out to be obvious to everyone, but of
course that leads to more irritation, until at some point it become
“obvious” that the real argument is about a subject that isn’t
being mentioned until the very end.

MMT: Referring to the experiment
in which the subject provides an overt response as to which of four
intervals contained a signal tone only if the experimenter requests it,
using a modality the experimenter changes with each request, and in which
the experimenter’s requests come after the trial following the one for
which the response is requested.

BP: As I understand the experiment you describe, the subject hears short
periods of white noise, in some of which there is also a faint tone. The
task is to say whether the tone was present or absent, with the added
variation that the presentation which is asked about is not the
just-previous one, but the one before that, requiring that the subject
remember and give that answer rather than answering the same question
about the current presentation. The current answer must also be selected
and remembered while the other is being given, because the experimenter
might also ask about it after the next presentation). I don’t know what
“using a modality which is changed with every request” means
unless you mean the request varies among being given in spoken or written
form or is a signal involving some other sense, such as tactile. That
doesn’t seem to be what you mean, though.

This is a quite complex experiment and I’m not sure what the point of
adding these provisions is, or whether we’re talking any longer about a
real experiment or an imagined one. I am guessing that the additions are
offered in defense of the idea that a higher control system can issue a
reference signal to a lower system that turns it into an overt reply to a
question, and that this overt response is not sensed by the higher system
as part of its control behavior. This justifies treating the overt
answer, therefore, as if it is open-loop relative to the higher system.
And as I understand the reason for this, it is so that a calculation of
some relationship between a stimulus and a response can be made as if the
relationship involved only forward temporal characteristics of the
system. And apparently (to me anyway) the reason for wanting to establish
this idea is to validate some experiments that have already been
done.

If we could return to the basic experiment, the task seems to require
using two perceptual input functions at the same time, one detecting
white noise and the other detecting a tone. Each of these PIFs would
produce a perceptual signal, with the tone signal presumably being quite
small relative to the signal in the other PIF that perceives noise, and
more variable because of the presence of the noise. The two perceptual
signals would be reported upon by a higher-order system as a noise signal
present on every trial, accompanied in some trials by a somewhat
unreliable tone signal from the other PIF.

This doesn’t address the main question; I offer this proposed analysis
just to give a PCT interpretation of the perceptual situation.

The next question is whether we always issue a verbal report on a
perception without considering the relationship of the actual (overt)
report to the perception. I’m rather sure that we all do this
occasionally, and that it is one reason we sometimes have difficulty in
communicating. We often fail to read what we write before sending it, and
to hear what we are actually saying to other people rather than just
attending to the imagined meanings we hope to get across. But the fact
that there can be a difference between what we imagine and what we
actually do is good enough reason to propose that this sort of open-loop
process doesn’t give good control of communication, and is probably
mostly replaced by a better-organized system after some experience with
the consequences. So it is for this reason that I am skeptical about the
idea that overt answers are not perceived in relationship to the signal
or other occasion for the answer. I am skeptical about the implication
that an error in this relationship would not be detected and immediately
followed by at least an attempt to correct it, however futile in a task
involving discrete variables.

I don’t think we can take this discussion much further than that without
actually doing some PCT experiments to establish the facts. Do subjects
behave as if they are insensitive to errors in the relationship of the
overt “response” to the signal calling for it? It doesn’t seem
that anyone has asked that question or answered it, so any conclusions
that depend on the answer must be postponed.

Best,

Bill P.

P.S.I head home from Manchester at 1:30 tomorrow and will be home by 6:30
that evening, thanks to time zones and chasing the Sun.

···

Martin Taylor (2010.07.22.23.41)

Rick Marken (2010.07.22.1010)--

RM:  ...So I think we have to set up a force choice

tone detection experiment where we can disturb pI-pA.

MT: How could you do that when pA is a controlled variable?

RM: By adding a disturbance to pl-pA: (pl-pA)+ d.

1. pI is fixed. It’s the disturbance to the relationship perception.

Actually, the disturbance is I, the interval in which the tone occurs. It
is the independent variable in the experiment to it can be manipulated.
So pI is not necessarily fixed; it can presumably be varied by varying
I.

That happens on the next trial. We are discussing what happens within a
single trial. I is fixed, and so is pI, although the memory mpI may
differ from pI by the time the following trial has been presented.

2. pA is controlled.

Yes.

