Signal paths (was Controlled Variables (was...))

MartinT · April 10, 2009, 9:28pm

[Martin Taylor 2009.04.10.13.52]

Happy Easter, if that fits your background…

[From Bill Powers (200904.07.0706 MDT)]

Martin Taylor 2009.04.06.14.11 –

I’m not happy with my reply to this post yesterday. As I read it over
today, I see that I never succeeded in focusing on what bothers me, and
ended up saying things that are irrelevant or pointless. This
communication business is not as easy as it looks. I’m going to try
again.

So I’ll answer this one before the earlier one.

Let’s make another
generic
claim: “All human psychology is an attempt to find out something
about the way a mind works”, where “a mind” represents
everything that goes on between the sense organs and the muscular or
chemical output processes.

I now see the problem here. It is the implication that behavior starts
with an input and ends with an output, so “mind” mediates
between the sense organs and the muscular or chemical output processes.
I
don’t know if this is what you intended to say, but it’s what your
words
get turned into when they end up inside me.

The PCT view is very different. The muscular or chemical outputs are
adjusted by the mind so as to alter the sensory inputs and make all
that
is constructed from them match the states that the mind prefers. I’m
sure
you understand this as well as I do. Why is it, then, that your words
appear to imply the old theory in which inputs from the environment
cause
outputs to the environment and that is the end until the next cycle? Is
this only my peculiar way of listening, or is your language drawn from
an
older paradigm? But if the latter is the case, how can it be that you
show such a clear understanding of PCT and such a lucid way of
expressing
yourself when you’re explaining it to others?

Perhaps your way of listening combines with my way of expressing,
leading to misapprehensions. But perhaps there is another possibility?
Since you believe I have a clear understanding of PCT, could it be that
I sometimes see implications of PCT that you have not seriously
considered? Might those implication not sometimes be correct?
In looking for something quite different, I came across this, from
[Martin Taylor 971212 10:45]:
*"I think you really do have the right idea. But you don’t have to
assume I have the wrong one. I did say that I was only trying to state
the same as you, but from a different viewpoint, after all!

If the
meaning were what you assume,
What you assume I assume. Why didn’t you assume I mean what you mean,
instead? It’s the easier assumption, and the correct one, too. As has
often been the case in earlier discussions."*
Perhaps the consistency with which you choose to assume I mean the
sillier of two possibilities might explain something of the frustration
I often feel during our interactions. However, let’s proceed.
Here’s a thread of possibility.
I see four distinct “schools” of psychological thought over the last
100 years (ignoring clinical schools, as well as the Gestaltists, who
dealt primarily with what we would call perceptual input functions).
These, in a deliberate ordering, are Behaviourist, Cognitive
Psychologists, TOTE, and PCT. The ordering is based on how much of the
system they considered important. Call them B, C, T, and P.
All these schools have some things in common. The direction of signal
flow is one. All agree that signals flow from the sensory systems into
the mystery inside the body, and flow out of the body to the
environment by physical and chemical means. In other words, all agree
on there being sensory INPUT and muscular and chemical OUTPUT. Calling
these interfaces “input” and “output” signifies an arbitrary point of
view. They could as easily be called, respectively “output” from the
environment to the living entity and “input” to the environment from
the entity. But we don’t usually take that environment-centred point of
view. It’s a convention to say “sensory input” and “muscular output”,
possibly because most of us find it easier to imagine being in the
position of the living object of analysis rather than that of the
physical environment.
To me, the key difference among the schools of thought is on what
signal channels they presume to exist and the importance they assign to
those channels. In gross terms, calling the mush inside the head “mind”
or M, the Behaviourists considered anything inside M to be
unexaminable, and the only signal paths to be S->M and M->R. M
acted as a very high-dimensional function, so R = M(S).
Cognitive Psychologists did differentiate M into modules, among which
were modules that modelled the environment and could algorithmically
determine what actions would have what effects on the environment. Call
those modules of M, Me (which isn’t a bad menemonic, because “self”
might be found there). The rest of the “mind mush” we can call Mm. This
separation allows for more signal channels, so that, speaking in gross
terms, the important channels were S->Mm, Mm->Me, Me->Mm,
Mm->R. Of course there are lots of modules, but for this purpose,
they can all be combined
TOTE includes everything the cognitive psychologists did, but adds an
important extra: R->E->S. (where E is the environment)
Nevertheless, if I understand TOTE correctly, this added channel is
used only as a kind of check on the cognitive processes, so the
important channels still remain those of the Cognitive Psychologists.
PCT incorporates all the channels used by TOTE. The segregation of M
into Me and the ill-defined M of Cognitive Psychology is further
refined so that there really is no “mind mush” left except for issues
of detail. Another difference is that in the same way the Cognitive
Psychologists partitioned the “mind mush”, PCT partitions the
environment, in a way that provides a new signal path. The environment
in PCT consists of those aspects influenced by R and those aspects
influenced by independent sources, collectively called “D” for
“Disturbances”. The two sets of aspects influenced by R and by D are
highly overlapped, and it might well be that most of the aspects
influenced by R in any particular circumstance are also influenced by
D. Nevertheless, it is convenient to segregate the effects into two
distinct signal paths, R->E and D->E, which merge into a signal
path E->S. In thinking of the signal paths, I combine in the
notation all the different levels of control just as I do for the
modules of cognitive psychology, and treat levels at “logic” and above
as Me. In contrast to the Behaviourists, PCT considers it possible to
examine what might be happening inside M, by simulating it, using as
data the observables in the environment, and seeing whether the model
would have produced nearly the same data under the same conditions. If
it does, the real M might plausibly include functions homologous to
those used in the model.
The key difference between PCT and TOTE is that the acronym “TOTE”
finishes with “E” = “End”. In PCT thought, there is no “End”. Control
is continuous, so that the path S->M->R is completed by
R->E->S at all times that some element of S is actually being
controlled. This, together with the prominence of D, raises the
importance of the R->…->S pathway to equality with that of the
pathways inside M.
Here’s a small table summarizing all this:
School
B C T P
S->M x X
X X
M->Me x
X X
Me->M x
X X
M->R x X
X X
R->E
x X
D->E
X
E->S
x X
This table may be a little misleading, inasmuch as B takes R to
be a function of S, whereas the others do not. But I think the gist of
it is a reasonable cartoon of the main differences among the schools of
thought.
Since all of the schools use the signal paths S->M and M->R, and
do not use paths M->S and R->M (though such paths can and may
exist), I think it is not at all unreasonable, nor inconsistent with a
pure PCT view, to say: "All human psychology is an attempt to find
out something
about the way a mind works", where “a mind” represents
everything that goes on between the sense organs and the muscular or
chemical output processes.

Now let’s consider the pathways that are new in PCT, since it
goes back to an old thread (I don’t know how old – apparently it was
some years old in 1997 and I haven’t found it in Dag’s archives yet),
recently revived between Bruce A and Bill (e.g. From
Bruce Abbott (2009.04.10.1045 EDT)] – that only a small
selection of perceptions are at any moment being controlled. As Bruce
says: “In
a computer, the routines exist in memory
but are not actively doing anything until those stored instructions are
executed. I really don’t have a sense that my “spoon-to-mouth”
control system is active while I’m typing this message.”

The new pathways in PCT are the R->E, D->E, and E->S pathways.
It’s easy enough to list them like that, but to do so hides the fact
that R can use only a few heavy, slow movers, whereas E->S uses a
huge number of very fast sensors. That discrepancy ensures that only a
very small proportion of the possible perceptions at any one level of
the hierarchy cold be under control at any moment. That means one or
both of two things: (1) only a small proportion of the possible
perceptions at any level are actually produced, and (2) only a small
proportion of the perceptions that are produced are controlled. This
being the case, there must be some mechanism for switching control from
one perception at a level to another at that level. Discussion of that
mechanism is better treated in a different thread, perhaps the one Bill
and Bruce are pursuing. Here I want to consider a different implication.

Many perceptions are functions of more than one lower-level perception
or sensory input (according to HPCT and a nearly 200-year tradition in
psychology). However, the value of any one perception is a scalar (in
HPCT). The implication is that control of this multi-input perception
can often be achieved by manipulation of only one of its several
inputs, leaving the others to vary as they might. The other inputs, in
combination, come from the set of aspects of the environment called D
above, whereas the controlled one is from the overlap of D and R, or
perhaps of R alone.

We can describe this set of circumstances as P = p((s1, s2,…,sn), sc)
where sc is the sole input that is influenced by the person’s actions.
Another way of writing the same thing is sc = F(P, (s1, s2,…, sn)),
where the functional relationship is descriptive and carries no
implication of causality. When P is well under control, it is
approximately constant, so one can write, for any specific value of P,
sc = F((s1, s2,…,sn) | P). This is the situation for most of our
controlled perceptions, though sometimes we control two or rarely three
of the inputs. In most PCT experiments, it’s only one. We might be able
to move from controlling one of the inputs to controlling another, but
it’s not easy to maintain simultaneous control of two, let alone three.

This holds at any one level. But it also holds at the next higher
level. The lower level perception controlled by influencing one of its
several inputs is just one of several inputs to the next higher level
perception, and yet that higher level perception can be controlled
using a feedback path that influences only the single lower-level one.
In the diagram, the controlling scalar-all-the-way feedback loop is
shown black, the other influences on the controlled perceptions are
shown red.

Notice that these uncontrolled inputs to the perceptual functions are
not shown where disturbances are usually shown in a PCT diagram, but
they are nevertheless disturbances. They are disturbances that are not
countered in the environment by any of the controller’s actions, but
their effect on the controlled perceptions are nevertheless countered
by the control action. Collectively, to the degree to which control is
successful, at each level they are functions of the single control
input.

I think that’s about all I want to put into this message. I think my
main point is really the same one as is exemplified in my complaint of
12 years ago: “I think you really do have the right idea. But you don’t
have to assume I have the wrong one. … Why didn’t you assume I mean
what you mean, instead? It’s the easier assumption, and the correct
one, too. As has often been the case in earlier discussions.”

Why is it that when I say something that could with some stretching be
interpreted as being silly, the silly interpretation is what you so
often assume to be my meaning? You’ve been doing it for over a decade
and a half, and from time to time it has even led me to consider
leaving CSGnet. It is incredibly hard to write anything that can’t be
given a different interpretation, without resorting to the kind of
language used in legal documents. I don’t think you would like that,
even if I could manage it. Instead, I’d prefer it if you might use
interpretations based on what you said: “you
show such a clear understanding of PCT.”

Martin

rsmarken · April 11, 2009, 2:16am

[From Rick Marken (2009.04.10.1915)]

Martin Taylor (2009.04.10.13.52)–

Bill Powers (200904.07.0706 MDT)–

I now see the problem here. It is the implication that behavior starts
with an input and ends with an output, so “mind” mediates
between the sense organs and the muscular or chemical output processes.
I
don’t know if this is what you intended to say, but it’s what your
words
get turned into when they end up inside me.

The PCT view is very different. The muscular or chemical outputs are
adjusted by the mind so as to alter the sensory inputs and make all
that
is constructed from them match the states that the mind prefers.

I see four distinct “schools” of psychological thought …Behaviourist, Cognitive
Psychologists, TOTE, and PCT.

All these schools have some things in common. The direction of signal
flow is one [way?–RM]. All agree that signals flow from the sensory systems into
the mystery inside the body, and flow out of the body to the
environment by physical and chemical means.

Actually, one of those schools doesn’t see signals flowing that way; PCT sees reference signals flowing “outward” from the mind (specifying the “states that the mind prefers”, as Bill put it so nicely) toward the sensory signals that are flowing “inward” toward the mind.

To me, the key difference among the schools of thought is on what
signal channels they presume to exist… Behaviourists considered…the only signal paths to be S->M and M->R…

Cognitive Psychologists… [the signal paths] …were S->Mm, Mm->Me, Me->Mm,
Mm->R.

TOTE includes everything the cognitive psychologists did, but adds an
important extra: R->E->S.

The key difference between PCT and TOTE is … Control
is continuous, so that the path S->M->R is completed by
R->E->S at all times that some element of S is actually being
controlled.

What distinguishes PCT, I think, is that it is the only theory that recognizes the possible existence of the outgoing signal path, from reference, Ref, to sensory (really perceptual) input, S: Ref>S. The existence of this path means that PCT does not see signals flowing one way though the organism, from S > M > R. The signal flow through the organisms in PCT (using your notation) is:

<Ref <M
V
R

(I hope this prints out right: the R (response) is coming down from where S and Ref come together).

The signal path through the organism is not one way. Oh, and, of course, S is a controlled perceptual variable, something that is not mentioned in any other theory; not even in closed loop TOTE theory. Miller, Galanter and Pribrium just didn’t understand that closed loop control is the control of input. That’s why Bill called his book “Behavior: The control of perception”; because that’s the really revolutionary aspect of his theory: mind specifying what it wants to experience and acting to get experience that way. And, of course, you can only study such systems correctly using the appropriate methodology; testing for controlled variables rather than causal relationships.

