From[Bill Williams 27 May 2004 2:50 AM CST]
[From Bill Powers (2004.05.25.1352 MDT)]
Bill Williams 25 May 2004 11:40 AM CST --
>In anycase I starting with, as I understand it the implications of
control
>theory, I have a arrived at a notion of "social" that may stand up to
>criticism. Given that it has been derived in a context of control
theory, as
>I understand it, is a bit different than other conceptions of "social."
This a very clear exposition of the way control theory enters into social
phenomena, and after one pass I can't remember disagreeing with anything
in
it.
Much as I value Bruce Nevin's joining into this discussion, I would like to
preserve
this element of agreement that we apparently reached. ( Not that you can't
decide
after re-reading and further reflection that my exposition contains a falacy
or falacies
that weren't apparent on a first reading. ) However what, I would like to
preserve is a
common understanding, if that is what it is, regarding how agents ( which
is as good
a word as any for present purposes ) interact with each other. I will
bring Bruce back
later to help out my argument.
First: it is physiologically implausible that any organism would be
constructed so that
its functioning could be controlled by an external cause. Autonomy seems to
be a
bed rock principle of organic behavior.
Second: Organic autonomy is not an unconditional autonomy. Rather it is an
autonomy
that is exercised in the context of some specific enviornment. The
implication of this is
that while the enviornment does not determine, or even influence behavior,
the
autonomous behavior of an organism will be different in different
environments. This,
as I understand it, is entirely consistent with a proper understanding of
control theory.
The control theory model with out any contradiction asserts that organic
behavior is
completely autonomous, and also that the behavior of an organism can and
will be
different depending upon the particular features of the environment.
These two propositions may appear to be entirely elementary and
unexceptional. At least
I hope they are. However, they specify a relationship between organic
behavior and the
environment which is different than the one that is ordinarily assumed.
Consider an organism in two enviornments A and B. Suppose we switch the
organism
from enviorment A to environment B. And suppose as a result the organism
changes its
behavior. Are we jusified in saying the switch from A to B _caused_ the
organism to
change its behavior? This question appears to me to be crucial in
understanding how
adopting the principles of control theory will neccesarily generate a change
in the conception
of social phenomena.
Ordinarily I think most people would say that switching the enviorment from
A to B _caused_
a change in the organisms behavior. However, if my premise that an
organism is autonomous
then can it be said that the switch from A to B _caused_ a change in an
autonomous creature?
I think there is a lot of opportunity for equvocation here. And, I think
the reason for this is that the
context of such a discussion introduces some abiguities. The context
involves an organism and
an environment. This is a different context than one in which "All we can
know is what we perceive.
If the slogan, "All we can know is what we perceive." is taken literally,
then it would appear that
the very notion of an enviornment is forbidden. We may not, in some sense,
be in a position to
know the environment other than through our perception of it. But, our
control theory, it would
seem does tell us that there _is_ an environment. And, that is knowing
_something_ about the
enviornment. So, having adopted a control theory ontology I feel confident
that I can refute the
claim that "All we can know is what we percieve." All I have to do is open
my little book, and
there in the diagram is an organism and an environment.
Now at present I am not sure how I am going to prove that there are other
people. Not that I really
doubt that there really are other people. Maybe the answer is to write a
book _Social Control
Theory_ by Bruce and Bill. But, first I would have to convince myself that
there really is a Bruce.
But, having done that, we could get to work, and eventually we would have a
book. Then when
people got tired of looking in control theory books written by solipists,
they could read this book
by Bruce and Bill. And, they could look at the title page, and they would
see "By george they have
solved the riddle of solipsism-- there are two people right here-- Bruce and
Bill." And, the solipists
will knash their teeth and wonder, however did they do that?