3. If you can’t change pA and you can’t change pI, how do you disturb
   pI-pA?

By changing I (the interval containing the tone); that will presumably
change pI and thus pI-pA.

Next trial, you can change I, but for this trial I is fixed. WITHIN a
trial, how do you disturb pI-pA independently of disturbing pA?

Now it becomes a question of logic. pI is fixed. There’s nothing you
can do about that. You want to disturb pI-pA, not just pA, and you want
to do this without changing pI. How, logically, can that be
possible?

Your premise is incorrect, I believe. We can disturb the
hypothetical controlled variable in my model (pI-pA) by manipulating I.
One way to manipulate I is to have the interval containing the tone (I)
vary during a trial.

The tone occurs in exactly one interval of the four. How do you vary it
within a trial? Either it is in interval 1 and none of the others, or in
interval 2 and none of the others or … What you suggest is weirder than
quantum entanglement, though it has much the same flavour of things being
in two states at once.

I would suggest changing I when
A begins.

When A begins, the fourth interval has long since been completed. In
fact, in the experiment as described, the following trial has also been
completed – four more presentation intervals.

If pI-pA is not controlled, then
such a change should have no effect on the answer (A) that is being
generated in response to control of pI-iA, where iA is the imagined
answer to the initial value of I during the trial.

MT:. I certainly do NOT control pI-pA. To do so would make no sense,
so far as I can see.

It doesn’t matter whether it makes sense to you or not. It’s a model of
how things work and I’m pretty sure I could implement it as a computer
program that would imitate the behavior in a forced choice experiment.

This, I gotta see! Time reversal and quantum entanglement in a single
simulation on a classical computer! Go for it!

I’m sure a computational version
of your model, which controls pI-iA as well as pA, would also produce
behavior that matches that seen in a forced choice experiment. What I’m
looking for is an experiment that will discriminate between these two
models. I think the one I describe above – where I is varied during a
trial – would do the job. Don’t you think?

No. I think it is procedural fantasy to require time reversal in a
real-world experiment, or even in a thought-experiment.

Let’s consider this example, because it is pertinent. Yes, the mouse
position is controlled. It is a location on the physical desk, not a
location on the screen. The cursor is a location on the screen, so the
expression c - d + m is a category error. Control of the mouse position
has a side-effect, which is that the position of the cursor on the screen
changes. But note that it is a side-effect, in that the controller of
mouse position knows nothing of the cursor. You could disturb the mouse
position control system by pushing the mouse physically. You can’t do
that with the cursor. If you want to disturb the cursor position, you
have to do that in the electronics.

Yes, I understand this. That’s why our models work so well. Sometimes we
explicitly put in a lower level system that controls m. Usually we just
assume that m is a controlled output, without specifying how it is
controlled (via muscle forces); then m corresponds to the controlled
mouse position read by the computer and c = d + m is not a category
error.

Your “m” corresponds to my “A”. It’s a controlled
output in the sense that it is an output that produces directly a
controlled perception that is brought to a defined reference value. But
to call it a controlled output is to confuse people who do believe that
we can control our outputs. We can create outputs that bring perceptions
(perhaps of those outputs) to their reference values, but we only control
inputs, and it would be better if we tried to not use casual language
that might lead some people to think otherwise.