Best regards

Rick

···

–
Richard S. Marken PhD

rsmarken@gmail.com

MartinT · April 11, 2009, 3:59am

[Martin Taylor 2009.04.10.23.35]

[From Rick Marken (2009.04.10.1915)]

Martin Taylor (2009.04.10.13.52)–

Bill Powers (200904.07.0706 MDT)–

I now see the problem here. It is the implication that behavior starts
with an input and ends with an output, so “mind” mediates
between the sense organs and the muscular or chemical output processes.
I
don’t know if this is what you intended to say, but it’s what your
words
get turned into when they end up inside me.

The PCT view is very different. The muscular or chemical outputs are
adjusted by the mind so as to alter the sensory inputs and make all
that
is constructed from them match the states that the mind prefers.

I see four distinct “schools” of psychological thought …Behaviourist,
Cognitive
Psychologists, TOTE, and PCT.

All these schools have some things in common. The direction of signal
flow is one [way?–RM]. All agree that signals flow from the sensory
systems into
the mystery inside the body, and flow out of the body to the
environment by physical and chemical means.

Actually, one of those schools doesn’t see signals flowing that way;
PCT sees reference signals flowing “outward” from the mind (specifying
the “states that the mind prefers”, as Bill put it so nicely) toward
the sensory signals that are flowing “inward” toward the mind.

Two, and I think three (I’m not 100% clear about TOTE – I think MGP
must be hidden behind several years of “The Magazine of Fantasy and
Science Fiction” and not readily accessible) see it as you describe.
Only the Behaviourists don’t. But the signals you mention are not
between the environment and the living organism. Physiologists also see
signals flowing outward to the sensory systems and inward from the
muscles. Those signals remain inside the skin.

To me, the key difference
among the schools of thought is on what
signal channels they presume to exist… Behaviourists considered…the
only signal paths to be S->M and M->R…

Cognitive Psychologists… [the signal paths] …were S->Mm,
Mm->Me, Me->Mm,
Mm->R.

TOTE includes everything the cognitive psychologists did, but adds an
important extra: R->E->S.

The key difference between PCT and TOTE is … Control
is continuous, so that the path S->M->R is completed by
R->E->S at all times that some element of S is actually being
controlled.

What distinguishes PCT, I think, is that it is the only theory that
recognizes the possible existence of the outgoing signal path, from
reference, Ref, to sensory (really perceptual) input, S: Ref>S. The
existence of this path means that PCT does not see signals flowing
one way though the organism, from S > M > R. The signal flow
through the organisms in PCT (using your notation) is:

<Ref <M
 V

 R
(I hope this prints out right: the R (response) is coming down from
where S and Ref come together).

Your diagram seems to single out the comparator in one individual
elementary control system, which is a component of my “M”. It’s fine to
do that. We do it implicitly whenever we talk about specific control
systems. But you really should have put M all round the bit you drew,
not just feeding the Reference level. That comparator gets its S from
another part of M and sends it R to yet another part of M. Furthermore,
the concept of a comparator isn’t unique to PCT. The Cognitive
Psychologists would have a very similar pattern, though they would use
a different name for the lowest variable. They would say: “This is what
I want to be the case; that is the current situation; given that
situation and knowing the way that pert of the world works, I can make
this plan that will get me to where I want to be.” The diagram would be
the same, except that the bottom letter would be P (for Plan). TOTE
also compares the state with what it should be, and the comparator
diagram would look the same.

With some stretching, I might also interpret the “V” in the diagram as
a control unit, but that wouldn’t alter the fact that the other two
schools would legitimately use the same diagram. They just wouldn’t
call “V” a control unit; a “Plan Unit”, perhaps. Anyway, it’s not
relevant to the point, and I don’t think it implies a problem with my
table of signal paths, since that was done to describe a more gross
level of analysis.

The signal path through the organism is not one way.

It is in PCT as far as the flow is FROM the environment into the mind
by way of the senses and TO the environment from the mind by way of the
muscles and chemical emissions. The directions I capitalized are what
was at issue, since the main reason for my message is that Bill claimed
that to say so was to accept a Behaviourist viewpoint. Internally,
there’s lots else going on, no matter which school you are discussing.
Even the behaviourists’ R = f(S) has a very complicated “f()”.

Oh, and, of course, S is a controlled perceptual
variable, something that is not mentioned in any other theory; not
even in closed loop TOTE theory. Miller, Galanter and Pribrium just
didn’t understand that closed loop control is the control of input.
That’s why Bill called his book “Behavior: The control of perception”;
because that’s the really revolutionary aspect of his theory: mind
specifying what it wants to experience and acting to get experience
that way.

Yep. Very true, and you quoted one bit of my message where I made that
point; but what does it have to do with the observation that in PCT, as
in all the other schools, the signals from and to the environment are
assumed to go into the sense organs and come out of the muscles? It’s
been rather a long time since people believed that vision was the
result of some emanation from the eyes!

And, of course, you can only study such systems correctly
using the appropriate methodology; testing for controlled variables
rather than causal relationships.

As you might well have expected, I don’t accept “of course” and “only”
as being anything other than expressions of religious faith. I have
come to expect any rational argument that there are other valid methods
to be treated very much as a Christian fundamentalist would treat
archaeological methods of looking for the truths behind the Gospels, or
as they do treat the evidence on how evolution works. In other words, I
expect critiques to be of the kind “the only appropriate methodology is
testing for the controlled variable” no matter what other methods I
might describe. And I expect those critiques to be strongly worded, and
not to take account of the details of whatever I might have suggested.
But of course, this is all prediction, based on a linear extrapolation
of the past, and the future is notoriously unpredictable

I suppose there are things that I take as articles of faith, too, such
as the coherence of the Universe and others things of which I may not
be aware, but even so, I don’t have to accept your faiths as necessary
truths. In fact, I think yours are self-contradictory, there being a
very clean argument that you can’t test for controlled variables unless
you accept that the relationships throughout every control loop are
causal.

Martin

Bill_Powers1 · April 11, 2009, 6:57am

[From Bill Powers (2009.04.10.2300 MDT)]

Martin Taylor 2009.04.10.13.52 –

Perhaps your way of listening
combines with my way of expressing, leading to misapprehensions. But
perhaps there is another possibility? Since you believe I have a clear
understanding of PCT, could it be that I sometimes see implications of
PCT that you have not seriously considered? Might those implication not
sometimes be correct?

Certainly, I can agree that you may see things about PCT that I haven’t
seen yet. I often have trouble seeing immediately where my own thoughts
are heading, much less someone else’s, especially when I’m listening to
persuasive and well-reasoned arguments. I tend to get caught up in the
other person’s assumptions, often not even realizing at first what I am
accepting that, later on, I might wish I hadn’t been so quick to accept.
But sometimes, of course, I might wish I had been quicker.

In looking for something quite
different, I came across this, from [Martin Taylor 971212
10:45]:
*"I think you really do have the right idea. But you don’t have
to assume I have the wrong one. I did say that I was only trying to
state the same as you, but from a different viewpoint, after
all!

If the
meaning were what you assume,
What you assume I assume. Why didn’t you assume I mean what you mean,
instead? It’s the easier assumption, and the correct one, too. As has
often been the case in earlier discussions."*

Perhaps the consistency with which you choose to assume I mean the
sillier of two possibilities might explain something of the frustration I
often feel during our interactions.

Yes, that would be most unjust. But part of that problem is simply my
style; I sometimes argue ad absurdum, pointing out what I think
are unwanted implications of assumptions by presenting the implications
as if the assumer were supporting them, which I know he is not. If you
don’t see those implications of the assumptions (or if I have made a
mistake in deducing the implications), it will look to the assumer as if
I have said he believes an absdurdity. I hope that’s the real explanation
– I’ll try to remember to say what I’m doing when I do that. Or maybe I
should just give up that kind of argument; it seems to backfire more
often than it works.

I see four distinct
“schools” of psychological thought over the last 100 years
(ignoring clinical schools, as well as the Gestaltists, who dealt
primarily with what we would call perceptual input functions). These, in
a deliberate ordering, are Behaviourist, Cognitive Psychologists, TOTE,
and PCT. The ordering is based on how much of the system they considered
important. Call them B, C, T, and P.

All these schools have some things in common. The direction of signal
flow is one. All agree that signals flow from the sensory systems into
the mystery inside the body, and flow out of the body to the environment
by physical and chemical means.

Good. While I disagree with your classification, it is a very useful
starting point and it clarified for me how to make the most important
difference between PCT and the other three clear – even though it has
taken the better part of the day, and a nap, to see what the issue is. I
think Rick saw most of it before I did.

In other words, all agree
on there being sensory INPUT and muscular and chemical
OUTPUT.

If I point out that the real problem is in this assumed agreement, will
that jog you enough to see what I’m going to say before I say
it?
Let’s consider the tracking experiment we’ve been discussing. What is the
sensory input, and what is the motor output? And what is the part of the
Mind that we’re interested in?
Here’s what the subject sees on the screen:

==== ====

<—>
The equal signs are the (split) target that moves up and down, and the
line with arrows at both ends is the cursor that also moves up and down.
We can easily establish that the controlled variable in this experiment
is some aspect of this display. That aspect, presumably, is the input
variable of interest. We determine its reference condition by seeing what
condition would generate no behavior to correct it.
I think you will agree that the controlled variable is the vertical
distance between the target and the cursor. The subject can set a
reference condition of zero distance, or any positive or negative
distance, and defend it against disturbances. The above diagram might
show the reference condition of the display. So in your chart, the
sensory input is this vertical distance, rather than the cursor alone or
the target alone.
The behavior of the subject is to change the position of a mouse by
pushing it away or pulling it closer. This action is indicated by a
number in the computer which is updated 60 times per second, and also by
the position of the cursor which is updated at the same time. So we have
the following definitions, in simplified form:
input: Cursor position - Target position
mind: integral((reference - input))
output: mouse position or cursor position
I think you can see the problem immediately. Neither of the objects on
the screen, the target or the cursor, is the input to the aspect of the
mind we are discussing. Yet the behaviorist would probably assume that
his manipulation of the target position is the input.
Whether we say there are two inputs or that the difference between them
is the input, the input is not independent of the output. The
experimenter can move the target to change the input, but that movement
is not the actual input; the actual input is cursor - target. As the
experimenter begins to change the target position and before he finishes
changing it, the behavior begins to change, causing the cursor position
to change and reducing the amount of change in the actual input variable.
If the experimenter moves the target 10 centimeters in 500 milliseconds,
by the time the movement is finished the cursor will have moved perhaps 8
centimeters, so the input quantity has changed only by 2 centimeters.
Inside another 500 milliseconds, that change in the input quantity has
dropped back nearly to zero as the cursor approaches the target. What the
experimenter thought was a step-change in the input was actually a brief
impulse with an initial magnitude of 10 centimeters followed by a rapid
return to zero.
So the stimulus that the behaviorist applies is not the actual input to
the subject. The actual input is a function of the stimulus and the
behavior.

The usual objection to this analysis is that the behavior is delayed, so
the stimulus occurs first, and then the behavior follows, producing the
next stimulus. This way of seeing it can be made to appear irrefutable
simply by making the target jump to another position followed immediately
by a return to the initial position, the whole stimulus event lasting
around 100 milliseconds. What will then be observed is a response of the
mouse and cursor toward the displaced position of the target, starting
about 150 - 200 milliseconds after the start of the target event and
returning to zero immediately. The entire stimulus event is finished
before the response even starts. So now we have the appearance of a
discrete stimulus event followed after some delay by a discrete target
event.

Notice how changing the experimental conditions changes the appearance of
what is happening, and how it suggests an entirely different organization
inside the subject. Clearly, the input cannot be a function of the output
if the input is finished before the output even starts. In fact, if one
uses discrete stimuli as a matter of course, feedback effects that modify
the input while it is happening will not even be suspected; instead, the
concept of “chaining” will be used; a stimulus gives rise to
the next stimulus, or the response gives rise to the next stimulus. An
excellent example of this is the knee-jerk reflex that I have mentioned
before. The brief force impulse from the hammer is finished before the
kick even starts. Yet we know that we are looking at the behavior of a
continuously-active control system.

Notice, too, that in the “cognitive” model, the environment
still plays a part in generating the behavior: it is the analysis of the
environment that is used to compute what commands are required to produce
a desired result. So mistaking the distal stimulus for the actual input
is just as misleading as it is for the behaviorist. The TOTE unit, which
monitors input variables during the Test phase, does this even more
explicitly.

As Rick pointed out, the biggest difference between PCT and the other
three approaches you describe is in defining the input to the mind. I
tried to allude to this but the point slipped past without comment. I
mentioned a possible variant on the Schouten experiment in which the
subject operates a continuous rheostat rather than simply clicking a
button. That would slow down the process enough to make it clear that the
action was affecting the light intensity at the same time the light
intensity was affecting the action. Then we would see that the level that
decides to respond to a light has to sense the state of the light rather
than responding open-loop. And that in turn would show that the
underlying system we are studying is not an open-loop system at any
level. In the Schouten experiment, we are seeing a control system
operating under conditions which artificially limit its behavior to what
seems a stimulus-response form – but there is no reason to think that is
what is really there.

You have repeatedly asserted that the Schouten experiment shows us an
open-loop response at the input. But the input you defined would not be
the input that actually enters the mind, because you have not given the
behavior its proper effect on that input. The subject would continue to
observe the relationship between the button and the lights even after the
data have been recorded for one trial. Of course the behavior can’t
affect the light input any more than the cursor position can affect the
target position in the tracking experiment. But the controlled variable
is the relationship, which is a function of both behavior and light
intensity. An experiment which allowed time for the control system to
function normally would reveal that.

