Basics, disturbance causes error?

[From Rick Marken (930321.1200)]

Martin Taylor (930320 11:00) --

so p(t) = d(t) + o(t) cannot be a meaningful statement.

If you really think that p(t) = d(t) + o(t) is somehow not a
meaningful statement, then we really are poles apart; your statement
seems to me to be a denial of the physical situation in a
compensatory tracking task. If you take a look at the HyperCard
conflict stack that I sent you, you will find in the tracking loop
the following code

put trunc (A1*H1+B1*H2+distx) into xp

xp is a number representing the horizontal position of the point to
be plotted; it corresponds to p(t) -- with t being the index of
the iteration on which this is happening. A1 and B1 are constant multi-
pliers. H1 and H2 are the numbers corresponding to the x and y positions
of the mouse on the current iteration -- they are equivalent to o(t);
finally, distx is the value of the disturbance on this iteration;
of course, it corresponds to d(t). So, on each iteration of
the program, the horizontal position of the cursor on the screen
is proportional to the sum of the current disturbance value (distx)
and the current value of the mouse output (H1 and H2). You keep saying
that there might be a lag in the production of o(t) and I keep trying
to point out that this is not important. I hope thinking in terms of
the computer program will help you understand this. H1 and H2 are the
CURRENT values of the mouse; they might have been caused (inside the
subject) by processes that happened hours earlier -- but that doesn't
matter in terms of what the subject is SEEING NOW; what the subject
is seeing (p(t)) is the combined effects of the PRESENT distrubance
(distx) and the PRESENT output (H1 and H2). So can we please agree
that, at least in a compensatory tracking task (like the one in
the conflict stack) p(t) (or, at least, the CEV -- the position of
the cursor on the screen) is ALWAYS proportional to the SUM of
dirturbance and output -- ie. p(t) = o(t) + d(t)? Run the HyperCard
stack and see if you can convince yourself this is really what is
happening -- what you are seeing when you do the tracking task (unless
you don't move the mouse at all) is o(t) + d(t). Of course, when you
are not moving the mouse at all you are not controlling -- so
there is information about d(t) in uncontrolled variables, as Bill
P. has so politely pointed out.

This remarkably simple (but apparently unpleasant) little fact is
the reason why I keep saying that in a control system "there is no
information about the disturbance in p(t)". All you are ever perceiving
(in tracking task or "real life") is the combined result of what is
"happening" (d(t)) and what you are doing while it's happening (o(t)).

Martin says:

Now let's answer the "focus" questions:

1) Do you believe that the disturbance to a CEV (and the perecption
thereof) typically causes the error (discrepency between reference and
perception) that leads to the output that opposes that disturbance?

I think one has to define the disturbance in respect to the error exactly
as one defines the CEV in respect to the percept. The perceptual input
function defines the CEV (provided you include the operations of all
the PIFs in lower ECSs that provide the sensory signals to this ECS).
The perceptual signal is the current representation of the state of the
CEV. From an outsider's viewpoint, in which the perceptual signal and
the CEV can be independently measured, they may differ. The same applies
to the error and the disturbance. From inside the ECS, the perceptual
signal is all there is.

Did anyone see an answer to my question in there? I still don't know what
you believe, but the correct answer is very simple: No. Since
e = r-p and p = o + d, e = r - o + d so the error is determined by
the combination of output and disturbance (assuming constant r). The
disturbance DOES NOT cause the error in a control system.

2) Do you believe that it is necessary that the disturbance at least
occasionally be represented in the CEV?

The disturbance is not "represented" in the CEV. The disturbing variable
may affect the CEV, but representation occurs within the ECS.

So the answer is no? Good. But what's this "the representation occurs
within the ECS"? You don't mean that it is represented in p(t)?
Good. So we have two remaining signals in an ECS -- reference and
error. Where is the disturbance represented? The correct answer is?
In neither of them. It's obviously not in r(t) because d(t) does not show up
in it's calculation. It's not in e(t) either because e(t) = r - o(t) + d(t) so
again d(t) itself is not represented in the signal -- just the sum -- d(t)
+ o(t). So there is no representation of d(t) ANYWHERE in a control
system -- not in p(t), r(t) or e(t); it only exists in o(t) as the
integrated, amplified error -- even though d(t) is NOT represented in
the error itself.

3) If you answered "yes" to (2), why do you believe this?

I didn't say "yes."

Correct. So I still don't know what you believe.

So, for my attempts to analyze the control LOOP to be
characterized (repeatedly) as a support for S-R approaches or as an
attempt to maintain an S-R view strikes me as a bit ludicrous.

I'm not trying to make a political point; I think it's wonderful that
you don't support S-R approaches to behavior. It's just that you
keep describing the operation of a control system incorrectly, and
the mistakes you make (such as saying that there is information about
the disturbance in p(t) or, now that you have apparently backed off of
that claim (see above), that such information resides somewhere inside the
control system) are precisely those that are necessary to maintain
an input-output view of the operation of such a system. It makes it
possible to view the ultimate cause of output (o(t)) as something
outside the system (the information about d(t)).

But I
can accept that you might see my comments as lending aid and comfort
to the enemy.

I think we have different "enemies". You have problems with SR
theories of behavior. I have problems with those ideas too -- but,
more importantly, I have problems with the whole conventional approach
to studying behavior -- both its goals (finding relationships
between IVs and DVs) and it's assumptions (that such relationship
reveal something about how behavior works). I am arguing so much
with you because I think your view leaves open the door to conventioal
research methodology. After all, if there is information about d(t)
at some point along the path from p(t) to o(t) then it seems reasonable
to assume that one could learn something about how that information is
"processed" by looking at relationships between d(t) and o(t). But
this, according to PCT, is what the conventional IV-DV approach is
already looking at -- and finding at least statistical relationships
between these variables. But we (PCTers) already know that if the
system you are studying is a control system then the function relating
d(t) to o(t) is the inverse of the feedback function relating o(t) to
p(t) -- it has NOTHING TO DO with the processes inside the organism that
convert d(t) into o(t) (in fact, they are converting p(t) into o(t)).

I am beginning, at last, to have the feeling that we really do see in
the same way what happens in a control loop.

I hope so. There are some easy ways to test this. You could try again to
answer my question 1) above so I could see how you think the nitty gritty of
a control loop works -- is error caused by the disturbance in a control
loop? Or you could tell me (as an opponent of SR theory) whether the
typical IV-DV experiment reveals anything about the workings of the
control systems under study (assuming that they are control systems); for
example, what do we learn about the subject in an experiment in which we
manipulate the amount of information in the stimulus -- the IV -- and
measure the time to respond to each stimulus -- DV)? My answer would
be "nearly nothing".

I am adament about this because I know (from experience with the
Karolyans) that unless people start doing their research in a new
way (trying to test for controlled variables, for example) PCT will
just become a new language for talking about the SOC (same old crap).
PCT is not a new language; it is a completely new model of behavior
in which behavior (as dependent variable) is controlled perception,
not caused output. In order to test this model, a whole new approach
to research is needed -- one that is not only non-statistical (one
subject at a time) and highly precise (accept only findings with less
than 1% error) but based a proper understanding of the control model.