[From Rick Marken (2010.07.23.1010)] referring to an experimental
variant in which the relationship control produces an answer for the
interval containing the signal on each trial, but the subject is never
asked to make a response that identifies which interval that was.
Instead, the subject is asked only to compare the answers produced by the
relationship control system on trial N with the answer it produced on
trial N-1. I argued that since the subject never makes a response that
corresponds to the signal interval, the relationship control system that
decides the appropriate answer on each trial cannot be using pA (the
perceived actual response) as one of its inputs. It must be using an
imagined answer which is remembered for comparison with the answer on the
next interval.
[RM] I just don't see how this test distinguish between control of pI-iA
vs pI-pA.
[MT] Because pI is of the form "it sounded louder in interval 3 than
the other three intervals", whereas pA has only two possible values
"Same" and "Different". I don't see how it can be any
clearer, unless you are Lord Nelson, putting your cognitive telescope to
your cognitive blind eye.
[RM] But they are being asked to control something
else, which is the relationship between their Same/Different response
and a perception of whether the tone intervals on successive trials
were the same or different.
[MT] Of course they are being asked to control that relationship, but in
order to generate the "I" input to that relationship, they have
to have controlled the relationship that is the point of the experiment,
the relationship between the perception of the interval and the
perception of the (real or imagined) answer. I'm not concerned as to
whether the "Same-Different" relationship control system has a
perceived real response input or an imagined response as its other input.
That's irrelevant to the demonstration that the interval relationship
control does not involve the perception of any physical response.
The only way you can get around this is by invoking a memory system even
more baroque than the one you produced to squirm out of the issue
addressed in the first experiment (discussed in the first part of this
message). Not only do you have to find a way for the subject to remember
the phsyical sound patterns of two sets of intervals, you then have to
have the subject match not the remembered sounds but the interpretation
of those sounds in terms of an interval number which at that point they
have not generated by controlling any kind of relationship. I think
baroque isn't the word. Rococo might be better -- pleasing in its imagery
of swirling interwoven curlicues, but totally non-functional.
[RM] I think we are agreed that the difference between
our models is a difference in what variables we think the subject
controls. My model says that the subject controls a perception of the
relationship between perceived tone interval and reported tone
interval; your model says that the subject controls this perception
only in imagination. So, according to your model, if we disturb the
relationship between perceived and reported tone interval, there
should be no resistance to this disturbance.
[MT] I gather you still haven't figured out the difference between
controlling imagination and controlling IN imagination. Yes, there would
be resistance. If you could disturb I without getting into time reversal
and quantum entanglement, then the imagined Answer would change to
compensate, as would the overt response that corresponds to it because of
the reference value supplied to the output hierarchy. There would be no
difference in that respect between the models.
-----------------
Howsoever that may be, are you still in agreement with yourself that both
models involve a relationship control system that has inputs pI and (real
or imagined) pA (i.e. is controlling either pI-pA (yours) or pi-iA
(mine)), and that both will reliably produce an overt response that
corresponds closely to the interval in which the subject perceived the
signal to have occurred? Your agreements with yourself don't have much
staying power, so despite that you say it again in the paragraph quoted
immediately above, I think it is necessary to ask.
Martin

[rsmarken]

[From Rick Marken (2010.07.27.1020)]

Bill Powers (2010.07.26.1601 GMT) to Martin Taylor 2010.07.27.17.19 --

I don't think we can take this discussion much further than that without
actually doing some PCT experiments to establish the facts. Do subjects
behave as if they are insensitive to errors in the relationship of the overt
"response" to the signal calling for it? It doesn't seem that anyone has
asked that question or answered it, so any conclusions that depend on the
answer must be postponed.

Perhaps you didn't read my last post ([From Rick Marken
(2010.07.26.2300)]) but in it I suggest an experiment that represents
an attempt to answer precisely this question: Do subjects behave as if
they are insensitive to errors in the relationship of the overt
"response" to the signal calling for it? Here is my suggestion from
that post:

RM: There could be 4 boxes on the screen (the 4 intervals). The subject is
to move a cursor to the box containing the target (an easily visible dot would
be nice) as soon as the target appears. When the mouse starts moving the
box containing the target will change (on half the trials) or not. The cursor
would be influenced by a disturbance, of course. According to your model
(as I understand it), as soon as the target comes on in one of the boxes the
subject imagines the response and then a few msecs later sends a reference
to the response system, which will actually move the cursor to what had been
the imagined answer box, and do this in a controlled way. So when the target
changes to another box while the response is in progress, the response would
still continue to the "old" target for some time. In my model, the response
would start changing instantly and proceed to the new target.

I framed it as a test of a difference between two models, Martin's
control of imagination model and my control of perception model. But I
think that's what we are doing -- testing a difference between models
of behavior in the forced choice experiment -- when we are asking the
question about the subjects being insensitive to errors in the
relationship of the overt "response" to the signal calling for it.
Martin is proposing a model of the behavior in these experiments where
subjects are indeed insensitive to such errors (which I would be
producing in my proposed experiment by moving the target at the start
of the response during a trial) whereas my model (which I believe is
the same as yours) predicts that subjects are sensitive to such errors
(behavior the are controlling the perceived relationship between
signal and overt response) and will act to correct such errors during
a trial when they exist.

Whaddaya think?

Hope you're having a nice time in England. It was great to see you.