Over to you.

Best,

Bill P.

···

Calling these interfaces
“input” and “output” signifies an arbitrary point of
view. They could as easily be called, respectively “output”
from the environment to the living entity and “input” to the
environment from the entity. But we don’t usually take that
environment-centred point of view. It’s a convention to say “sensory
input” and “muscular output”, possibly because most of us
find it easier to imagine being in the position of the living object of
analysis rather than that of the physical environment.
To me, the key difference among the schools of thought is on what signal
channels they presume to exist and the importance they assign to those
channels. In gross terms, calling the mush inside the head
“mind” or M, the Behaviourists considered anything inside M to
be unexaminable, and the only signal paths to be S->M and M->R. M
acted as a very high-dimensional function, so R = M(S).
Cognitive Psychologists did differentiate M into modules, among which
were modules that modelled the environment and could algorithmically
determine what actions would have what effects on the environment. Call
those modules of M, Me (which isn’t a bad menemonic, because
“self” might be found there). The rest of the “mind
mush” we can call Mm. This separation allows for more signal
channels, so that, speaking in gross terms, the important channels were
S->Mm, Mm->Me, Me->Mm, Mm->R. Of course there are lots of
modules, but for this purpose, they can all be combined
TOTE includes everything the cognitive psychologists did, but adds an
important extra: R->E->S. (where E is the environment)
Nevertheless, if I understand TOTE correctly, this added channel is used
only as a kind of check on the cognitive processes, so the important
channels still remain those of the Cognitive Psychologists.
PCT incorporates all the channels used by TOTE. The segregation of M into
Me and the ill-defined M of Cognitive Psychology is further refined so
that there really is no “mind mush” left except for issues of
detail. Another difference is that in the same way the Cognitive
Psychologists partitioned the “mind mush”, PCT partitions the
environment, in a way that provides a new signal path. The environment in
PCT consists of those aspects influenced by R and those aspects
influenced by independent sources, collectively called “D” for
“Disturbances”. The two sets of aspects influenced by R and by
D are highly overlapped, and it might well be that most of the aspects
influenced by R in any particular circumstance are also influenced by D.
Nevertheless, it is convenient to segregate the effects into two distinct
signal paths, R->E and D->E, which merge into a signal path
E->S. In thinking of the signal paths, I combine in the notation all
the different levels of control just as I do for the modules of cognitive
psychology, and treat levels at “logic” and above as Me. In
contrast to the Behaviourists, PCT considers it possible to examine what
might be happening inside M, by simulating it, using as data the
observables in the environment, and seeing whether the model would have
produced nearly the same data under the same conditions. If it does, the
real M might plausibly include functions homologous to those used in the
model.
The key difference between PCT and TOTE is that the acronym
“TOTE” finishes with “E” = “End”. In PCT
thought, there is no “End”. Control is continuous, so that the
path S->M->R is completed by R->E->S at all times that some
element of S is actually being controlled. This, together with the
prominence of D, raises the importance of the R->…->S pathway to
equality with that of the pathways inside M.
Here’s a small table summarizing all this:

School
B
C
T
P
S->M
x
X
X
X
M->Me
x
X
X
Me->M
x
X
X
M->R
x
X
X
X
R->E
x
X
D->E
X
E->S
x
X
This table may be a little misleading, inasmuch as B takes R to be
a function of S, whereas the others do not. But I think the gist of it is
a reasonable cartoon of the main differences among the schools of
thought.
Since all of the schools use the signal paths S->M and M->R, and do
not use paths M->S and R->M (though such paths can and may exist),
I think it is not at all unreasonable, nor inconsistent with a pure PCT
view, to say: "All human psychology is an attempt to find out
something about the way a mind works", where “a mind”
represents everything that goes on between the sense organs and the
muscular or chemical output processes.

Now let’s consider the pathways that are new in PCT, since it goes
back to an old thread (I don’t know how old – apparently it was some
years old in 1997 and I haven’t found it in Dag’s archives yet), recently
revived between Bruce A and Bill (e.g. From Bruce Abbott (2009.04.10.1045
EDT)] – that only a small selection of perceptions are at any moment
being controlled. As Bruce says: “In a computer, the routines exist
in memory but are not actively doing anything until those stored
instructions are executed. I really donï¿½t have a sense that my
ï¿½spoon-to-mouthï¿½ control system is active while Iï¿½m typing this
message.”

The new pathways in PCT are the R->E, D->E, and E->S pathways.
It’s easy enough to list them like that, but to do so hides the fact that
R can use only a few heavy, slow movers, whereas E->S uses a huge
number of very fast sensors. That discrepancy ensures that only a very
small proportion of the possible perceptions at any one level of the
hierarchy cold be under control at any moment. That means one or both of
two things: (1) only a small proportion of the possible perceptions at
any level are actually produced, and (2) only a small proportion of the
perceptions that are produced are controlled. This being the case, there
must be some mechanism for switching control from one perception at a
level to another at that level. Discussion of that mechanism is better
treated in a different thread, perhaps the one Bill and Bruce are
pursuing. Here I want to consider a different implication.

Many perceptions are functions of more than one lower-level perception or
sensory input (according to HPCT and a nearly 200-year tradition in
psychology). However, the value of any one perception is a scalar (in
HPCT). The implication is that control of this multi-input perception can
often be achieved by manipulation of only one of its several inputs,
leaving the others to vary as they might. The other inputs, in
combination, come from the set of aspects of the environment called D
above, whereas the controlled one is from the overlap of D and R, or
perhaps of R alone.

We can describe this set of circumstances as P = p((s1, s2,…,sn), sc)
where sc is the sole input that is influenced by the person’s actions.
Another way of writing the same thing is sc = F(P, (s1, s2,…, sn)),
where the functional relationship is descriptive and carries no
implication of causality. When P is well under control, it is
approximately constant, so one can write, for any specific value of P, sc
= F((s1, s2,…,sn) | P). This is the situation for most of our
controlled perceptions, though sometimes we control two or rarely three
of the inputs. In most PCT experiments, it’s only one. We might be able
to move from controlling one of the inputs to controlling another, but
it’s not easy to maintain simultaneous control of two, let alone
three.

This holds at any one level. But it also holds at the next higher level.
The lower level perception controlled by influencing one of its several
inputs is just one of several inputs to the next higher level perception,
and yet that higher level perception can be controlled using a feedback
path that influences only the single lower-level one. In the diagram, the
controlling scalar-all-the-way feedback loop is shown black, the other
influences on the controlled perceptions are shown red.

Notice that these uncontrolled inputs to the perceptual functions are not
shown where disturbances are usually shown in a PCT diagram, but they are
nevertheless disturbances. They are disturbances that are not countered
in the environment by any of the controller’s actions, but their effect
on the controlled perceptions are nevertheless countered by the control
action. Collectively, to the degree to which control is successful, at
each level they are functions of the single control input.

I think that’s about all I want to put into this message. I think my main
point is really the same one as is exemplified in my complaint of 12
years ago: “I think you really do have the right idea. But you don’t
have to assume I have the wrong one. … Why didn’t you assume I mean
what you mean, instead? It’s the easier assumption, and the correct one,
too. As has often been the case in earlier discussions.”

Why is it that when I say something that could with some stretching be
interpreted as being silly, the silly interpretation is what you so often
assume to be my meaning? You’ve been doing it for over a decade and a
half, and from time to time it has even led me to consider leaving
CSGnet. It is incredibly hard to write anything that can’t be given a
different interpretation, without resorting to the kind of language used
in legal documents. I don’t think you would like that, even if I could
manage it. Instead, I’d prefer it if you might use interpretations based
on what you said: “you show such a clear understanding of
PCT.”

Martin

No virus found in this incoming message.

Checked by AVG -
www.avg.com

Version: 8.0.238 / Virus Database: 270.11.51/2052 - Release Date:
04/10/09 06:39:00

rsmarken · April 11, 2009, 3:58pm

[From Rick Marken (2009.04.11.0900)]

Bill Powers (2009.04.10.2300 MDT)–

This post merits a paper of its own. I’m sure you have better things to do so I’ll try to sketch one out myself as soon as I can get to it. This is the real heart of the matter for me:

This way of seeing it [as S-R – RM] can be made to appear irrefutable
simply by making the target jump to another position followed immediately
by a return to the initial position, the whole stimulus event lasting
around 100 milliseconds…So now we have the appearance of a
discrete stimulus event followed after some delay by a discrete target
event.

Notice how changing the experimental conditions changes the appearance of
what is happening, and how it suggests an entirely different organization
inside the subject… In fact, if one
uses discrete stimuli as a matter of course, feedback effects that modify
the input while it is happening will not even be suspected…An
excellent example of this is the knee-jerk reflex that I have mentioned
before. The brief force impulse from the hammer is finished before the
kick even starts. Yet we know that we are looking at the behavior of a
continuously-active control system.

…In the Schouten experiment, we are seeing a control system
operating under conditions which artificially limit its behavior to what
seems a stimulus-response form – but there is no reason to think that is
what is really there.

I wish I had included this kind of discussion in my “Revolution” paper but, of course, I couldn’t because I had never had it pointed out so clearly before. (This post from Bill certainly shows the incredible scientific value of CSGNet, for me anyway). But it could make a great – hugely great – paper. It explains, better than anything I said in the “Revolution” paper, why psychologists (of all stripes) have been able to labor under the illusion that organisms are S-M-R (using Martin’s notation) systems. It’s because they set up their experiments using stimuli that come and go so quickly (like the people in Oz) that the situation is guaranteed to make their controlling look like S-R.

Thanks, Bill!

Best

Rick

PS. I computed the data for my experiment on difficulty level incorrectly; it turns out that when I do it correctly a one level model with the same Gain and Dampling accounts for the behavior in all three difficulty conditions.

···

–
Richard S. Marken PhD
rsmarken@gmail.com

MartinT · April 14, 2009, 4:41am

[Martin Taylor 2009.04.14.00.04]

[From Bill Powers (2009.04.10.2300 MDT)]

Martin Taylor 2009.04.10.13.52 –

I see four distinct
“schools” of psychological thought over the last 100 years
(ignoring clinical schools, as well as the Gestaltists, who dealt
primarily with what we would call perceptual input functions). These,
in
a deliberate ordering, are Behaviourist, Cognitive Psychologists, TOTE,
and PCT. The ordering is based on how much of the system they
considered
important. Call them B, C, T, and P.

All these schools have some things in common. The direction of signal
flow is one. All agree that signals flow from the sensory systems into
the mystery inside the body, and flow out of the body to the
environment
by physical and chemical means.

Good. While I disagree with your classification, it is a very useful
starting point and it clarified for me how to make the most important
difference between PCT and the other three clear – even though it has
taken the better part of the day, and a nap, to see what the issue is.
I
think Rick saw most of it before I did.

In other words, all
agree
on there being sensory INPUT and muscular and chemical
OUTPUT.

If I point out that the real problem is in this assumed agreement, will
that jog you enough to see what I’m going to say before I say
it?

You know I will have read further before I post my response, but I am
writing this before reading further. At this point I cannot believe you
are going to argue that in PCT there is output from the senses to the
environment and input to the muscles from the environment. But that is,
after all, the only other possibility. So “No” I cannot see where you
are going before I read how you will come to this remarkable conclusion.

Let’s consider the tracking experiment we’ve been discussing. What is
the
sensory input, and what is the motor output? And what is the part of
the
Mind that we’re interested in?

Here’s what the subject sees on the screen:

==== ====

<—>

The equal signs are the (split) target that moves up and down, and the
line with arrows at both ends is the cursor that also moves up and
down.
We can easily establish that the controlled variable in this experiment
is some aspect of this display. That aspect, presumably, is the input
variable of interest. We determine its reference condition by seeing
what
condition would generate no behavior to correct it.

In PCT, yes. I suppose it would be possible to figure out what the
other schools might say about it, but there wouldn’t be much point,
would there?

I think you will agree that the controlled variable is the vertical
distance between the target and the cursor. The subject can set a
reference condition of zero distance, or any positive or negative
distance, and defend it against disturbances. The above diagram might
show the reference condition of the display. So in your chart, the
sensory input is this vertical distance, rather than the cursor alone
or
the target alone.

No. In my diagram, the sensory input is a pattern of light and dark on
the retina. The vertical distance is part of what I called Me, the
computed stuff. But no matter, let’s proceed.

The behavior of the subject is to change the position of a mouse by
pushing it away or pulling it closer. This action is indicated by a
number in the computer which is updated 60 times per second, and also
by
the position of the cursor which is updated at the same time. So we
have
the following definitions, in simplified form:

input: Cursor position - Target position

mind: integral((reference - input))

output: mouse position or cursor position

I think you can see the problem immediately. Neither of the objects on
the screen, the target or the cursor, is the input to the aspect of the
mind we are discussing.

You may be discussing “aspects of the mind”, but I was not. I believe
you must have read my message, in which I listed a very small number of
gross signal pathways admitted by the different schools. In it, you
would not have seen any reference to “aspects of mind” other than a
generic segregation that becomes more and more explicit as one moves
from behaviourist to Cognitive, to TOTE, to PCT.

Yet the behaviorist would probably assume that
his manipulation of the target position is the input.

