Discouraged, depressed and reorganizing

[From Rick Marken (961119.2000)]

Peter J. Burke (951119.1900)--

I get discouraged and depressed reading all the posts arguing about how
the cat can get out of the box, etc. I see nothing in these discussions that
add to our understanding of PCT.

What a coincidence. I've been getting discouraged and depressed reading
these posts as well. But it in my case it was because people seem to see
nothing in these discussions that add to their understanding of PCT. It's
nice to know that my discouragement and depression (like my paranoia) is
not baseless;-)

There is no question (in my mind) that PCT has a much better approach
than most others to understanding behavior,

Why are you so sure that PCT is better than other approach to
understanding behavior? I really would like to know why you think
that this is so. Also, if PCT is better than "most" other approaches,
which other approaches are better than (or tied with) PCT?

but I still say work needs to be done on (re)organization and plausalble
models need to be built and experimented with. PCT remains terrible
incomplete without a strong theory of (re)organization.

Why do you say that work needs to be done on reorganization? Have you
collected data on that cannot be explained in terms of ordinary control
(variations in action to protect intended results from disturbance) and seems
to require an explanation in terms of reorganization? As far as plausible
models go, they have been built and experimented with; they're there if you
want them (see below).

I would like to see the modelers generate some plausiable models that
might serve as theoretical guides to the process of empirical research.

One simple model is this: change a parameter of a control loop, such as the
gain, g, in an effort to bring a measure of the error in that control loop,
such as the average error over the last minute, e', to zero. So the
reorganizing system controls a perception, e', by varying a paerameter of
the control loop, g, in which e' exists; the reorganizing system trys to make
a percpetoin of e'match a reference of 0.

The reorganizing system can't know how to change g in order to make e'
approach 0 so it makes random changes in g randomly. It does this by
periodically adding a delta to g(g := g + delta). The time between additions
of delta depends on the size of e'; when e' is large the time between
additions of delta is short (high rate of reorganization); when e' is small
the time between additions is long (low rate of reorganization). The same
value of delta is added to g as long as e' is approaching 0; when the absolute
value of e' starts to increase, a new delta is randomly selected; the new
delta is randomly selected from some range of values; -0.05 to +0.05, for
example.

Best

Rick

[Peter Burke (11/20/95 10:46 )]

[From Rick Marken (961119.2000)]

but I still say work needs to be done on (re)organization and plausalble
models need to be built and experimented with. PCT remains terrible
incomplete without a strong theory of (re)organization.

Why do you say that work needs to be done on reorganization? Have you
collected data on that cannot be explained in terms of ordinary control
(variations in action to protect intended results from disturbance) and seems
to require an explanation in terms of reorganization? As far as plausible
models go, they have been built and experimented with; they're there if you
want them (see below).

Are you familiar with the work of John Holland on classifier systems? Here is a
adaptive model that basically starts with nothing except perceptions and
organizes itself to "survive" in an enviornment. Clearly, it is not quite
that simple, but (re)organization is an essential part of the basic model. I
have done some work that modifies the basic classifier system to introduce
goals (standard) and an "error" signal so that it becomes an ECU. The work
in only preliminary, but it seems fruitful so far. Why does work need to be
done on (re)organization? Because the basic ECU is not adaptive. Because we
cannot explain its emergence. Because we cannot explain how it will continue
to survive in a changing environment. Because we don't know how an ECU can
become a hierarchically organized set of ECUs. Lots of reasons.

I would like to see the modelers generate some plausiable models that
might serve as theoretical guides to the process of empirical research.

One simple model is this: change a parameter of a control loop, such as the
gain, g, in an effort to bring a measure of the error in that control loop,
such as the average error over the last minute, e', to zero. So the
reorganizing system controls a perception, e', by varying a paerameter of
the control loop, g, in which e' exists; the reorganizing system trys to make
a percpetoin of e'match a reference of 0.

The reorganizing system can't know how to change g in order to make e'
approach 0 so it makes random changes in g randomly. It does this by
periodically adding a delta to g(g := g + delta). The time between additions
of delta depends on the size of e'; when e' is large the time between
additions of delta is short (high rate of reorganization); when e' is small
the time between additions is long (low rate of reorganization). The same
value of delta is added to g as long as e' is approaching 0; when the absolute
value of e' starts to increase, a new delta is randomly selected; the new
delta is randomly selected from some range of values; -0.05 to +0.05, for
example.

Nice suggestion. I had thought of something like that, but had not yet
implemented it. Do you have any working models with this principle built in?

Peter

ยทยทยท

At 08:01 PM 11/19/95 -0800, you wrote:
------------------------------------------------------------------------
Peter J. Burke Phone: 509/335-3249
Sociology Fax: 509/335-6419
Washington State University
Pullman, WA 99164-4020 E-mail: burkep@unicorn.it.wsu.edu
------------------------------------------------------------------------