Best

Rick

···

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

[rsmarken]

[From Rick Marken (2010.07.27.1030)]

Oops. I made a weird response error (how apropos) that I am now correcting. The penultimate sentence in [From Rick Marken (2010.07.27.1020)] container an error; the corrected words are shown in red:

Martin is proposing a model of the behavior in these experiments where subjects are indeed insensitive to such errors (which I would be producing in my proposed experiment by moving the target at the start of the response during a trial) whereas my model (which I believe is the same as yours) predicts that subjects are sensitive to such errors (because they are controlling the perceived relationship between signal and overt response) and will act to correct such errors during a trial when they exist.

Best

Rick

···

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

[MartinT]

[Martin Taylor 2010.07.27.14.52]

[From Rick Marken (2010.07.26.2300)]

Martin Taylor (2010.07.27.17.19)--
I waited a few days before responding to this message [From Rick Marken
(2010.07.23.0900)] and to the immediately following one [From Rick Marken
(2010.07.23.1010)]

Nice to have you back.

I wasn't away. I was waiting in hope that you might continue to think about the issues, and would correct your earlier messages.

But I think you and I may be the only two who
are still interested in this.

Maybe because we were once psychophysicists? I don't know of any others on CSGnet, though they may well be lurking out there.

Perhaps Bill will comment when he gets
back. But I really would like to get this turned into a piece of
research so any comments would be welcome.

Rick Marken (2010.07.22.1010)--
RM: Actually, the disturbance is I, the interval in which the tone occurs.

MT: That happens on the next trial. We are discussing what happens within a
single trial.

The value of I (the interval containing the tone on each trial) is the
_value_ of the IV on a trial, and, of course, that value varies over
trials. You can also vary I within a trial if you like; but the
structure of a trial in a tone detection task would have to be changed
to make such variation experimentally useful. One way to do this would
be to change the sequence of intervals that might contain a tone into
parallel intervals. For example, on a trial the tone could either be
presented to the left, right or both ears (centered). Then within a
trial the tone could shift from left to center, center to right, etc.

That would introduce a perceptual uncertainty into the detection problem, which would make the detection more difficult, but it wouldn't affect which interval held the tone. Anyway, are we not trying to address the validity within PCT of the standard N-interval forced choice experiment? Why add this complication? My variants did not change the detection problem. They only changed the manner and timing of the possible responses.

MT: 3. If you can't change pA and you can't change pI, how do you disturb
pI-pA?

RM: By changing I (the interval containing the tone); that will presumably
change pI and thus pI-pA.

MT:Next trial, you can change I, but for this trial I is fixed. WITHIN a trial,
how do you disturb pI-pA independently of disturbing pA?

In the auditory task it would take some ingenuity, but that's the art
of experimentation.

I think that's a cop-out. Look, the question is always how well a person can detect the presence of some "signal" defined by mutual agreement between the experimenter and the subject before the trial begins. Whether the experiment is the N-AFC experiment we have been concentrating on or some variant, if you change the nature of the presentation, that is very likely to change how well the subject can detect the presence of the signal. To change "I" within a trial is to change the nature of the presentation, and very probably to alter the subject's ability to detect the signal.

Now it becomes a question of logic. pI is fixed. There's nothing you can
do about that. You want to disturb pI-pA, not just pA, and you want to do
this without changing pI. How, logically, can that be possible?

It's not a question of logic. It's a question of designing an
experiment to test whether the subject is controlling for a match
between interval (I) and answer (A) only in imagination (as per your
model) or in reality (perception thereof, anyway).

I think that changing (a) when the answer is required, (b) the subject's knowledge of whether an answer will be required, and (c) the nature of the answer if and when it is required (my Experiment 1) will accomplish this. My experiment 2 accomplishes it in a different way, by decoupling the control of the choice of interval (choice of 1, 2, 3, or 4) from the nature of the Answer (choice of "Same" or "Different").

Under the conditions of either experiment, if the "in reality" perception of the answer is to be the "answer" input to the PIF of the control unit that controls the match of perceived interval to answer, then the sensory data must, at some level of uncontrolled perception, be stored and recovered appropriately before being identified with an interval number. It's possible that this could be done, but the design of the experiments was intended to make that storage and retrieval as difficult as possible. To model it, you must include a model of how this sensory storage is done.