I imagine so.

Whether we say there are two inputs or that the difference between them
is the input, the input is not independent of the output.

I know we agree that the other three schools are demonstrably
inadequate in dealing with even this simple situation. I’m afraid I
still don’t see where you are going, how you are going to show that you
don’t agree that input is through the senses and output through the
muscles.

So the stimulus that the behaviorist applies is not the actual input to
the subject. The actual input is a function of the stimulus and the
behavior.

Yes. Does that input not go in through the senses? Is not the signal
path within the person FROM the senses rather than TO the senses?

Notice, too, that in the “cognitive” model, the environment
still plays a part in generating the behavior: it is the analysis of
the
environment that is used to compute what commands are required to
produce
a desired result. So mistaking the distal stimulus for the actual input
is just as misleading as it is for the behaviorist. The TOTE unit,
which
monitors input variables during the Test phase, does this even more
explicitly.

Yes. All true. But I still haven’t seen how you are going to get to the
position you said up top that you were going to, and I can already see
the end of your message.

As Rick pointed out, the biggest difference between PCT and the other
three approaches you describe is in defining the input to the mind.

I think it was I who pointed it out before Rick. Might it be worthwhile
to quote from my message: “In PCT thought, there is no “End”. Control
is continuous, so that the path S->M->R is completed by
R->E->S at all times that some element of S is actually being
controlled. This, together with the prominence of D, raises the
importance of the R->…->S pathway to equality with that of the
pathways inside M.”

I
tried to allude to this but the point slipped past without comment. I
mentioned a possible variant on the Schouten experiment in which the
subject operates a continuous rheostat rather than simply clicking a
button. That would slow down the process enough to make it clear that
the
action was affecting the light intensity at the same time the light
intensity was affecting the action.

Remember that in the Schouten experiment, the subject has NO influence
on the light. You are proposing a quite different experiment, one that
would no doubt elucidate some questions, but it would not get at what
Schouten was getting at, nor would it give the kinds of results that
interested me about it – namely the linear increase in information
about the light that is manifest in the proportion of correct responses
as a function of delay time.

Then we would see that the level that
decides to respond to a light has to sense the state of the light
rather
than responding open-loop. And that in turn would show that the
underlying system we are studying is not an open-loop system at any
level. In the Schouten experiment, we are seeing a control system
operating under conditions which artificially limit its behavior to
what
seems a stimulus-response form – but there is no reason to think that
is
what is really there.

No, indeed. And in none of my diagrams dealing with different aspects
of the experiment has there been any such suggestion. You are knocking
down straw men with great facility. I claim only that in the Schouten
experiment and in many detection and discrimination experiments, the
subject is unable to influence the presentation (the disturbing
variable) in any way.

You have repeatedly asserted that the Schouten experiment shows us an
open-loop response at the input.

I would make a slight edit to that sentence, replacing “repeatedly”
with “never”. That would be rather more accurate.

I think the rest of my message, on which you did not comment, makes
this even more explicit. Rather than repeating why that is so, I will
simply present again the diagram that accompanied it.

Having now read your entire message, I’m leaving my interpolated
comments as they were when I made them during a read-through of your
message. Having finished the message I cannot make any sense of your
comment up front:

[MT] In other words, all agree
on there being sensory INPUT and muscular and chemical
OUTPUT.

[BP]If I point out that the real problem is in this assumed
agreement, will
that jog you enough to see what I’m going to say before I say
it?

In this diagram, is the signal flow from A to C by A>B>C or

C ?

Martin

rsmarken · April 14, 2009, 6:21am

[From Rick Marken (2009.04.13.2320)]

Martin Taylor (2009.04.14.00.04) to Bill Powers (2009.04.10.2300 MDT)

Thanks for answering Bill’s post. I’ll let Bill handle it on his own – what gems it yields – but I just have to remark on one point that you keep making that I think is rather misleading:

Remember that in the Schouten experiment, the subject has NO influence
on the light.

You say this as though this is what distinguishes the Schouten experiment from other psychological experiments, making it possible to learn about the perceptual transfer function from the relationship between light onset and the subject’s response… But the onset of the light is an independent variable (IV) in this experiment and like the IV in every psychological experiment subjects have no influence over it.

So the Schouten experiment is no different than any other conventional experiment. The light onset is precisely equivalent to the rubber hammer that is use to tap the patellar tendon to produce the “knee jerk” reflex. The knee jerk response, like the button press in the Schouten experiment, looks like an open loop response to the IV (tap strength or light onset) over which the subject has no influence. But in both cases we know that the system under study is a control system and that the apparent open loop response to the IV is actually a control action that occurs too late to correct for the effects of the IV (tap or light). The IV in both cases is (as Bill explained) a sudden and, thus, uncorrectable disturbance to a controlled variable (tendon length on the knee jerk, light-press relationship in the Schouten experiment).

It is by using such disturbances that psychologists (like Schouten, I’m afraid) have been able to maintain the illusion that they are dealing with an S-R system. But, as with all conventional experiments, the Schouten experiment tells us no more about the open loop characteristics of the light perception system than the knee jerk experiment tells us about the open loop characteristics of the tendon length perception system.

Best regards

Rick

···

–
Richard S. Marken PhD
rsmarken@gmail.com

MartinT · April 14, 2009, 2:42pm

[Martin Taylor 2009.04.14.10.29]

[From Rick Marken (2009.04.13.2320)]

Martin Taylor
(2009.04.14.00.04) to Bill Powers (2009.04.10.2300 MDT)

Thanks for answering Bill’s post. I’ll let Bill handle it on his own –
what gems it yields – but I just have to remark on one point that you
keep making that I think is rather misleading:

Remember that in the Schouten
experiment, the subject has NO influence
on the light.

You say this as though this is what distinguishes the Schouten
experiment from other psychological experiments, making it possible to
learn about the perceptual transfer function from the relationship
between light onset and the subject’s response… But the onset of the
light is an independent variable (IV) in this experiment and like the
IV in every psychological experiment subjects have no influence over
it.

Not all psychological experiments, and not all detection and
discrimination experiments. Many of them do allow the subject to
influence the presentation and thereby to control their perception of
it. On the other hand, many do not, and the Schouten study is one of
them. A pursuit tracking study is also of this class.

So the Schouten experiment is no different than any other conventional
experiment. The light onset is precisely equivalent to the rubber
hammer that is use to tap the patellar tendon to produce the “knee
jerk” reflex. The knee jerk response, like the button press in the
Schouten experiment, looks like an open loop response to the IV (tap
strength or light onset) over which the subject has no influence.

In both the Schouten study and the knee-jerk response, the subject DOES
have control of the perception of the response, yes. In the Schouten
case it is the button choice that the subject can control; in the
knee-jerk case it is the tendon-length that the subject can control. In
the Schouten case the subject cannot influence the perception of which
light was lit; in the knee-jerk case the subject cannot influence the
perception of being tapped.

But in both cases we know that the system under study is a
control system and that the apparent open loop response to the IV is
actually a control action that occurs too late to correct for the
effects of the IV (tap or light).

If you simply imagine either the knee-jerk or the Schouten-type
discrimination experiment with an infinitely fast control loop, how
would the response in any way affect the perception of the presentation
together with the instantaneous feedback? In the case of the knee-jerk,
what would be physically influenced? The knee angle, I would guess. How
would that influence the perception of being tapped? In the case of the
Schouten study, or others of the kind, what effect on the light would
the immediate (and presumably correct) button push have?

There is no relation between the speed of the presentation and of the
feedback loop and the fact that the response has no possible effect on
the perception of the presentation. To bring this up is a total red
herring, designed to divert people’s attention from the issues at hand.

The IV in both cases is (as Bill explained) a sudden and,
thus, uncorrectable disturbance to a controlled variable (tendon length
on the knee jerk, light-press relationship in the Schouten experiment).

It is by using such disturbances that psychologists (like Schouten, I’m
afraid) have been able to maintain the illusion that they are dealing
with an S-R system. But, as with all conventional experiments, the
Schouten experiment tells us no more about the open loop
characteristics of the light perception system than the knee jerk
experiment tells us about the open loop characteristics of the tendon
length perception system.

Do you see why this claim is wrong – totally wrong, not just a little
wrong?

Martin

rsmarken · April 14, 2009, 3:39pm

[From Rick Marken (2009.04.14.0840)]

Martin Taylor (2009.04.14.10.29)–

Rick Marken (2009.04.13.2320)]

But the onset of the
light is an independent variable (IV) in this experiment and like the
IV in every psychological experiment subjects have no influence over
it.

Not all psychological experiments, and not all detection and
discrimination experiments. Many of them do allow the subject to
influence the presentation and thereby to control their perception of
it.

In which case the “presentation” is not an IV. For example, in the method of adjustment, the intensity of the light is adjusted by the subject; it is not independent of the subject’s actions so it is not an IV. In fact, it would be considered a DV in a conventional experiment. The IV would be the intensity of the background light or of a comparison light or something like that. An IV is, by definition, independent of influence by the subject. This is true in every experiment.

On the other hand, many do not, and the Schouten study is one of
them. A pursuit tracking study is also of this class.

Both pursuit and compensatory tracking tasks are “in this class” (experiments where the subject has no influence over the IV), as are all experiments, even properly conducted tests for the controlled variable (where the IV is properly recognized as a possible disturbance to a controlled variable). Perhaps you are thinking of an IV as being the same as an input to the system. This is not the case. An IV is a disturbance to a controlled input (CV). The “direction of effect” looks like this:

IV—> input (CV)<—DV.

Note that the direction of effect is not one way. The IV (disturbance) might be a visible component of the CV (as it is in pursuit tracking and the Schouten experiment) or it might not (as in compensatory tracking). But the IV (disturbance) in an experiment is always independent of the subject’s actions; the CV (which is a function of both IV and actions) is not.

The knee jerk response, like the button press in the
Schouten experiment, looks like an open loop response to the IV (tap
strength or light onset) over which the subject has no influence.

In both the Schouten study and the knee-jerk response, the subject DOES
have control of the perception of the response, yes.

No, they have control over a controlled variable (CV). The response (knee jerk or button press) is the action the would protect the CV from disturbance if the disturbance did not come and go so quickly.

If you simply imagine either the knee-jerk or the Schouten-type
discrimination experiment with an infinitely fast control loop, how
would the response in any way affect the perception of the presentation
together with the instantaneous feedback?

The response does not affect the presentation; it affect the CV. The “knee jerk” brings the patellar tendon’s length (the CV) back to the reference length after the disturbance created by the uninfluenable “presentation” of the tap; the “button press” brings the relationship between light onset and button press (the CV) back to the reference relationship after the disturbance created by the uninfluencable presentation of the button light.

There is no relation between the speed of the presentation and of the
feedback loop and the fact that the response has no possible effect on
the perception of the presentation. To bring this up is a total red
herring, designed to divert people’s attention from the issues at hand.

It’s not just the speed of presentation that prevents the response from having an effect; in the case of the knee jerk it is the immediate removal of the force from the tendon that does it. In the Schouten study it’s the abruptness of the onset that makes the press look like a response to the presentation; the press is actually probably effective at keeping the CV (the relationship between presentation and press) under control.

It is by using such disturbances that psychologists (like Schouten, I’m
afraid) have been able to maintain the illusion that they are dealing
with an S-R system. But, as with all conventional experiments, the
Schouten experiment tells us no more about the open loop
characteristics of the light perception system than the knee jerk
experiment tells us about the open loop characteristics of the tendon
length perception system.

Do you see why this claim is wrong – totally wrong, not just a little
wrong?

No. I see that it’s right – totally right, not just a little right. I think you would see it too if you worked a bit on understanding the difference between an IV and a CV in an actual, real world experiment, like Schouten’s.

Best regards

Rick

···

–
Richard S. Marken PhD
rsmarken@gmail.com

Bill_Powers1 · April 14, 2009, 5:41pm

[From Bill Powers (2009.04.16.0854 MDT)]

Martin Taylor 2009.04.14.00.04 –

I’m answering this out of sequence because I made a bad mistake and need
to correct it, and because it’s clear that I didn’t make my point
previously.

The mistake was in saying

BP then:

I mentioned a possible
variant on the Schouten experiment in which the subject operates a
continuous rheostat rather than simply clicking a button. That would slow
down the process enough to make it clear that the action was affecting
the light intensity at the same time the light intensity was affecting
the action.

BP Now:

Instead of “light intensity” I should have said “perceived
variable.” With the continuous control involving the rheostat, the
presented light intensity is not what the subject sees: the subject sees,
at the relevant level, the net result of the presented light intensity
and the setting of the rheostat. The presented light intensity is the
same as the perceived variable only for as long as it takes to start
turning the rheostat. I’ll get back to the Schouten setup later, in which
the environmment is arranged to make sure that the action can’t affect
the light intensity during the defined period of one trial.

One point I’ve been trying to make is that what the experimenter presents
to the subject is not necessarily what the subject perceives. The
experimenter doesn’t know what the subject perceives, but only what the
experimenter perceives as the presentation. What the subject perceives is
the result of combining the presentation with an effect of the
action.

