[From Bill Powers (920325.1100)]
Jeff Dooley (920324) --
Astute of you to pick up Kauffman's idea. I recognize what he is talking
about from experiments with reorganization, the E. coli models. Here are
some impressions I picked up, generalizations without benefit of systematic
verification.
There appears to be an optimum loop gain in a reorganizing system. If the
gain's too low, even large errors don't produce very frequent "tumbles," so
progress toward lower error is very slow. But if it's too high, favorable
directions of movement still produce a relatively short interval between
changes, so there isn't a chance for much progress toward the goal even
when the correct direction happens to result. Somewhere in between there's
an optimum relationship between the error signal and the tumbling rate.
Intuitively, I can see that reorganization has to allow enough time for
evaluating the consequences of the new "direction of movement" and to take
advantage of favorable directions.
I don't think the edge of chaos is as "razor thin" as you suggest. There's
probably a rather broad range over which the goal would be reached in time
for survival. Of course Kauffman seems to be thinking in terms of the scope
of the reorganizing changes rather than their frequency, but even so I
would think that there would be some leeway -- E. Coli's method of travel
seems to work just as well with three-dimensional tumbles as with 1-
dimensional ones, so the success isn't extremely sensitive to the number of
degrees of freedom simultaneously being changed.
I'll let Rick's good answer to your previous post stand in for mine.
ยทยทยท
--------------------------------------------------------------
Tom Bourbon (920325) --
You will be the envy of all pure CSGers. Some advice from my days long ago
at the VA Research Hospital, where I was supposed to have half of my time
to devote to control theory: learn quickly how to say no. When people see
that you actually know something, they will try to drag you into their
worst-conceived projects to rescue them. Let them sink. If you say yes, as
I did far too often, you'll end up wasting all your time on trivial
projects that need total revision to work. To reveal my deepest prejudices
based on too small a data base, my strongest advice is to avoid surgeons.
All of them I met had incredibly inflated concepts of the brilliance of
their own ideas and they assumed that everyone would just go along. I think
it has something to do with medical school.
On the other hand, maybe I just got a bad batch. But try to stick to
control theory if they'll let you.
You comments about baby spiders are excellent. To build on that, I think we
also have to remember that the chances of arriving in one evolutionary jump
at the production of baby spiders who disperse this way is zero. Before
they could disperse this way, they had to be able to spin silk at an early
age, climb plants and trees to high places, learn to jump off instead of
hang on, and of course do all the things that are needed to locomote up
irregular surfaces in a systematic way, counter to gravity. The few
examples of open-loop behavior we see -- and apparently the only kinds of
behavior that traditional scientists are capable of recognizing when they
see them -- are supported by a vast hierarchical structure of control
systems. If there really are any open-loop behaviors, their components are
all control behaviors, and as you say, they will remain only until long-
term environmental disturbances make them counterproductive again.
Evolution might throw up an occasional open-loop reponse to a stimulus, but
its chances of survival in the company of other variants that do the same
thing by control -- in a disturbance-resistant way -- are pretty poor. It's
sad, but when the only behavior you can recognize is a response to a
stimulus, you're going to miss practically all of what is going on.
As to control vs. noncontrol, we're up against history. The only widely
known alternative to the "scientific" concept of purpose (outcome) is the
metaphysical concept. To say that organisms have INNER purposes is, in the
view of most scientists, to classify yourself as a metaphysician. Somehow
scientists have to learn that there's an alternative both to the current
scientific view and to metaphysics. Not many of them go far enough into
control theory to realize that they're looking at something radically new.
What we need is come sort of quantum tunneling argument.
--------------------------------------------------------------------
Greg Williams (920324): a very astute comment.
I'm pleased that so many different people are independently coming up with
the correct arguments against Beer and (sorry Chris) Malcolm. The CT
concept of purpose does NOT mean just a consequence of an action, but an
INTENDED consequence. In all systems without feedback, it is only the human
designer and user who can see whether the consequence of a "purposeful"
(useful) action is the desired one, and take action to correct the result
if it's not. You can say that the purpose of a lawnmower is to cut grass,
but if you were to go right now to the place where you keep your lawnmower,
I'll bet any amount it would just be sitting there, not cutting grass or
accomplishing any other useful purpose. The human designer and operator of
these so-called "purposive" open-loop systems seems to be completely
invisible. We have to keep calling attention to that man behind the
curtain madly working the levers while the Great Oz roars.
--------------------------------------------------------------------
Best to all
Bill P.