  I think an
experiment, where the location of the tone (left, right or center)
changes during some trials just as the subject starts to answer would
test the difference between our models. If your model is correct the
subject will continue to give the answer appropriate to the interval
where the tone first occurred; if my model is correct, the subject
will will start changing the answer immediately after the tone occurs
in the new interval.

What "new interval"? You are simply describing a tracking study in which a tone moves around in space. You can do that in continuous form, too. It's not a detection study of the kind we have been discussing, for two reasons. Firstly, if the signal is in an uncertain location, it's harder to detect than if it is in a location (and pitch and harmonic structure, etc.) known to the subject. Secondly, if it moves it is harder to detect than if it is stationary. Thirdly, now you have three spatially disparate noise sources rather than one, and you don't know how the noise sources in the locations other than the one containing the signal will influence the subject's ability to detect teh signal. You would have to do a whole mess of experiments to determine what we might call the auditory system's "noise spatial bandwidth" to get a handle on that.

Even if you discount the theoretical reasons (adduced in the 1950s and 1960s) why this is so, and even if you say that the experiments showing it to be so are PCT-invalid, you have to acknowledge to possibility that the statements might be true, and that their possible truth invalidates your proposal as a way of studying the subject's ability to detect the signal (other than under your specific proposed conditions).

RM: The tone occurs in exactly one interval of the four. How do you vary it
within a trial?

By changing the interval (spatial location) containing the tone during
the trial. If the trial starts with the tone centered then, after say
100 msec, change the location (on half the trials, so the subject
won't know whether it's going to change or not) to the left or right.

Yes, I understand your experimental setup (I think). I don't believe it would test the difference between our models even if it didn't influence the subject's ability to detect the signal.

RM: I would suggest changing I when A begins.

When A begins, the fourth interval has long since been completed.

Good point. As I said, we would have to use some ingenuity so that
this could be done. Rather than using tones I think it would be easier
to do a visual version of this forced choice task. There could be 4
boxes on the screen (the 4 intervals). The subject is to move a cursor
to the box containing the target (an easily visible dot would be nice)
as soon as the target appears.

The point of the experiment is to determine how well the subject can detect the signal, not to track an easily detected signal as it moves. Let me suggest changing the display. Have your four boxes, and in each one present say 20 dots moving quasi-randomly. In three of the boxes, the global mean location of all the dots is always in the middle of the box, but in one box the global mean location has a slow drift in one direction.

When the mouse starts moving the box
containing the target will change (on half the trials) or not. The
cursor would be influenced by a disturbance, of course. According to
your model (as I understand it), as soon as the target comes on in one
of the boxes the subject imagines the response and then a few msecs
later sends a reference to the response system, which will actually
move the cursor to what had been the imagined answer box, and do this
in a controlled way. So when the target changes to another box while
the response is in progress, the response would still continue to the
"old" target for some time. In my model, the response would start
changing instantly and proceed to the new target.

So you are proposing a reaction-time experiment, are you? I thought we were avoiding those as being impossible to interpret. However, let me just point out that in both models, the time for the sensory input (pI in your earlier terminology) to change would be the same. In both models the outflow pathways from the relationship control system to the overt response are the same. The only difference is that the feedback pathways from the perception of the actual response in my model go only to the control units dealing with making the response, whereas in yours they go back further, to the relationship control system itself. Looking at the circuitry this way makes it seem as though my model might result usually in a quicker change when the stimulus changes. But I strongly doubt that there would be a measureable difference, and if there was, that it would be possible to interpret the difference as a difference between models.

If you were to do an actual experiment, I suppose you would have to find a way to get the subject to control sometimes in imagination and sometimes by matching the perceived "real" response to the perceived sensory input. I wonder how you would do that?

MT:. I certainly do NOT control pI-pA. To do so would make no sense, so
far as I can see.

RM: It doesn't matter whether it makes sense to you or not. It's a model of how
things work and I'm pretty sure I could implement it as a computer program
that would imitate the behavior in a forced choice experiment.

MT: This, I gotta see! Time reversal and quantum entanglement in a single
simulation on a classical computer! Go for it!

...There is also no time travel in my model, which
controls the present time perception of the relationship between the
perceived interval containing the target (pI) and the perceived answer
(pA); the perception that I call pI-pA.

So, for you, the "perceived interval" is produced in an S-R process, not the relationship control system we have been assuming? It must be, because the sensory data has long gone when pA is available. Your pI either has been produced by some mechanism prior to the relationship control (in which case, why have the relationship control process at all?) or is a memory of the sensory data, which the experimental design is intended to make difficult.