If what the experimenter presents to the subject is not necessarily what
the subject perceives, it is incumbent on the experimenter to demonstrate
in some way what the subject is actually perceiving, whether it is
different from what the experimenter perceives or the same. Simply the
chance that it could be different is enough to demand that a
determination be made. In PCT we have a way of doing that; I don’t know
of any other approach that does. As long as there is any real possibility
that what the subject perceives is not what the experimenter thinks is
being perceived, there is only one acceptable course of action: find out
the truth. Otherwise the experiment will mean nothing.

BP earlier:

Here’s what the subject sees on
the screen:

==== ====

<—>

The equal signs are the (split) target that moves up and down, and the
line with arrows at both ends is the cursor that also moves up and down.
We can easily establish that the controlled variable in this experiment
is some aspect of this display. That aspect, presumably, is the input
variable of interest. We determine its reference condition by seeing what
condition would generate no behavior to correct it.

MT: In PCT, yes. I suppose it would be possible to figure out what the
other schools might say about it, but there wouldn’t be much point, would
there?

BP now:

The point is the same as the one described above. You can look at that
display on the screen, but you don’t know what the subject sees there.
That’s the whole point of the test for the controlled variable: without
it, YOU DON’T KNOW. You’re only guessing.

BP: Earlier:

I think you will agree that the
controlled variable is the vertical distance between the target and the
cursor. The subject can set a reference condition of zero distance, or
any positive or negative distance, and defend it against disturbances.
The above diagram might show the reference condition of the display. So
in your chart, the sensory input is this vertical distance, rather than
the cursor alone or the target alone.

MT:

No. In my diagram, the sensory
input is a pattern of light and dark on the retina. The vertical distance
is part of what I called Me, the computed stuff. But no matter, let’s
proceed.

BP:
I trust you don’t mean that the pattern of light and dark is S1, the
input from the light. If you meant that, the “information
processing” you said you are measuring would be occurring in the
path between the light bulb and the retina.
You make my point very nicely. If the experimenter could watch the images
of that target-cursor pattern on the retina as the subject does the
pursuit tracking task, he might be very surprised. On the retina, he
would not see a moving target being pursued by a moving cursor. He would
see a roughly stationary target with the cursor moving above and below
it, but not nearly as much as the real cursor moves on the screen. This
is because in the pursuit task, the eyes follow the target up and down
fairly closely, keeping its image relatively stationary on the retina –
certainly much more stationary than what the experimenter sees on the
computer screen.
So what the experimenter sees as the perceptual situation is not even
what is presented to the retina of the subject’s eyes. And if what is
being controlled is a derived variable like the distance between two
objects, even the retinal pattern of light and dark intensities is not
the relevant perception. The relevant perception is the
relationship of distance, which is not explicitly represented, according
to PCT, below the fifth level of perceptual input functions in what you
(and I, of course) call the mind.

This is what I was suggesting you might realize when I told you which of
your statements was the first I disagreed with:

MT earlier:

In other words, all agree
on there being sensory INPUT and muscular and chemical
OUTPUT.

WP earlier:

If I point out that the real
problem is in this assumed agreement, will that jog you enough to see
what I’m going to say before I say it?

BP now:

What we do not and cannot agree on without proof of some sort is the
aspect of the “sensory input” that is relevant to the muscular
or chemical output. We might agree that there is some sort of sensory
input, but just what it is remains unknown to the naive external observer
– even at the level of primary sensory stimulation.

You said “Is not the signal path within the person FROM the senses
rather than TO the senses?” That was not what I hoped you might
say.

BP earlier:

The behavior of the subject is
to change the position of a mouse by pushing it away or pulling it
closer. This action is indicated by a number in the computer which is
updated 60 times per second, and also by the position of the cursor which
is updated at the same time. So we have the following definitions, in
simplified form:

input: Cursor position - Target position

mind: integral((reference - input))

output: mouse position or cursor position

I think you can see the problem immediately. Neither of the objects on
the screen, the target or the cursor, is the input to the aspect of the
mind we are discussing.

MT:

You may be discussing
“aspects of the mind”, but I was not.

BP:

Yes, you were. You spoke of the “distance” between cursor and
target in this post. You spoke of perception of a light, and of light
being “on” or “off.” Those are perceptions are
aspects of the mind’s way of perceiving, not phenomena in the
environment.

MT:

I believe you must have
read my message, in which I listed a very small number of gross signal
pathways admitted by the different schools. In it, you would not have
seen any reference to “aspects of mind” other than a generic
segregation that becomes more and more explicit as one moves from
behaviourist to Cognitive, to TOTE, to PCT.

BP:

I don’t know what you’re getting at here. Aren’t signal pathways part of
the mind? Isn’t perception of a relationship an aspect of mind? Or are
you referring only to the terminology to be found in other theories?

MT:

I know we agree that the other
three schools are demonstrably inadequate in dealing with even this
simple situation. I’m afraid I still don’t see where you are going, how
you are going to show that you don’t agree that input is through the
senses and output through the muscles.

That was your guess at an answer to my hint about the statement I
disagreed with. Input is certainly through the senses – I don’t think
you really thought I believed otherwise. But what is being perceived
can’t be deduced simply from tracking the physical inputs to the point
where they enter the sensory interface from outside. What is perceived
depends crucially on the perceptual input functions that receive the
incoming stimuli and convert them into perceptual signals that are
functions of the incoming stimuli. If you don’t know how to identify
those functions, you don’t know what is being reported to the brain, the
mind. With PCT we have developed a method for finding out which aspects
of the incoming stimuli are actually being controlled, which is evidence
about what is being perceived – model-dependent evidence, but that’s
better than no evidence at all. That’s why both Rick and I consider some
form of the Test to be an essential part of valid psychological
experimentation. Without it, all you know is how your manipulations of
the environment looked to you, and what aspect of the observable motor
actions seemed worthy of your attention.

BP earlier:

So the stimulus that the
behaviorist applies is not the actual input to the subject. The actual
input is a function of the stimulus and the
behavior.

MT:

Yes. Does that input not go in
through the senses? Is not the signal path within the person FROM the
senses rather than TO the senses?

It should be clear by now that this is a perseveration from the initial
wrong guess about what I meant. Get that out of your mind. The stimulus
input enters the sensory interface. No question about that. It is the
functional relationship of the stimuli to the resulting perceptual
signals that remains unknown until you do some adequate formal
investigation of them, which PCT alone provides. The subject experiences
the perceptual signals, not the stimuli.

MT:

Yes. All true. But I still
haven’t seen how you are going to get to the position you said up top
that you were going to, and I can already see the end of your
message.

BP:

You won’t find what you’re looking for because what you’re looking for
was not what I meant.

BP earlier:

As Rick pointed out, the biggest
difference between PCT and the other three approaches you describe is in
defining the input to the mind.

MT:

I think it was I who pointed it
out before Rick. Might it be worthwhile to quote from my message:
“In PCT thought, there is no “End”. Control is continuous,
so that the path S->M->R is completed by R->E->S at all times
that some element of S is actually being controlled. This, together with
the prominence of D, raises the importance of the R->…->S pathway
to equality with that of the pathways inside M.”

BP:

That shows that you misunderstood Rick, too. Rick was pointing out that
the correct expression should be written

f(R,S) → M → R → E → S

This expression is not in final form; the variable S appears twice. This
means that the observed relationship between S and R does not represent M
alone, but contains properties of both M and E. When control is very
good, in fact, the influence of M on the observed relationship between R
and S all but vanishes; the observed relationship can be predicted almost
perfectly from the properties of E.

If you wish you can put time indices on the variables, but since these
are continuous variables they always have a value, and the correct method
of analysis is not a sequential analysis, but the convolution theorem.
And we do not consider the path S->M->R to be separable from the
path R->E->S because these are really two simultaneous
equations. The input to M is not S, as the first equation suggests, but
f(R,S). Again, this could be written more exactly using the convolution
theorem, but the point is that the input to M is not S alone.

BP earlier:

I tried to allude to this but
the point slipped past without comment. I mentioned a possible variant on
the Schouten experiment in which the subject operates a continuous
rheostat rather than simply clicking a button. That would slow down the
process enough to make it clear that the action was affecting the light
intensity at the same time the light intensity was affecting the action.

Remember that in the Schouten
experiment, the subject has NO influence on the light.

Hark back to the opening of this post: I should have said the effect of
the action on the relevant perception, not on the light intensity. You’re
perfectly right; the action has no effect on the light intensity. But if
PCT is correct, the only way for that to generate an action is for the
light intensity to play the part of a disturbance, an arbitrary influence
on the real controlled variable. What, then, is the real controlled
variable? Or to cast that in more scientific terms, from what
experimental evidence, and by what reasoning, do you identify the
controlled variable? Isn’t it reasonable to ask for experimental evidence
and the method of reasoning?

You are proposing a quite
different experiment, one that would no doubt elucidate some questions,
but it would not get at what Schouten was getting at, nor would it give
the kinds of results that interested me about it – namely the linear
increase in information about the light that is manifest in the
proportion of correct responses as a function of delay
time.

What if the Schouten experiment did not involve the control system you
imagined and the one you initially drew, with the imagination connection
at the second level but no input to the second level indicating the state
of the answer as indicated by the state of the button? What if the actual
system at the second level was continually perceiving the
relationship between the perceived state of the light and the
perceived state of the button? I drew a correction to your diagram by
adding a blue line carrying that information to the second level.

If the latter control system were present, wouldn’t you see exactly the
same behavior? At least you would see it up to the point of the initial
button press. What you would not see, since it wouldn’t be counted as
data, would be the immediate reaction to perceiving that the wrong
relationship had resulted. The physical setup is arranged so such
an attempt to correct the error would not be observed or counted as
correct. We can’t tell from what data are still available all these years
later whether your model or that model with my addition of a signal path
is correct, or (because no tests were ever done) whether they are both
wrong.

Would this affect your interpretation of the data? Maybe not. Maybe it
would. It’s possible that this would remain a clever way of looking at
details of perception that occur between the stimulus and the response,
or between a disturbance and a corrective action, or between a
disturbance and an imagined action. It’s possible that those details
could be profitably interpreted in the language of information theory.
And it’s possible that what is actually going on inside the system
between onset of light intensity and the end of a button-press, which are
the only data we have, is totally different from what we’re imagining and
that there is an entirely different reason for the observed relationship
between the distribution of guesses and the definition of d’.

Here is another mistake I made in that post:

BP earlier:

Then we would see that the level
that decides to respond to a light has to sense the state of the light
rather than responding open-loop. And that in turn would show that the
underlying system we are studying is not an open-loop system at any
level.

This was a flat-out slip of the mind. I meant to say that the level that
decides to respond to a light has to sense the state of the BUTTON rather
than responding open-loop. The error conditions are (a) light and
not-button and (b) button and not-light. The no-error conditions are (c)
light and button, and (d) not-light and not-button. The reference state
boils down to

Light XOR not-button

The second level has to perceive whether the answer-control system is
perceiving the requested answer or not.

Of course a behaviorist can argue that the second level simply responds
to the light by emitting an output that causes the button to be pressed,
and without any tests there is no way to prove that proposal wrong. As
far as I can see, that model would also be consistent with the data and
your analysis.

BP earlier:

In the Schouten
experiment, we are seeing a control system operating under conditions
which artificially limit its behavior to what seems a stimulus-response
form – but there is no reason to think that is what is really
there.

MT:

No, indeed. And in none of my
diagrams dealing with different aspects of the experiment has there been
any such suggestion. You are knocking down straw men with great facility.
I claim only that in the Schouten experiment and in many detection and
discrimination experiments, the subject is unable to influence the
presentation (the disturbing variable) in any way.

Is it knocking down a straw man to point out that the Schouten experiment
does not distinguish between PCT and SR theory?

Your initial diagram also made the claim that the state of the button
perceived at the second level was entirely imagined, so there was no
actual feedback connection about the state of the button at the second
level. You still have made no comments about this.

MT earlier:

You have repeatedly asserted
that the Schouten experiment shows us an open-loop response at the input.

MT earlier:

would make a slight edit to that
sentence, replacing “repeatedly” with “never”. That
would be rather more accurate.

OK, I used the word “response” ambiguously and you took it to
mean the output behavior. I meant “perceptual response to the
light.” And that you did say was open-loop. However, by omitting
feedback about the state of the button to the second level, you also made
sure that the second level was indistinguishable from a purely open-loop
system.

The second diagram, the one you are now drawing, seems to incorporate the
change I recommended, though you haven’t drawn attention to this
fact.

MT:

I think the rest of my message,
on which you did not comment, makes this even more explicit. Rather than
repeating why that is so, I will simply present again the diagram that
accompanied it.

BP:

Now there is a line to the higher input coming from the perceptual signal
in the “answer-control” system which receives the
“answer” output from the higher system, and the higher system
is no longer open-loop. Its perception now depends on the light input
(one of the red input lines to the upper input function) and the state of
the button perceived as the requested answer (the black line entering the
upper input function). The latter is the line I recommended adding to
your original diagram.

Having now read your entire
message, I’m leaving my interpolated comments as they were when I made
them during a read-through of your message. Having finished the message I
cannot make any sense of your comment up front:

[MT] In other words, all agree on there being sensory INPUT
and muscular and chemical OUTPUT.

[BP]If I point out that the real problem is in this assumed
agreement, will that jog you enough to see what I’m going to say before I
say it?

In this diagram, is the signal flow from A to C by A>B>C or

C ?

