where's the disturbance?

[From Rick Marken (930318.2200)]

Martin Taylor (930318 15:20) --

Let's look at the mechanics of what happens in a control loop, taking
Bill's example as a starting point.

In the following, B represents a CEV, in the specific example an object
connected by springs to A and C. The movements of A and C affect the
movement of B in opposing directions.

A control system has sensory input representing the position of B (and
nothing else).

Right!!

In the diagrams, at the moment the disturbance step occurs, it is fully
and completely represented in the CEV (and, by hypothesis, in the
perceptual signal, though that is a point we will wish to dispute in
later postings).

I am willing to agree that, at the moment the disturbance step occurs,
the disturbance IS COMPLETELY REPRESENTED IN THE CEV. I'm even willing
to assume that the control system is both sluggish and that it has a
substantial trasport lag -- so the output doesn't even start to
compensate for the disturbance for some time after the start of the
step. So for some time (as long as you like) the behavior of the
CEV is due solely and completely to the disturbance. Now the only
problem is HOW DOES THE CONTROL SYSTEM KNOW THAT THIS IS THE CASE?
That is, how does the control system know when the disturbance step
started? How does the control system know when the variations in the
CEV that it is looking at are due only to the disturbance and when they
are the combined result of its own effects and those of the disturbance?

In order for the control system to know when the disturbance has started
or stopped it would have to be able to perceive the disturbance. But you
have assumed (above) that "the control system has sensory information
about the position of B [the CEV] (and nothing else)". So the only person
who knows that the CEV (during your example) is the result of the
disturbance alone is the observer of the control system -- ie. you.

As the effect of the disturbance is used by the
ECS, through changes in its output, the representation of the disturbance
is, so to speak, "bled off" the perceptual signal, to find representation
in the output signal.

This is a nice try at saving the S-R point of view but it can't work; the
control system cannot possibly (for the reasons I mentioned above) know
when to do this "bleeding off". Exactly the same changes in the CEV could
be the result of the combined effect of the disturbance and the control
control system's own outputs or of those outputs only (with the disturbance
constant) -- in fact the response to disturbance that is reflected in the
CEV in your example is exactly the same as the transient response of the CEV
to an output resulting from the addition of an electrial impulse
"disturbace" to the error signal in the control system; if the control
system "bled off" the same output that it made when the source of
this variation in the CEV was caused by the disturbance, there
would be positive feedback -- but, as I have shown, the addition of
such a disturbance to the error signal is "compensated for" just as
effectively as if the disturbance had come from the outside world.

The point of all this is to try to demystify the "magical" fact that
there is no representation of the disturbance in the perceptual signal,
although there is an almost perfect representation of it in the output
signal.

It seems magical only when the control loop is looked at in
S-R terms (with the CEV being the S and the output being the R).
There was never any magic from a PCT perspective -- there is no
perceptible representation of the disturbance in the perceptual
signal; output is non-magically made to mirror the disturbance
by a system the generates outputs that are proportional to r-p,
the difference between current perception and a reference for
that perception.

The world can't tell you what to do -- and, frankly, it probably
doesn't much care.

Is this an agreeable reformulation of what Bill wrote a couple of days
ago explaining the events going on around a real loop with finite transport
lag?

Nope (see above). But I'll let Bill speak for himself.

Bottom line: the dynamics of the effects of disturbance and output
on the controlled variable don't bail out what is basically the
S-R view of control. There is NO information about the disturbance
in the CEV (or the perception thereof). Thus, the basic assumption of
scientific psychology for the last 100 plus years is out the window;
this is why the idea that there is no representation of the disturbance
in the perception is being so strongly resisted (and will continue to
be so -- despite the evidence); no one wants to give up comfortable
assumptions -- but facts is facts. I'm willing to give up my
comfortable assumption (that there is no recoverable representation
of the disturbance in the CEV) is someone would just show me how
to recover that representation -- and, hence, the disturbance (without
being given outside information about the output and/or the disturbance
itself).

Best

Rick

[From Oded Maler (930319.0915 ET]

ยทยทยท

*
* [From Rick Marken (930318.2200)]
*
* Martin Taylor (930318 15:20) --

Etc.

I have a lot to ketchup with the info stuff, but let me try to
formulate it the way I see it. Suppose there is one CEV x in one
dimension. It was at position x0 at t=0 (suppose for the moment that
perception is precise), and for a period t the control system pushed
it toward the right (positive). Then at time t, the variable is
observed to be in the left (x<x0). In this case you can be sure that
during that interval of time the net effect of the disturbance (its
integral or whatever) was to push it to the left. Of course you cannot
exactly know which of the infinitely many curves having the same
integral was the actual disturbance, but you can exclude many of
the other possibilities (you can be sure that the disturbance was not
in the same direction of you efforts). This, I think, is the sense
in which the perceptual signal *does* provide information about the
disturbance. All this can be done in a probabilistic framework, with
noise in the perception, action etc., and with some prior probabilities
on the form of the disturbance. Even if you cannot measure your own
actions, just know that they are bounded, still the velue of the
perceptual signal can tell you something about more likely or less
likely form of the disturbance.

Btw, I like the idea of calling all the real external world a
"disturbance", it's anthropocentrism in its best! ("Life is what
happens to you when you try to do other things" -J. Lennon).

--Oded

p.s.

I talked to a friend who is a mathemtical control theorist (= the
abstract version of a control engineer) and to his mind, every
control is feed-back control.

--

Oded Maler, LGI-IMAG (Campus), B.P. 53x, 38041 Grenoble, France
Phone: 76635846 Fax: 76446675 e-mail: maler@vercors.imag.fr