MT: [From Rick Marken (2010.07.23.1010)] referring to an experimental
variant in which the relationship control produces an answer for the interval
containing the signal on each trial, but the subject is never asked to make
a response that identifies which interval that was...
[RM] I just don't see how this test distinguish between control of pI-iA vs
pI-pA.

[MT] Because pI is of the form "it sounded louder in interval 3 than the
other three intervals", whereas pA has only two possible values "Same" and
"Different".

What I don't see is how our models make different predictions in this
experiment.

Your model requires the addition of a model of sensory storage, or it requires the assumption that the interval can be chosen without using the pI-pA control system that both models have been assuming to produce the choice of interval. If your model can turn the sensory data (the original pI) into the choice of interval (your new pI) without using a relationship control unit, then your whole structure becomes the S-R structure that for several months you said my model was.

Howsoever that may be, are you still in agreement with yourself that both
models involve a relationship control system that has inputs pI and (real or
imagined) pA (i.e. is controlling either pI-pA (yours) or pi-iA (mine)), and
that both will reliably produce an overt response that corresponds closely
to the interval in which the subject perceived the signal to have occurred?

Yes.

I'll take that as a "Yes".

Much follows from that "Yes". Thank you.

Martin

[rsmarken]

[From Rick Marken (2010.07.27.2220)]

�Martin Taylor (2010.07.27.14.52)

MT: Anyway, are we not trying to address the validity within PCT of the standard
N-interval forced choice experiment?

RM: I suppose you could put it this way. The results of the standard
N-interval forced choice experiment are valid if the observed S-R
relationship reflects the forward, open-loop characteristics of the
system. Your model of the behavior in this experiment says that the
results of such experiments are valid in this sense. My model says
that they are not. So we can test whether, in fact, the results of
such an experiment are valid by testing to see which of our models is
correct.

MT: Why add this complication?

RM: What you call a "complication" I call an experimental manipulation
that makes it possible to compare the predictions of the two models.

MT: My variants did not change the detection problem. They only changed the
manner and timing of the possible responses.

RM: But it's not the detection aspect of the models that needs to be
tested. What we have to test is whether the subject in a N-interval
forced choice experiment is responding without considering the
relationship of the actual (overt) response to the stimulus (as per
your model) or is controlling a perception of the relationship between
the stimulus and overt response (as per mine).

RM: In the auditory task it would take some ingenuity, but that's the art
of experimentation.

MT: I think that's a cop-out. Look, the question is always how well a person can
detect the presence of some "signal" defined by mutual agreement between the
experimenter and the subject before the trial begins.

RM: That's not the question addressed by this research. The question
is whether the behavior in the experiment is closed-loop (as per my
model) or not (as per yours).

MT: Yes, I understand your experimental setup (I think). I don't believe it would
test the difference between our models even if it didn't influence the subject's
ability to detect the signal.

RM: Why do you think that? Please explain in terms of the models we've
agreed on.

MT: The point of the experiment is to determine how well the subject can detect
the signal, not to track an easily detected signal as it moves.

RM: That's the point from the point of view of the researcher using
this method, of course. And the observed relationship between S and R
would tell the reseacher something about the subject as a detection
system if the subject's behavior in the experiment were open-loop, as
per your model. But that's what I am questioning; my model of the
behavior in the experiment says that the behavior is closed-loop. If
my model is right, then the S-R relationships observed in these
experiment do not tell the researcher quite what he thinks they do.

[MT] Because pI is of the form "it sounded louder in interval 3 than the
other three intervals", whereas pA has only two possible values "Same" and
"Different".

RM: What I don't see is how our models make different predictions in this
experiment.

MT: Your model requires the addition of a model of sensory storage, or it
requires the assumption that the interval can be chosen without using the
pI-pA control system that both models have been assuming to produce the
choice of interval. If your model can turn the sensory data (the original pI)
into the choice of interval (your new pI) without using a relationship control
unit, then your whole structure becomes the S-R structure that for several
months you said my model was.

RM: This is not a difference in what our models predict. It is simply
a claim that my model of the experiment you propose would have to be
S-R. Since you can say that before we have even run the experiment it
is clear that your approach to science doesn't require that you
collect any data. Just propose a thought experiment and then claim
that no model besides yours can handle it. This is not the way I think
science should be done.