BP:

It is from A to B to C. If B is an input function, however, you cannot
know what information the arrow from B to C is carrying unless you know
the form of the input function, as well as the state of A. Knowing A is
not enough. It does not tell you what the input to C, coming from B, is.
And that is what I have meant from the start, whether I communicated that
adequately or poorly.

Best,

Bill P.

P.S. In the Schouten experiment, I wonder what would happen if the
subject were told, “We’re going to pay you $50 for participating in
this experiment, but we’re going to deduct 10 cents for each mistake you
make.”

Bill_Powers1 · April 14, 2009, 7:26pm

[From Bill Powers (2009.04.16.1317 MDT)]

Martin Taylor 2009.04.14.00.04 --

I'm making all sorts of mistakes today.

I said the right expression was

f(R,S) -> M -> R -> E -> S

But this was done too hastily. We consider the Stimulus to be a disturbance if the response can't affect it, and the role of the environment is already contained in the function f, so I should have written

f(R,S) -> M -> R

It's R, not S that is present twice in the expression.

Best,

Bill P.

rsmarken · April 14, 2009, 7:59pm

[From Rick Marken (2009.04.14.1300)]

Bill Powers (2009.04.16.1317 MDT) to Martin Taylor (2009.04.14.00.04)–

f(R,S) → M → R

It’s R, not S that is present twice in the expression.

And, of course, f(R,S) is the what I mean by CV, the controlled perceptual variable, f() being the perceptual function.

Best regards

Rick

···

–
Richard S. Marken PhD
rsmarken@gmail.com

Bill_Powers1 · April 14, 2009, 8:27pm

[From Bill Powers (2009.04.14.1426 MDT)]

Rick Marken (2009.04.14.1300) --

And, of course, f(R,S) is the what I mean by CV, the controlled perceptual variable, f() being the perceptual function.

Yes. The CV is a bit hard to pin down, since we usually think of it as being in the environment. But you're right, the CV is defined by the perceptual input function, and unless we have a similar one we won't recognize it when we see it.

Best,

Bill P.

MartinT · April 17, 2009, 2:41pm

[Martin Taylor 2009.04.17.10.15]

[From Bill Powers (2009.04.16.1317 MDT)]

Martin Taylor 2009.04.14.00.04 –

I’m making all sorts of mistakes today.

I said the right expression was

f(R,S) → M → R → E → S

But this was done too hastily. We consider the Stimulus to be a
disturbance if the response can’t affect it, and the role of the
environment is already contained in the function f, so I should have
written

f(R,S) → M → R

It’s R, not S that is present twice in the expression.

“It’s R, not S that is present twice in the expression” is
really the central point of why I originally introduced the Schouten
experiment, isn’t it?
Other than that, the only issues you and Rick have wanted to discuss
for three months now are about the mechanisms that exist in the
pathways that result in the observable response (button push, saying
“Left”, waving hand, …). Those issues are perfectly reasonable
topics of discussion, but they should be in separate threads.

···

In another message [From Bill Powers (2009.04.16.0854 MDT)], you say: “Is
it knocking down a straw man to point out that the Schouten experiment
does not distinguish between PCT and SR theory?” Yes, it most
definitely is.
Is it necessary for every experiment to distinguish between PCT and SR?
Have there not been enough studies that do so distinguish?
The straw man is that you make it sound as though an inability of the
experiment to distinguish between PCT and SR is relevant to “It’s R,
not S that is present twice in the expression”, or, to be more
direct, as though it is relevant to the fact that whatever perception
might be being controlled, it has as at least one input a disturbance
manipulated by the experimenter and not by the subject, and the output
actions observable by the experimenter are functionally related to that
disturbance. The fact that a functional relationship exists is
irrelevant to which theory of mind is more probably correct. The facts
of other experiments have made that discrimination crystal clear.

Martin

Bill_Powers1 · April 17, 2009, 4:55pm

[From Bill Powers (2009.04.17.0904 MDT)]

Martin Taylor 2009.04.17.10.15 –

“It’s R, not S that is
present twice in the expression” is really the central point of
why I originally introduced the Schouten experiment, isn’t
it?

I don’t know. Is it? I thought the central point was the relationship
between the time delay and the proportion of correct responses. You’re
going to have to explain how the fact that R shows up twice relates to
that. I don’t have one of those minds that quickly sees all the
implications in cryptic allusions.

Other than that, the only issues
you and Rick have wanted to discuss for three months now are about the
mechanisms that exist in the pathways that result in the observable
response (button push, saying “Left”, waving hand, …).
Those issues are perfectly reasonable topics of discussion, but they
should be in separate threads.

I think we operate on different principles here. For a hundred years,
large numbers of psychologists and other life scientists have simply
taken it for granted that behavior consists of responses to stimuli, and
that there is no need to prove that this is true in every experiment.
Most of them wouldn’t even know how to find out if that is true, because
they don’t know of any alternative. But I have taken the position that is
IS necessary to prove it in every experiment the interpretation of which
depends, to any extent, on the kind of model assumed. It may turn out
that the model doesn’t make any difference. If so, a little extra effort
has been wasted. You also don’t have much of a model. But there are many
cases, I think, in which the model makes a great deal of difference, so
much that if the model is wrong, the conclusions are wrong. And that is
worth wasting a little extra effort to discover.

Perhaps I err in the other direction. I think every experiment should
test a model. To me, this is simply the best way to see if a theory
predicts correctly. A model behaves according to the rules you have built
into it, so when you use the model you commit yourself to those rules.
Your hopes, beliefs, interests, and preferences can no longer influence
the outcome; you can wave your arms and make all the excuses you want to,
but the model is telling you the actual implications of your assumptions.
Before you can say anything at all about the meaning or implication of
any aspect of the experiment, you have to establish first that your model
correctly predicts the behavior that’s actually observed.

When you presented those drawings of the model you imagined for the
Schouten experiment, the first thing, and really the only thing, I wanted
to know was whether that model could behave as subjects behaved in that
experiment. I didn’t care what assumptions you used, as long as the model
conformed to them. I didn’t care about the proportion of correct
responses or whether the measure of discriminability had any particular
relationship to the time delay. I just wanted to see if this model would
actually work.

I realized quickly that as drawn, the model would not work. A permanent
imagination connection at the upper level meant that that level could not
function as a control system relative to the real environment. It was set
up so that environmental influences (other than the light) could not
disturb the relationship it was controlling. That, in turn, meant that
what you drew could be modeled just as well by a simple S-R connection
with the right characteristics. At that point I lost interest because in
my way of thinking, that makes everything you say that relates to parts
of the model irrelevant.

Your rejoinder at one point of my objections was, in effect, “But
think of all those other control systems working in the background, and
all the things we know about control systems in general that have already
been established.” That’s a violation of my basic principle, which
demands that we demonstrate rather than assume whenever the assumption
would make a critical difference in the conclusions. Anything about those
background control systems and prior results from experiments that would
have an effect on this experiment should be part of the model, and there
should be ways to observe their effects. Otherwise they’re nothing but
fudge factors that can be used to make the conclusions come out any way
you want.

I realize that this is an ideal and we can’t always live up to it
completely. But in my opinion we should do our best. If there’s any
assumption that is really critical, we should include at least a
first-approximation test of it in the experiment, so we can compare a
model that includes the assumption against an experiment in which the
assumption might turn out to be false in this one case.

I understood, with a little help, that the “open-loop”
relationship you were talking about was actually in a part of the model
where no loop ever existed – in the perceptual input path. As you said,
all components in a control system are open-loop as we usually represent
them, so why not this one? That was OK with me once I grasped exactly
what part of the model you were talking about. But then I wanted to see
if this model made it possible to identify the particular intervening
variables you were proposing. And it didn’t.

Here’s what we observe:

Light onsets → Organism → % correct judgments at delay =
tau.

You proposed that between the light and some stage of the model
inside the organism, there was some kind of information processing going
on, and that the third beep specified exactly when the output indicating
the judgment should occur. You proposed that part of this process was a
set of control systems that could press a button when told to perceive
that event. But the point inside the organism when that judgment was made
is inaccessible; there is nothing observable to indicate it, or to
indicate that there are actually any internal control systems at work.
You said that we know the path from the point of judgment to the output,
so we could deduce the input part of the path, but we can’te observe
anything indicating the output part of the path either. All we observe is
the overall relationship between light intensity and button press. In the
absence of a verifiable model that shows where we could observe something
indicating the intermediate process, we are free to imagine any internal
processes we like. And that very freedom tells us that the experiment, or
the model, is inadequate.

If we’re free to assume any value or even the existence or nonexistence
of a variable in a model without changing the result, that is sufficient
proof that the result does not depend on that value or variable, and the
variable doesn’t belong in the model. Even the old-time behaviorists
understood this when they objected to the use of imaginary intervening
variables in psychological explanations. They got that one right.

In another message [From Bill
Powers (2009.04.16.0854 MDT)], you say: “Is it knocking down a
straw man to point out that the Schouten experiment does not distinguish
between PCT and SR theory?” Yes, it most definitely is.

Is it necessary for every experiment to distinguish between PCT and SR?
Have there not been enough studies that do so distinguish?

Yes, it’s necessary and No there haven’t been enough and never will be,
and I guess this is a real difference between us. There will never be so
many studies that we can simply take PCT for granted from then on. If a
control experiment can’t show that we are dealing, in this one case in a
hundred, with an S-R system, it’s a bad experiment and if the conclusions
depend on its being a PCT system, we can’t accept them. Loyalty to PCT is
a handicap. We can’t adhere to this principle with total fidelity in
every detail, but we can at least stick to it insofar as it affects the
main idea in any experiment, or its most important findings.

The straw man is that you make
it sound as though an inability of the experiment to distinguish between
PCT and SR is relevant to “It’s R, not S that is present twice in
the expression”, or, to be more direct, as though it is relevant
to the fact that whatever perception might be being controlled, it has as
at least one input a disturbance manipulated by the experimenter and not
by the subject, and the output actions observable by the experimenter are
functionally related to that disturbance. The fact that a functional
relationship exists is irrelevant to which theory of mind is more
probably correct. The facts of other experiments have made that
discrimination crystal clear.

Forget the other experiments. It’s THIS experiment I’m concerned with.
What you don’t appear to see (though I hestitate to believe this) is that
the appearance of R in the expression for the input means that S IS
NOT THE ACTUAL INPUT TO THE BEHAVING SYSTEM WE ARE STUDYING. That,
surely, is not a straw man.
If there is a control system sensing both the light and the
“answer” perceptual signal coming from a lower control system,
and if it is controlling the relationship between the light and the
answer (the actual answer given, not an imagined one), then that
relationship is the input to the control system of interest. The state of
that relationship can be affected – but not instantly – by the action
of pressing the button, even though the other term in it, the state of
the light, can’t be affected by that action.
That is how a model of this system might be organized and how it would
behave. Clearly, control is limited by the speed of the control system;
disturbances that appear instantly will cause errors that remain
uncorrected for some short time, but then they will be corrected by the
action. A disturbance that appears and then disappears before the control
system can act will result in an action that causes an error rather than
corrects one: pressing a button when the light is not on is an error,
just as not pressing it when the light is on is an error.
By wasting a little effort we could see if such a control system is
actually present. Instead of just doing the same presentation over and
over and measuring the same data again and again, we could add the
recording of all button presses even when they are wrong and even those
that take place after the initial press but before the next trial starts.
Our model tells us that it will try to correct incorrect presses of
either kind (wrong press or failure to press). It tells us that it will
respond to a brief flash of the light with a delayed response that seems
to echo the flash. We could adjust the model so the responses and delays
match those of the real subject. Whatever we say about events inside the
organism would then have at least a little evidential support. We would
know at least one way in which the system could be organized to behave in
the way we observe. That is a big step upward from pure
imagination.
If you read carefully what I say about any of my experiments and
demonstrations, you will see how often I will qualify a statement by
adding, “According to PCT” or “If PCT is correct.”
Those additions are not pro forma boilerplate. I am reminding myself as
well as the reader that there has to be a way to tell, in every
experiment, if the PCT model is NOT the right one. If that ability is not
there, we don’t have an experiment – it’s just an exercise. Every
experiment should test a model, and that means it has to be possible that
the test could be failed. Otherwise it’s not a test. I don’t mean that we
can test everything every time, but if we don’t test anything
we’re not doing science.

Best,

Bill P.

rsmarken · April 20, 2009, 9:56pm

[From Rick Marken (2009.04.20.1500)]

Bill Powers (2009.04.17.0904 MDT)--

>Martin Taylor 2009.04.17.10.15 --

"It's R, not S that is present twice in the expression" is really the
central point of why I originally introduced the Schouten experiment, isn't
it?

I don't know. Is it?

I would like to hear Martin's answer to this question.

Other than that, the only issues you and Rick have wanted to discuss for
three months now are about the mechanisms that exist in the pathways that
result in the observable response (button push, saying "Left", waving hand,

>....).ï¿½ Those issues are perfectly reasonable topics of discussion, but they
>should be in separate threads.

