Purpose, Failure of PCT

[From Rick Marken (920324 8:20)]

First -- a note to Chuck Tucker. I read your "study guide" on
sociological views of crime that you sent to me by snail mail.
Thanks so much. It was really excellent. Not only
well written -- but I could even warm up to the conclusion.
I'm surprised to find that Durkheim was such a marvelously
innovative thinker. Really nice paper Chuck.

Jeff Dooley (920323.1100)

Welcome to CSGNet.

Your comments on "purpose" and "open loop" behavior were right
on target. Kuhn would have a great time with this one all right.

You also say:

I'd like to suggest the idea, blending concepts of control
theory purposiveness with a thesis of biologist Stuart
Kauffman's: that species (complex, adaptive systems) evolve
(drive themselves) to the edge of chaos and maintain
themselves there.

I think I need some clarification of the "edge of chaos" idea --
maybe you could make it a bit more concrete. It seems to have a
family resemblance to the idea that evolutionary reorganization
(which seems a bit chaotic) is the result of an increased rate
of random mutation (over generations, of course) resulting from
chronic "error" at the genetic level.

Chris Malcolm, in response to the following comment by me:

I just can't understand how any "behavior" that consistently
achieves some result (escape) could be completely open loop (unless it were
always performed in the same environment, from the same orientation, with the
same motor charateristics, etc.). The achievment of consistent results
in a variable environment is control, by definition.


This is a common misconception. It is possible for behaviour to be
adaptive in the sense of changing appropriately in response to the
environment without even involving any sensors, let alone controlled
variables. There are two ways it can be done. The first is to use
unstable motor behaviour where the environment selects the appropriate

That's one smart environment you got out there, Chris.

The second is where the appropriate behaviour concerns part
of the environment rather than the creature, e.g., putting something
into a hole. This can be done by using fixed predetermined behaviour on
the part of the creature to determine only part of the behaviour of the
item in question, and letting the local environment also affect it, with
the net result that fixed behaviour on the part of the creature results
in goal-seeking behaviour on the part of the item in question (the thing
which is being put in the hole, for example).

I don't know whether to laugh or cry. First I read this; then I go to
the mail box and find that the symposium for APS that I proposed was
rejected (Bill, Joel and Tom -- you're welcome to come to San Diego and
party at my condo down their anyway; you can leave your senses at home,
of course, because all we'll really be doing down there is behaving --
adaptively, I hope, but my experience has been that the San Diego
environment gives great adaptive behavior -- if you consider playing golf
and swimming to be adaptive).

Imagine my disappointment when I learned that my concept of behavior as
the control of perception is a common misconception; and on the same day
that I find out that the major national association of scientific
psychologists has no interest in a symposium on the implications of
purpose for the study of behavior. Boy, there's 12 years down the tubes.
I'm sorry all you folk's on CSGNet had to waste all this time and
bandwidth on arguments over a common misconception. But at least Chris
left the door open to the possibilty that SOME "adaptive" behavior
involves sensory input. So there are controlled variables -- they just
don't really need all this control theory stuff to understand them.

Chris goes on to say:

These two cases seem to many people to be so counter-intuitively absurd
that they cannot believe it possible, but as the history of science has
so often shown, failure of our human imaginations is not a reliable guide
to impossibility in our universe!

Well, it was my impression that many people feel these two cases to be
so obvious that they are the basis of all studies in the life sciences.
Guess I run into the wrong people.

In order to convince students of this I have built a couple of simple
robot systems which demonstrate these two cases. Look Ma, adaptive
behaviour with no sensors!

After a suitable pause for speculative ingenuity I will describe these

Please -- don't pause for too long. Seal my doom -- drive the stake into
the heart of PCT. Describe for us, in detail, these examples of adaptive
behavior (by which I presume you mean control since that was what I was
referring to in the original statement) that are open loop.

But just one teensy-weensy question before you do this -- so you can
be sure that your aim is true and you go straight to the heart: in
your "putting item into the hole" robot -- does it work no matter
what the size and shape of the item (even if it's wider than the hole)?
Does it work on a windy day? If the robot moves or is pushed (with continuously
varying force)? If you block the hole?

If "item in hole" is the reference state of a controlled variable
then that result should be achieved in the context of any disturbance
that could change that result -- to "item next to hole", for example. Does the
robot do that? Does it get the item into the hole in spite of reasonable
disturbances? If so, we control theorists have sure made a BIG mistake.
You will forgive me for thinking that it can't do it -- but if you demonstrate
to me that you have a robot that can control a variable without sensing
it, then I promise to be a good scientist and admit that I was as "mis-
conceptioned" as I could be -- and admit that SR is an important model
of adaptive behavior.

One last little point -- I am assuming that you mean the same thing by
"adaptive behavior" as I mean by control. If not -- if, for example,
all you mean is that you can design a system that generates interesting
results (like putting an item into a hole) then we are not talking about
the same thing at all (as you should be able to tell by my original
statement). I am not claiming the you need to control sensory input in
order to produce "interesting" or "useful" results -- you can produce
many useful results by accident. I am saying that systems cannot produce
CONSISTENT RESULTS -- whether they are interesting or useful or whatever
from YOUR point of view -- in normal environments -- ie -- ones which
also have variable effects on these results -- unless the systems can
SENSE these results. If this is a misconception -- common or not -- I
would apreciate being straightened out on it as soon as possble;
certainly before I waste anymore trees writing papers about it.


The loose canon



Richard S. Marken USMail: 10459 Holman Ave
The Aerospace Corporation Los Angeles, CA 90024
(310) 336-6214 (day)
(310) 474-0313 (evening)