RM: The way I like to do science is to start with the models and then
design experiments to test them. The models should be formulated well
enough so that an experiment can be set up where the models make
clearly different predictions -- like the transit of Mercury
experiment that compared relativity to Newtons model of gravity. We
have spent quite a while getting to agreement about what our
alternative models of the N-interval forced choice experiment are. I
think we're there now. Yours controls the relationship between
perceived stimulus and imagined response until this relationship is
under control at which point the reference for the response is sent to
a lower level system which produces the overt response (under
control). Mine controls the relationship between perceived stimulus
and perceived overt response until this relationship is under control.
This little difference between our models is crucial and I think my
proposed experiment with target spots appearing in a box that might
change during a trial tests this difference between the models. If you
don't think it does test the difference then please explain why, in
terms of the two models, of course.

Best

Rick

···

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

[Bill_Powers1]

[From Bill Powers (2010.07.28.0825 GMT)]
I think we should make the experiment as much like Martin's experiment as possible. In his experiment, the only way to disturb the relationship without changing anything in the lower-order systems that press the button is not to disturb the lower-order system at all, as your idea of moving the squares would do. Perhaps what I suggest is impossible -- the presence or absence of feedback to the higher level isn't visible from outside the person and there really isn't any immediate feedback via the experimenter. It would take several trials before the experimenter realized that the button presses were all wrong and complained that the subject wasn't following the instructions (thinking of making the response invisible to the experimenter, as if the button didn't make contact).

Busy getting ready to come home. Manchester is definitely where the action is.

Best,

Bill P.

[rsmarken]

[From Rick Marken (2010.07.28.0940 PDT)]

Bill Powers (2010.07.28.0825 GMT)--

I think we should make the experiment as much like Martin's experiment as
possible.

I agree, as much as possible. But not so much that it is no longer a
test of the alternative models of the behavior in the experiment.

If I understand your suggestion, you would have the subject do the
standard N alternative forced choice detection task (Martin's
experiment), using button presses to indicate the interval in which
the tone was thought to occur. The only innovation would be that on
some (all?) trials the button would not make contact. The
experimenter, who is also a subject in the experiment, would be
monitoring the results of the button presses and would notice that the
subject is not producing the correct responses. So the experimenter
would inform the subject that he or she is not following instructions.
If the subject protests, it would show that the subject is controlling
for the relationship between the stimulus (tone interval) and their
overt response (the interval indicated by the button press).

The problem with this experiment (from my perspective) is that it
doesn't seem to test the essential difference between Martin's and my
models, as I understand them. Both my model and Martin's control the
perceived response (button press). In Martin's model the perception of
the button press is controlled independently of the perceived tone
interval; in mine the perception of the button press is controlled in
relationship to the perceived tone interval. I think both of our
models could predict that the subject would protest if told that
he/she were not following the instructions. And if the subject were
told by the experimenter that the button wasn't recording the correct
(or any) answer I think both our models could predict that the subject
would say something like "that's your problem, not mine; I'm doing
what I was asked to do".

While I think it is a good idea to make any experimental test of the
models of the behavior in Martin's experiment as much like that
experiment as possible, I also think that the reason we have models is
so that we can vary the situation in ways that make it possible to
test different predictions of the model. So while my proposed
experiment is not exactly like Martin's force choice experiment, I
think the variations I propose don't change the essential nature of
the task and, therefore, don't require a change to either of our
models. Both models can perform my proposed version of Martin's
experiment; the models just make different predictions about what will
happen in that experiment when the location of the signal changes
after the response is initiated.

I suppose I could try to make the experiment look a bit more like
Martin's original forced choice experiment. I'll think about it. But
right now I think that my proposed experiment would provide a pretty
cleat test of the difference between Martin's and my model.

Busy getting ready to come home. Manchester is definitely where the action is.

Have a nice, safe trip home. I'm so glad the action was there; it's
sure not here (depending on what action you're talking about, of
course;-)

Best

Rick

···

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

[rsmarken]

[From Rick Marken (2010.07.31.0940)]

Rick Marken (2010.07.28.0940 PDT)

I suppose I could try to make the experiment look a bit more like
Martin's original forced choice experiment. I'll think about it. But
right now I think that my proposed experiment would provide a pretty
cleat test of the difference between Martin's and my model.