I think we operate on different principles here. For a hundred years, large
numbers of psychologists and other life scientists have simply taken it for
granted that behavior consists of responses to stimuli, and that there is no
need to prove that this is true in every experiment. Most of them wouldn't
even know how to find out if that is true, because they don't know of any
alternative. But I have taken the position that is IS necessary to prove it
in every experiment the interpretation of which depends, to any extent, on
the kind of model assumed. It may turn out that the model doesn't make any
difference. If so, a little extra effort has been wasted. You also don't
have much of a model. But there are many cases, I think, in which the model
makes a great deal of difference, so much that if the model is wrong, the
conclusions are wrong. And that is worth wasting a little extra effort to
discover.

I agree, of course. What do you think, Martin? You (Martin) keep
saying that you liked my "Revolution" paper and then you go off and
dismiss everything I say in the paper (like my repeatedly pointing out
the importance -- for the reasons Bill gives above -- of always
testing for controlled variables). If you disagree with the arguments
in my paper then why not just disagree and explain your
disagreements? Why play games?

Is it necessary for every experiment to distinguish between PCT and SR? Have
there not been enough studies that do so distinguish?

Yes, it's necessary and No there haven't been enough and never will be, and
I guess this is a real difference between us.

I'd be interested in what you haev to say about this, Martin.

The straw man is that you make it sound as though an inability of the
experiment to distinguish between PCT and SR is relevant to "It's R, not S
that is present twice in the expression", or, to be more direct, as though
it is relevant to the fact that whatever perception might be being
controlled, it has as at least one input a disturbance manipulated by the
experimenter and not by the subject, and the output actions observable by
the experimenter are functionally related to that disturbance.

Well said, Bill. What say you, Martin?

What you don't appear to see (though I hesitate to believe this) is that the
appearance of R in the expression for the input means that S IS NOT THE
ACTUAL INPUT TO THE BEHAVING SYSTEM WE ARE STUDYING. That,
surely, is not a straw man.

What do you think about this, Martin?

If there is a control system sensing both the light and the "answer"
perceptual signal coming from a lower control system, and if it is
controlling the relationship between the light and the answer (the actual
answer given, not an imagined one), then that relationship is the input to
the control system of interest. The state of that relationship can be
affected -- but not instantly -- by the action of pressing the button, even
though the other term in it, the state of the light, can't be affected by
that action.

And what do you think about this, Martin?

By wasting a little effort we could see if such a control system is actually
present.

That wasted effort being, of course, testing for controlled variables.

Martin, do you still think that the test for controlled variables is
not a necessary part of psychological research? Or have you finally
been convinced against your will (and, thus, are of the same opinion
still;-)?

Best

Rick

···

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

MartinT · April 21, 2009, 4:46am

[Martin Taylor 2009.04.20.23.38]

I answer this in conflict, as I know it will be too short to reolve the
issues that ought already to have been resolved weeks ago. All the
questions have been answered before, and really, really, should not
need to be answered again. But just in case there is a possibility that
something might get through, I’ll make a few quick comments.

[From Rick Marken (2009.04.20.1500)]

Bill Powers (2009.04.17.0904 MDT)--

Martin Taylor 2009.04.17.10.15 --

"It's R, not S that is present twice in the expression" is really the
central point of why I originally introduced the Schouten experiment, isn't
it?

I don't know. Is it?

I would like to hear Martin's answer to this question.

Yes it is.
Quote from [Martin Taylor 2009.02.12.14.25]
P3. IV as a disturbance. It’s true that the IV is a disturbance to
some controlled variable that the user influences by the output, but it
is not necessarily true that the CV is the same as the one that leads
to the effect that the open-loop analyst considers. In many cases,
there is no environmental feedback connection possible between the DV
and the IV, a lack of connection explicitly arranged by the
experimenter. The DV affects a different perception, perhaps that the
experimenter is satisfied with the subject’s performance. The relation
between the IV and the DV that is reported in the experiment is quite
possibly exactly what the open-loop experimenter thinks it is, given
the reference levels set by the higher-level purpose of perceiving the
experimenter to be satisfied. (This ties into the “establishing
operations” considerations on p6).
At a lower level of control [Martin Taylor 2009.02.12.17.26]: I
think you are ignoring the fact that not all perceptions, even
conscious ones, are controlled at all times. Many, and I would say
most, of our perceptions are not being controlled at any one moment. I
can perceive and report that my neighbour has entered their house,
without there being any feedback connection that affects the location
of the neighbour. In an experiment, for example, the IV could be the
experimenter inserting a signal in interval 1 and the subject pushing
the button marked 1, which has no effect on either that signal event or
the next one.
Or [Martin Taylor 2009.02.17.14.28]: I hope you now realize that
“open-loop” in this
context means that the subject’s response about the nature of the
presentation has no influence on the actual nature of the presentation,
NOT that there is a stimulus-response, cause-effect relation between
the presentation and the subject’s report. I don’t know how to say that
without using the words open-loop for the relation between the IV
(presentation) and DV (subject’s report).

Analytically:

Bill said "We consider the Stimulus to be a disturbance if the response
can’t affect it, and the role of the environment is already contained
in the function f, so I should have written
f(R,S) → M → R.

If f(R,S) leads to M, which leads to R (in the sense of signal flow),
which leads back to R, S is unaffected by anything that happens in the
loop. If f(R,S) is a controlled perception, we have p = f(R,S) ~= k (k
is a constant), which leads to R = g(S, k). Since we have S = s(D) (D
is the disturbance), R=g(s(D),k), meaning that R is a function of the
disturbance. The question at issue is not the form of this function,
but its channel capacity, and where the measured capacity is
constrained. Since the loop R-> environment → f(R,S) →
M-> R is unaffected by the timing of D, changes in the relation of R
to D can be attributed to the function s().

Which is the “central point of why I originally introduced the Schouten
experiment, isn’t it”?

Other than that, the only issues you and Rick have wanted to discuss for
three months now are about the mechanisms that exist in the pathways that
result in the observable response (button push, saying "Left", waving hand,
....).  Those issues are perfectly reasonable topics of discussion, but they
should be in separate threads.

I think we operate on different principles here. For a hundred years, large
numbers of psychologists and other life scientists have simply taken it for
granted that behavior consists of responses to stimuli, and that there is no
need to prove that this is true in every experiment. Most of them wouldn't
even know how to find out if that is true, because they don't know of any
alternative. But I have taken the position that is IS necessary to prove it
in every experiment the interpretation of which depends, to any extent, on
the kind of model assumed. It may turn out that the model doesn't make any
difference. If so, a little extra effort has been wasted. You also don't
have much of a model. But there are many cases, I think, in which the model
makes a great deal of difference, so much that if the model is wrong, the
conclusions are wrong. And that is worth wasting a little extra effort to
discover.

I agree, of course. What do you think, Martin? You (Martin) keep
saying that you liked my "Revolution" paper and then you go off and
dismiss everything I say in the paper (like my repeatedly pointing out
the importance -- for the reasons Bill gives above -- of always
testing for controlled variables). If you disagree with the arguments
in my paper then why not just disagree and explain your
disagreements? Why play games?

I don’t. I liked an agreed with most of what you say in the Revolution
paper, and I was particularly pleased that you had specified one set of
conditions under which it was legitimate to examine the properties of
an open-loop segment of a control pathway. The fact that you have since
disavowed that passage of the paper is irrelevant to why I thought it
had been a very judicious point to make. With it, the paper made sense.
Without it, some of the arguments would have been open to question.

Is it necessary for every experiment to distinguish between PCT and SR? Have
there not been enough studies that do so distinguish?

Yes, it's necessary and No there haven't been enough and never will be, and
I guess this is a real difference between us.

I'd be interested in what you haev to say about this, Martin.

If I were looking to investigate the efficiency of an engine, I would
not conduct a study to show that THIS engine was not run by inflow and
outflow of phlogiston. That question was settled long ago.
Scientifically, the question of whether humans control their
perceptions is, I believe, well settled, and I really do disagree that
every experiment that looks at human (or other biological) activity
must have as its aim to buttress what is to me a well-established fact
of nature. If you want to do experiments for public-relations purposes,
to convince people that still think phlogiston flow runs engines, fine,
so be it. But don’t ask me to do that when I’m interested in the
thermal and physical properties of the elements of the engine. Bill may
still be fighting the old wars, and he has reason to worry that the
wider world of psychologists don’t found their work on PCT, but to
insist that every experiment must be done for propaganda purposes is to
obstruct the progress of the science.

The straw man is that you make it sound as though an inability of the
experiment to distinguish between PCT and SR is relevant to "It's R, not S
that is present twice in the expression", or, to be more direct, as though
it is relevant to the fact that whatever perception might be being
controlled, it has as at least one input a disturbance manipulated by the
experimenter and not by the subject, and the output actions observable by
the experimenter are functionally related to that disturbance.

Well said, Bill. What say you, Martin?

The straw man is that Bill makes it seem as though if an experiment can
give useful results in both SR and PCT analyses, it must be discarded
as invalid in both.

What you don't appear to see (though I hesitate to believe this) is that the
appearance of R in the expression for the input means that S IS NOT THE
ACTUAL INPUT TO THE BEHAVING SYSTEM WE ARE STUDYING. That,
surely, is not a straw man.

What do you think about this, Martin?

I do not agree that I don’t see what I have often expressed verbally,
in many diagrams and in equations. The straw man in this case is the
fantasy of my ignorance that f(x,y) has an argument other than x.

If there is a control system sensing both the light and the "answer"
perceptual signal coming from a lower control system, and if it is
controlling the relationship between the light and the answer (the actual
answer given, not an imagined one), then that relationship is the input to
the control system of interest. The state of that relationship can be
affected -- but not instantly -- by the action of pressing the button, even
though the other term in it, the state of the light, can't be affected by
that action.

And what do you think about this, Martin?

I think that when Bill says the relationship is the input to the
control system of interest, he is calling a controlled perception an
input. Usually, we use the term “input” to refer to the inputs to a
perceptual function, of which the perception is an output. The
perception is, of course, an input to a comparator, but that isn’t what
Bill says. He says “input to the control system of interest”. In most
diagrams of control systems, there are two inputs, a reference input
and a disturbance input. In that sense, only the category perception of
which light is lit is an input to the control system of interest. In a
looser sense, all inputs to the perceptual function could be called
inputs, placing only the perceptual function, the compaarator, and the
output function in “the control system”. But if we did that, “the
control system” would actually be a simple I-O circuit with several
inputs. I don’t think that’s such a good idea.
In sum, what I think about it is that Bill’s comment is not very well
formed.
However, I do accept that after the button is pushed, the subject is
quite liable to perceive that he gave the wrong answer. I don’t accept
that this can happen by the time the button must be pushed, since, as
Bill pointed out: [From Bill Powers (2009.02.20.1946 MST)]: “Yes,
Martin emphasized that the action does not affect the presentation.
That loop is not closed, right? It could be closed if the action did
affect the presentation, which isn’t impossible, but it doesn’t, so
that potential loop is not closed – which means it’s open, as we use
the term. There IS an open-loop relationship between S and R, isn’t
there?” Yes, there is.

By wasting a little effort we could see if such a control system is actually
present.


That wasted effort being, of course, testing for controlled variables.
Martin, do you still think that the test for controlled variables is
not a necessary part of psychological research?

Yes. It’s not a necessary part of every experiment, at least not
explicitly. However, almost all experiments I know of in perception
psychology do implicitly test casually for the controlled variable, in
that there are checks to see that the subject is doing the task as
requested. It’s pretty hard to fake trying to get a certain percentage
correct when you actually can see less, and it’s pretty hard to fake it
even when you could actually get it correct on every trial. The
statistics don’t pan out when people try that – or I’ve never seen
anyone who claimed to be able to do it (and I know I can’t).

Or have you finally
been convinced against your will (and, thus, are of the same opinion
still;-)?

By this long interchange, I have become convinced that it is very hard
to get people to believe something simply by offering logical arguments
that demonstrate it. If someone is clever enough, they can always find
some way of avoiding the evidence by looking in another part of the
room, and if they aren’t clever enough, they can just say the argument
is incomprehensible (I believe “convoluted” is a commonly used
expression).

I hope I can remember to refrain from further postings on this or other
topics, if they will take more than 5 minutes to conceive and execute.
This was supposed to be such a 5-minute response, and it has taken an
hour I can ill afford.

Martin

PS. I’m not going over this to edit out typos and brain farts, so you
may well find errors and omissions, which I hope will be treated with
generosity appropriate to an ally, rather than as weaknesses in an
enemy’s armour.

rsmarken · April 21, 2009, 6:19am

[From Rick Marken (2009.04.20.2320)]

Martin Taylor (2009.04.20.23.38)--

I answer this in conflict, as I know it will be too short to reolve the
issues that ought already to have been resolved weeks ago.

They probably are as resolved as they are going to get. You are going
to continue to believe what you want and I am going to continue to
believe what I want. I believe that your beliefs are a barrier to
progress in PCT research. But there's nothing I can do about it.

Quote from [Martin Taylor 2009.02.12.14.25]ï¿½ P3. IV as a disturbance. It's
true that the IV is a disturbance to some controlled variable that the user
influences by the output, but it is not necessarily true that the CV is the
same as the one that leads to the effect that the open-loop analyst
considers.

If there is a relationship between IV and DV in a closed loop system
then the IV must be a disturbance to a CV that is also affected by the
DV. That's why there _is_ a relationship between an IV and DV in a
closed loop system. Saying that there is a relationship between IV and
DV because the IV is a disturbance to a CV but that the DV is caused
by a different CV is just arguing that the system is really open-loop
even though it's closed-loop; in other words, it's nonsense.

