Beyond the tracking task

[From Rupert Young (970919.1130 BST)]

I think those of us who are in favor of PCT and wish to see it propagated are
rather like a group of peasants standing outside the castle of conventional
life sciences trying to knock it down by throwing stones at the castle walls.
Futile. Those inside the castle can afford to ignore the peasants as they are
secure and have the respect of the majority of peasants, the stones have no
effect. Perhaps, what us peasants should do is to take those stones and build
our own, bigger and better castle. Only then will those in the conventional
castle be tempted to leave it and join us.

I concur therefore with Rick and Bill's recent remarks.

(Rick Marken (970826.0850))

This (as you know) is my favorite theme. I have long felt that
the best way to promulgate PCT is to publish tons of papers
describing demonstrations of the _phenomena_ of control. I don't
think theoretical papers will help much anymore because the
assumption made by readers of these papers is that PCT explains the
_existing_ observations of behavior made by social scientists. But,
as we all know, existing observations of behavior are mainly
observations of aspects of control -- disturbance-output ("S-R")
relationships, feedback function - output ("operant") relationships,
changes in reference input ("cognitive processes") -- but not of
control itself.

Although the tracking task is a very powerful demonstration of control, I feel
we need to move beyond it, to extend PCT and show it working in other areas
and disciplines. I think it is somehow too simple for (some) people to grasp
the far reaching implications of the control of input from this demo. For PCT
to acquire converts from conventional science we need to produce more
"sophisticated" research that simulates or explains more complex behavior in
order to convince others that what we are doing does not involve only "toy"
problems (though our goal should be to do PCT not to convince others).

Here are three areas of research that I think would be the most useful and
productive ways to go.

1) Psychology

(Rick Marken (970826.0850))

What I would like to see are more publications that describe "flesh
and blood" examples of control. For example, I would like to see
some studies illustrating control occurring in conversations
between people

Yes this would be good. But I wonder how easy it is to actually measure what
is going on to "prove" that PCT is happening. In such situations it seems to
me that humans are too flexible in that reference signals are constantly
changing. i.e. you might be testing for one controlled variable when the
person has gone on to control for something else. Would it be better to do
the tests in controlled lab environments without distractions ? Anyway, in the
end are we not still just looking at observed behaviour and applying our own
PCT interpretations on what we see, which is our criticism of conventional
Psychology? Perhaps, more convincing Tests could be done on low-level
control systems that are more reliable ? Such as, we all know how intensity
is varied by the iris when we walk from the bright outside into a dark inside,
but does the iris adjust (slightly) perhaps, when we are trying to recognize
an object that is darker than its surroundings?

(Rick Marken (970826.0850))

I will try to start following my own advice; I will try to collect
and describe examples of the kind of control (purposeful behavior)
that goes on in everyday life.

(Bill Powers (970826.1053 MDT))

Maybe we should just collect these examples, put them in a standard format,
and when we have enough of them publish Human Control Behavior, Vol. 1. Then
start collecting more.

When we pile enough observations against that wall, it may collapse.

I think this is the way to go and build up a database of examples at all
levels of control that we can cite whenever we promote PCT.

2) Ethology - study of animal behavior

I wonder if doing the Test on animals instead of humans might be more
profitable. Especially on "lower" animals such as the wasp example I recently
gave. I should imagine that the variables that animals control are more
constant and reliable.

3) AI/ALife

The area of research that I would favor most is simulations of control systems
to produce simple animal behaviors. Ideal for this, I think, is java web
demos; easily accessible for all to play with and contribute. I would suggest
the best way to build such simulations is in an incremental fashion, starting
with the lowest levels and incorporating each higher level control system when
a more complex behaviour is called for. For example, it would be fairly simple
to produce java demos along the lines of the "crowd" demo. Initially, you
could have a single "animat" with a couple of motors that wanders around
controlling for a source of warmth or light. Building up to one that avoids
obstacles, has varying hunger levels, seeks out food and water, etc., etc. I
intend to start doing this when I've finished my Ph.D., though, of course,
anyone is welcome to start without me.

If those of us who "understand" how control theory applies to living systems
are unable to extend these areas of research then perhaps conventional
scientists are justified in rejecting PCT.

"Build it and they will come."

Ah, I've just seen Bill's Build-A-Bug post. This is great, just what I was
talking about!



Centre for Vision, Speech and Signal Processing University of Surrey
World Wide Web:
Office Telephone: 01483 259838 Home Telephone: 01483 533120