I guess this topic is no longer of interest? I suppose that I'll just
have to continue running this gauntlet of hostile reviewers alone, as
usual. Sure would have liked some help.

Best

Rick

···

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

[Bruce_Gregory2]

[From Bruce Gregory (2010.07.31.1310 EDT)]

[From Rick Marken (2010.07.31.0940)]

I guess this topic is no longer of interest? I suppose that I’ll just

have to continue running this gauntlet of hostile reviewers alone, as

usual. Sure would have liked some help.

Courage, Camille.

[Bill_Powers1]

[From Bill Powers (2010.08.01.0855 MDT)]

Rick Marken (2010.07.31.0940) --

> I suppose I could try to make the experiment look a bit more like
> Martin's original forced choice experiment. I'll think about it. But
> right now I think that my proposed experiment would provide a pretty
> cleat test of the difference between Martin's and my model.

I guess this topic is no longer of interest? I suppose that I'll just
have to continue running this gauntlet of hostile reviewers alone, as
usual. Sure would have liked some help.

I'm in the middle of recovering from the trip to England and musing over the state of CSGnet. It seems to me that the center of gravity of scientific PCT has moved across the Atlantic, where dozens of students are absorbing it into their learning and research and young lively professors are teaching it in ways that encourage and inspire. It's still not the Center for the Study of Living Control Systems, but it's a few steps closer to it than anything going on in the CSG or on CSGnet.

I don't have anything in particular against the topics that have been discussed in the last month or so on CSGnet, but they aren't going in directions that interest me and I have little to contribute to them that would satisfy any of the current discussants. Applying PCT-like terminology can be useful, but if the underlying ideas are still the same old ones that have been around for 50 to 200 years it's not likely that anything new will be added to PCT as a result.

An old manuscript showed up while I was looking for something else; it was the first draft of what eventually became Making Sense of Behavior. The working title was "Starting Over." That's still what I want to do, though it's getting a little late in the day. I want to find out if there are really levels of control, and see the research started that will begin to establish their nature. I want to know if apparent instances of "up-side-down" levels are real -- that is, wanting to teach algebra to make enough money to eat. Could there be some organization other than a hierarchy? I want to know how reference signals held in memory drift with time. I want to know how we reorganize in new environments to perceive their properties and learn to control them. I want to know if there is really an organized environment Out There, and if so, whether the kinds of things we perceive are linked in any systematic way to properties of reality. I (like Richard Kennaway) want to build PCT robots, or see them getting built; I want to help develop a curriculum for teaching the basics on which PCT rests.

There is a lifetime of projects concerning living control systems waiting for someone to get interested in them. Most of the old problems will eventually fade away as people lose interest in them, and PCT will introduce new ones that more directly concern the question of how behavior really works, how it is really organized, what we can really do about human difficulties.

Right now I don't quite know what comes next. It will be different.

Best,

Bill P.

[Bruce_Gregory2]

[From Bruce Gregory (2010.08.01.1658 EDT)]

[From Bill Powers (2010.08.01.0855 MDT)]

I don't have anything in particular against the topics that have been discussed in the last month or so on CSGnet, but they aren't going in directions that interest me and I have little to contribute to them that would satisfy any of the current discussants. Applying PCT-like terminology can be useful, but if the underlying ideas are still the same old ones that have been around for 50 to 200 years it's not likely that anything new will be added to PCT as a result.

BG: I'm a bit surprised that addressing the question of how ongoing research in the social sciences and behavior economics would be different if the experimenters knew PCT and what they might learn that they do no now learn from redesigned studies does not interest you. Let me suggest the possibility that your lack of interest in such questions may be mirrored in the lack of interest in PCT on the part of traditional social scientists.

There is a lifetime of projects concerning living control systems waiting for someone to get interested in them. Most of the old problems will eventually fade away as people lose interest in them, and PCT will introduce new ones that more directly concern the question of how behavior really works, how it is really organized, what we can really do about human difficulties.

BG: It may be, of course, that the discovery of control, like the discovery of the mechanisms by which signals are transmitted in the brain, while significant in itself, does not lead to great advances in learning how behavior "really works." You can get pretty far by simply assuming that once neural networks become established, it takes energy to rewire them. The brain, it seems, tends to resist extensive rewiring (for perfectly good reasons).

Right now I don't quite know what comes next. It will be different.

BG: Bonne chance.

Bruce