Bill Williams
Clark McPhail went through a similar analysis, arriving at similar
conclusions but in the context of sociology, where the resistance to
"individualism" must be at least as stiff as in economics.A useful term in this connection is "agent," unless that, too, has bad
connotations in economics. It's not quite as evocative of pins through
butterflies as "specimen", although Phil Runkel has taken away a lot of
that unwanted connotation. An agent is an entity capable of acting on the
world to produce effects, the implication being that the effects are
chosen
and intended. It might have an advantage in that "agent-based economics"
seems to be gaining ground, and may be approaching the principles of
control, if only tentatively. The term is just a nomination, not a
decision. No reason we can't use both.The idea of a theory of culture is attractive. From where I stand, the
chief difficulty would be in deciding just what we would want such a
theory
to accomplish. In the theory of agents, control theory can be used to
model
and predict the actions of a single agent in a specified set of
conditions,
and it also serves as a source of guidelines for research into the
internal
organization of single specimens. Out of control theory we get some very
general rules of thumb, about things like loop gain, stability,
relationships between disturbances and actions, and hierarchical control.So we can ask what sorts of general principles we would like to draw from
a
control-theoretic analysis of culture, and where we can get the data to
work from. There are, of course, tons of studies from a century or more of
social research, but unfortunately the quality of the data is such that
it's hard to draw conclusions from them with much predictive power. We may
be in a situation like that vis-a-vis psychology, where any value of past
research data for use in exploring control phenomena is strictly
accidental. The experimental designs simply didn't anticipate the needs of
research into control processes. This leaves us with little choice but to
do our own experiments, or get someone else to do them (which is a lot
harder, believe me).The other approach is to experiment with models. The "crowd" program was
designed for that purpose, and it revealed phenomena that, unknown to me,
were already well-known in the study of crowds -- the "arcs and rings"
configuration being one of the first to be noticed in the setup called
"guru." This is a branch of modeling that tends to show up in fields where
experimenting with real systems is difficult, expensive, or illegal. You
can answer such questions as "What is the effect on an astronaut's ability
to control position in space if one leg has to be amputated?" -- without
actually doing this to a real astronaut.Experimenting with models reveals emergent phenomena not obvious in the
initial design of the model. Consider a simple design consisting of two
control systems which are independent except that the actions of each one
symmetrically affect the controlled variable of the other (as well as
their
own). It's not obvious that this elementary relationship creates positive
feedback, but an experiment with a model will quickly show the instability
that appears when the strength of the mutual effects exceeds a certain
threshold. This sort of exploration may provide a different route to a
theory of culture, by suggesting non-obvious phenomena that can emerge
from
social interactions. Of course the underlying models of agents have to be
realistic, and the interactions tested have to be in the realm of the
possible, but this method is being used increasingly in other sciences,
such as astrophysics and materials sciences. We have used it in PCT to
explore the properties of systems in conflict (Kent McClelland) without
actually putting real people into conflict with each other (well, not on
purpose).The two approaches, experiments with real systems and experiments with
models, would, in the best of all possible worlds, be highly interactive,
with real-system experiments being used to develop and test models, and
experiments with models being used to point out new phenomena which can be
investigated in the real systems. It seems to me that this is bound to be
how we investigate social phenomena in the future.Finally, we can now do something that nobody could have done (on any
feasible budget) only ten or a dozen years ago: model large numbers of
agents interacting with each other in a complex environment. I'm talking
about hundreds, even thousands, of agents, in models that manipulate many
megabytes of data. So we can have our cake and eat it too: our macro
models can be built out of large numbers of microeconomic interactions, to
show the relationship between the detailed and general views of the
system.
Charlotte Bruun wrote eloquently about this aspect of modeling in her
thesis and the paper more specifically about her model (she hasn't
communicated for some time, but I'm leaving it up to her -- a busy
professional, a wife, and a mother of several very small children).The consumption-budget model we've been tossing around may be a place to
start, in that it involves a single agent, but is potentially part of a
larger system involving other sectors of the economy and other agents. If
we can get this one agent operating with acceptable realism in a sort of
amorphous fuzzy economy, we can then start giving form to other parts of
the economy with which this agent ineracts, as in econ004 and beyond. I
think we will discover things by experimenting with this model and its
successors, phenomena that we can hope to identify in the real economy. At
first, I'd be happy to "predict" things that everybody already knows --
that's how we make sure the model is working right. But I'm confident that
there are phenomena which no human being, however knowledgable or
insightful, could predict simply because the interactions are too complex
for anyone to comprehend without the aid of a model.I haven't even mentioned reorganization and learning. I'm irrationally
daunted by the prospect of setting up a big model and letting
reorganization loose to work on it. I guess I don't have much faith that
it
will really work. But what the hey, we'll try it one of these days.
Today I had a big break-through in converting Little Man Version 2 into
Delphi. It was originally in C, which of course is not compatible with
Delphi which uses Pascal, The existing code in Pascal did not contain all
the original features, and while I did (with a lot of help from my
friends)
salvage the original Pascal code for Version One (from the late 1980s), I
could not find any trace of an equivalent form of Version 2 in Pascal from
1989 and the early 90s -- the one with real physical dynamics in it. So I
decided to try to merge the two versions with the old C code as a guide,
and today I managed to get the little man pointing while looking at the
finger and target with two independently-aimed eyes in a movable head,
with
binocular vision for control in the depth dimension -- in Delphi. This
part
still works in the imagination mode (no physical dynamics yet), but I've
already tested the interface with the physical dynamics part of the
program
and joining the two parts should not take much longer. Then all that will
remain will be to use the Delphi tools for editing parameters, and
building
a "live block diagram" of the Version 2 model -- everyone liked that
feature of Version 1, which was left out of version two for lack of space
on the screen, so I want to include it in the Version 2 model.Then I think we will be able to turn to the economics modeling much more
seriously. I still have a lot of converting from Pascal to Delphi to do
(it
scares me to think that none of my old DOS-bsed programs will compile,
much
···
less run, on any working machine 10 years from now) but the Crowd
(converted) and the Little Man were the most complex programs and involved
learning a lot of new things about Delphi. The rest can go faster,
expecially if there is some volunteer help.Best,
Bill P.