In many cases, there is no environmental feedback connection
possible between the DV and the IV, a lack of connection explicitly arranged
by the experimenter.

There is _never_ a feedback connection between the DV and IV. Saying
that there sometimes is suggests that you don't understand what a
disturbance is.

Or [Martin Taylor 2009.02.17.14.28]: I hope you now realize that "open-loop"
in this context means that the subject's response about the nature of the
presentation has no influence on the actual nature of the presentation

The subject's response in an experiment (DV) never has an influence on
the nature of the presentation (IV). This is true whether the system
under study is open or closed loop. You either don't understand or
don't want to understand the relationships between IV, DV and CV an
experiment on a closed loop system.

I liked an agreed with most of what you say in the Revolution
paper, and I was particularly pleased that you had specified one set of
conditions under which it was legitimate to examine the properties of an
open-loop segment of a control pathway.

No, I specified the conditions under which the relationship between an
IV and a DV in an experiment using causal methodology (ie. a
conventional experiment) provides a valid picture of the open-loop
transfer function (the function that transforms sensory input into
behavioral output) of the organism. The condition is that the system
under study be open-loop. The only way to tell whether the system is
open or closed loop is by testing for the controlled variable.
Conventional psychologists don't do this test because, like you, they
just think it's unnecessary; so they simply _assume_ that the system
can be studies as an open loop system and that, therefore, it is
appropriate to use causal methodology to study these systems.

The fact that you have since
disavowed that passage of the paper is irrelevant to why I thought it had
been a very judicious point to make.

What passage did I disavow? I avow everything in that paper, though
I'm sure I could have said some things better.

With it, the paper made sense. Without
it, some of the arguments would have been open to question.

Well, I sure hope you'll tell me what passage that is. And if you can
show me my disavowal that would be helpful too.

The straw man is that Bill makes it seem as though if an experiment can give
useful results in both SR and PCT analyses, it must be discarded as invalid
in both.

I never heard Bill make things seem that way. What Bill seemed to be
saying to me (particularly in that great _Psych Review_ paper of 1978)
was that the apparently open-loop S-R relationships you observe when
studying a closed loop system are a behavioral illusion. Actually,
Bill said something more like R =1/f (S): the relationship between
stimulus (S) and response (R) in a closed-loop system is the inverse
of the feedback connection (1/f) between R and CV.

You're clearly not going to give up on seeing the merits of
conventional research; obviously you are not convinced that the IV-DV
relationships you see in experiments like Schouten's are a behavioral
illusion if the system under study is a control system. But I would
still like to see what I disavowed but left in the "Revolution" paper
anyway.

Best

Rick

···

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

Bill_Powers1 · April 21, 2009, 12:17pm

[From Bill Powers (2009.04.21.0537 MDT)]

Martin Taylor 2009.04.20.23.38 –

I answer this in conflict, as I
know it will be too short to reolve the issues that ought already to have
been resolved weeks ago. All the questions have been answered before, and
really, really, should not need to be answered again. But just in case
there is a possibility that something might get through, I’ll make a few
quick comments.

You’re inmpatient with your readers, and the result is that you don’t
spend the time needed to achieve clarity.
Quote from [Martin Taylor 2009.02.12.14.25] *P3. IV as a
disturbance. It’s true that the IV is a disturbance to some controlled
variable that the user influences by the output,*The “user?” Are you speaking of the experimenter and his
output? Do you mean the subject of the experiment, and if so, are you
speaking of the lower level of output or the upper level in your
diagram?
*but it is not necessarily true that the CV is the same as the
one that leads to the effect that the open-loop analyst
considers.*How can that occur? Do you mean the subject is controlling a variable
other than the one that the open-loop analyst considers? Other than the
one that is affected both by the IV and the action? Is an open-loop
analyst an analyst who thinks he is dealing with an open-loop system when
it is really closed-loop? Or do you mean that the variable the analyst
considers is not the CV, but some other variable in the system? If so,
the analyst isn’t considering a CV at all, is he? Your sentence seems to
say, at first, that he is.
*In many cases, there is no environmental feedback connection
possible between the DV and the IV, a lack of connection explicitly
arranged by the experimenter. The DV affects a different perception,
perhaps that the experimenter is satisfied with the subject’s
performance.*Are you describing a condition in which the DV cannot affect a
variable that is also being affected by the IV, and is prior to the
actual CV? Does the DV affect the different perception in addition to its
effect on the CV affected by by the IV, or instead of that?
*The relation between the IV and the DV that is reported in the
experiment is quite possibly exactly what the open-loop experimenter
thinks it is, given the reference levels set by the higher-level purpose
of perceiving the experimenter to be satisfied. (This ties into the
“establishing operations” considerations on p6).*Now I can’t tell if you’re referring to a different experiment in
which the only effect that is perceived by the subject is an effect on
the experimenter’s satisfaction with his performance. Your thoughts are
whizzing around in your head and you know what you mean, but what you
mean isn’t getting into the text. I am telling you that this layer of the
protocol isn’t working. Are you saying that the only effect of the DV in
the Schouten experiment is on the perception of the experimenter’s
satisfaction?

Your communications, instead of slowing down and getting more detailed
when difficulties arise, are getting more terse and hasty, making them
harder and harder to understand. The labor of figuring out what you’re
trying to say is increasing instead of decreasing. I know you’re under
time constraints, which may well account for this problem, but it’s
really useless to try to explain what you mean under such
conditions.

I attach two figures which show my current model of what is going on in
the Schouten experiment, except for the systems that adjust the timing so
the contact closures coincide with the third beep. One shows the
organization of the control systems, and the other shows the imagined
events that take place, with their time relationships. The timing diagram
shows the reference signal being turned on 50 ms after the start of the
rise of the perceptual signal; that would produce 95% correct button
presses. I assume that the identification of the light is always correct,
but that the error signal that exists at the time the reference signal is
turned on can be erroneous because the perceptual signal has not yet
reached its final value and represents various proportions of signal and
noise. Other possibilities exist since the part of the system responsible
for the observed delay effects can’t be determined from the data. My
diagram shows only one possibility.

I have added to the system diagram, in red, labels showing how the parts
of the relationship-controlling system correspond to the generic PCT
diagram.

When your obligations permit, perhaps you could take the time to lay out
your own diagram of what you think might be happening, for comparison
with mine. And please take your time, writing with the clarity that seems
to come naturally when you’re teaching people about PCT – just pretend
that we know as little as they do and need just as much patient
explanation as they do.

Best,

Bill P.

P.S. Yes, we are all on the same side. I noted long ago that cooperation
is a difficult matter because nobody knows exactly what another’s
perceptions are or what reference levels exist, even when a lot of time
is spent trying to find out.

Bill_Powers1 · April 23, 2009, 3:58pm

[From Bill Powers (2009.04.23.0823 MDT)]

Rick Marken (2009.04.20.2320) –

RM to MT:

The subject’s response in an experiment (DV) never has an influence
on

the nature of the presentation (IV). This is true whether the system

under study is open or closed loop. You either don’t understand or

don’t want to understand the relationships between IV, DV and CV an

experiment on a closed loop system.

It’s risky trying to guess what someone else understands, and even
riskier to broadcast accusations about what they do not want to
understand. Apparently, that simple fact is something you don’t want to
understand.
In the Schouten experiment, the controlled variable of interest is not
simply a variable that happens to be disturbed by an external influence,
the state of the light. A disturbance of the kind you have in mind would
occur if, for example, someone in addition to the subject could push the
button. Obviously, turning the light on and off has no effect on the
state of the button, so it doesn’t disturb the same variable the subject
is controlling at the lowest level in Martin’s diagram.

At the highest level in the diagram, Martin says the state of the button
is perceived in relationship to an answer to an implied question (such as
“is the light on?”). The state of the relationship can be
(light and answer), (not-light and not-answer) , (light and not-answer)
or (not-light and answer). The last two would be error conditions. I call
this a relationship perception but it could also be a logical
perception.

The subject can affect only one term in the relationship, the state of
the answer. The other term, the state of the light, can affect the state
of the perceived relationship independently of the feedback effect from
the subject’s action, so at the higher level the state of the light is a
disturbance of the relationship perception, the same perception being
controlled by pressing the button. That’s the sort of disturbance you
seem to be thinking about. At the second level, it appears that the
answer is affected by the light directly (the SR view), but in actuality
this effect is due to disturbing the relationship of the button
and the light. If the right button is already pressed when the light goes
on (because the subject was fooled by a random excursion of the
perceptual signal), there would be no error and thus no change in the
action, but if the wrong button had been pressed, 50 milliseconds or so
later the error signal would appear and the button would be
released.

Just to show Martin that I can generalize when I want to, this
second-level case actually applies to systems at all levels, because all
controlled perceptions are functions of multiple variables at a lower
level (even the lowest-level perceptions are influenced by multiple
environmental variables). In the equation p = f(x1, x2, x3 … xn), any
of the x’s can be influenced by

(1) the controller’s action alone,

(2) both the controller’s action and an independent influence, or

(3) independent influences alone.

In the Schouten experiment, only the first and third cases are
represented, as far as we know. In the case you’re thinking of, only the
second case is present. So what you’re talking about is not quite the
same thing Martin is talking about.

I think you are closer to the truth than Martin is, however, because in
all three cases above, there are two influences on the controlled
perception, one due to the controller’s action and the other
independent. Following our usual custom of drawing a virtual controlled
variable in the environment that corresponds to the controlled
perception, we would show the relationship as a single environmental
variable affected by the controller’s action and a disturbance. This is
how we experience abstract variables anyway; we see things like the
distance between two points as if it is actually out there in the
environment between the two points, even though there is nothing in that
location. So functionally we could diagram pursuit tracking this way
(courier font for alignment):

( to P.I.F.) (from system)

^

v

  target ----> distance <

-----cursor

(disturbance)

CV (system’s action)

We could draw the Schouten situation this way:

( to P.I.F.) (from system)

^

v

  light ----> relationship < -----

answer(button)

(disturbance)

CV (system’s
action)

I think that’s the way you are seeing it – as if the relationship is
really out in the environment.

There is really no way to tell where in the control system the observed
Schouten phenomena are occurring. The diagram I posted, accompanied by a
timing diagram, showed how the actual perception of the light could
always be correct, yet the wrong button could be pressed in just the
proportions observed as the timing of the third beep was varied. We can’t
rule out the case Martin analyzed, nor can we rule it in: the
experimental data do not suffice for that. The observations remain
interesting, but I think there is a simple way to explain them and
determine the maximum channel capacity, if that’s what we want to know. A
frequency-response description of that capacity would be just as useful
as a description in terms of entropy and bits, and would entail fewer
imaginary components such as entropy and information, which are not
observable variables.

If you will look at my timing diagram you will see that the actual
channel capacity, in the sense of the number of correct determinations
per second that this system could make, depends on when the reference
signal at the second level is turned on after the correct determination
has been made. Without further data, we can’t determine the channel
capacity of the first stage of perception when the correct determination
is made – which would be the maximum possible for the whole
system.

Best,

Bill P.

···

I liked an agreed with most
of what you say in the Revolution

paper, and I was particularly pleased that you had specified one set
of

conditions under which it was legitimate to examine the properties
of an

open-loop segment of a control pathway.

No, I specified the conditions under which the relationship between
an

IV and a DV in an experiment using causal methodology (ie. a

conventional experiment) provides a valid picture of the open-loop

transfer function (the function that transforms sensory input into

behavioral output) of the organism. The condition is that the system

under study be open-loop. The only way to tell whether the system is

open or closed loop is by testing for the controlled variable.

Conventional psychologists don’t do this test because, like you,
they

just think it’s unnecessary; so they simply assume that the system

can be studies as an open loop system and that, therefore, it is

appropriate to use causal methodology to study these systems.

The fact that you have since

disavowed that passage of the paper is irrelevant to why I thought
it had

been a very judicious point to make.

What passage did I disavow? I avow everything in that paper, though

I’m sure I could have said some things better.

With it, the paper made sense. Without

it, some of the arguments would have been open to question.

Well, I sure hope you’ll tell me what passage that is. And if you
can

show me my disavowal that would be helpful too.

The straw man is that Bill makes it seem as though if an experiment
can give

useful results in both SR and PCT analyses, it must be discarded as
invalid

in both.

I never heard Bill make things seem that way. What Bill seemed to
be

saying to me (particularly in that great Psych Review paper of
1978)

was that the apparently open-loop S-R relationships you observe when

studying a closed loop system are a behavioral illusion. Actually,

Bill said something more like R =1/f (S): the relationship between

stimulus (S) and response (R) in a closed-loop system is the inverse

of the feedback connection (1/f) between R and CV.

You’re clearly not going to give up on seeing the merits of

conventional research; obviously you are not convinced that the
IV-DV

relationships you see in experiments like Schouten’s are a
behavioral

illusion if the system under study is a control system. But I would

still like to see what I disavowed but left in the “Revolution”
paper

anyway.

Best

Rick

–

Richard S. Marken PhD

rsmarken@gmail.com

No virus found in this incoming message.

Checked by AVG -
www.avg.com

Version: 8.0.238 / Virus Database: 270.12.1/2069 - Release Date: 04/20/09
10:36:00