[From Bill Powers (940224.2030 MST)]
Cliff Joslyn (940224.1400) --
Regarding the underpinnings of PCT:
There was no one in cybernetics/systems theory after Ashby's book
in 1953 (Design for a Brain) from whom I learned anything about
control theory and its role in behavior. Wiener's book of 1948,
which I read in 1952 thanks to Kirk Sattley, got me started: the
concept of feedback control, and the particular relations to
behavior that he laid out, clicked in my mind as the obvious
successor to all the psychological models I had ever heard of,
including the one in which I then believed. Ashby's book gave me
an organized view of how one would start applying these
principles on a grander scale -- it was as much his organization
as his ideas that turned me on.
But Ashby lost me when he starting treating behavior as if it
came in little either-or packages -- I felt he had abandoned the
main trail and was going off in unproductive directions. I
especially felt, later, that his drive for the utmost generality
was premature and based on only a sketchy understanding of
control systems.
My main mentors were the control engineers themselves, and
especially the pioneers of analogue computing and simulation:
Philbrick, Korn and Korn, and Soroka, who not only provided the
machinery and systematized the art of analogue computing, but
developed penetrating insights into the principles of negative
feedback. I never met any of my mentors, in or out of
cybernetics: I just read their books and manuals. Wiener and
Ashby inspired me to go back to the sources of the ideas that
they had adopted. When I did, I gradually came to realize that
neither of them had learned very much about control systems.
ยทยทยท
-------------------------------------
You question the primacy of control theory as used in PCT:
(B) the particular negative feedback loop architecture that PCT
advocates.
Unlike many other approaches, PCT does not assume an architecture
and then look for phenomena which fit it. It starts with the
simple fact that organisms can produce regular and disturbance-
resistant outcomes despite the fact that their motor outputs have
highly variable effects on the local environment. As far as we
know, this can be explained only if the organism is able to
represent the outcome inside itself, compare the current state of
the outcome with an internally-define intended state, and convert
the difference into an amount and direction of action that will
keep the difference small. That is the basic architecture of PCT,
and the only one of which I have heard that can actually explain
what we observe.
... you have shown a very interesting result of SYSTEMS THEORY:
namely, that a particular real-world phenomenon requires a
particular system architecture, independent of the type of
components.
But isn't this a platitude? It would be more surprising if a
real-world phenomenon required NO particular system architecture.
The phenomenon is simply an expression of the architecture; a
different architecture would result in different phenomena. It
has been the case for over 300 years that when we observe a
phenomenon, we try to relate it to the properties of the objects
involved in it. If a general theory is to prove useful or
interesting, at some point it must tell us something we didn't
already know.
My beef with general systems theory is that while it purports to
apply to ALL systems, so far it has had to wait for others to
explain particular systems in detail before it can claim to have
known the result all along.
If BOTH (propositions mentioned) are true then you have
correctly defined PCT with respect to ST, namely that it
concerns systems OF ANY TYPE which demonstrate control
phenomena and, equivalently, have correctly constructed
negative feedback loops.
We have shown that a negative feedback system with a specific
architecture will reproduce the phenomenon we call control (as
opposed to what some others call control). Neither we nor any
other person knows whether some other kind of system could not
equally well explain the same phenomenon. We may not now know
what such a system might be, but simply to assume that no other
idea will ever be discovered is unwarranted; we have simply come
up with one positive instance of a type of system that will
create the observed phenomenon. To claim on this basis that PCT
is the ultimate general theory of control is not legitimate and I
do not make that claim. Any theory depends on the factual truth
of its postulates. This is the Achilles' heel of all claims about
"general" theories. You can show that a general theory is
consistent with its premises, but theorizing will not show
whether those premises are related to the real world or whether
some other set of premises would not serve just as well and will
not turn up tomorrow.
In discussing how ST people could be doing PCT "without knowing
it," you say
The idea is that (1) an ST person considers the operation of
living systems; (2) (s)he considers that feedback may be
important; (3) (s)he then uses feedback to describe some
interesting result. Bingo.
How many of these people, in considering the operation of living
systems, have considered the phenomena with which PCT is
concerned? How many, in considering that feedback may be
important, have correctly analyzed the way in which it is
important, and the consequences that it creates? How many, in
using feedback to describe some interesting result, have used it
correctly, and with respect to a result that actually occurs as
opposed to one that is only imagined? "Bingo" requires that you
have markers on all five numbers, and I have seen no evidence of
that outside PCT.
Also, it depends on if you take the term "living system" to
STRICTLY mean a single organism or merely a system which
INCLUDES an organism.
From your own writings, I glean that there is very little
agreement in ST on what constitutes a "system" or how a living
system differs from other sorts. If you can freely apply a basic
term to vastly different situations, you may create the illusion
of generality but what you actually achieve is vagueness. I don't
really care what you call "a system." The term is hopelessly
compromised by careless usage and lack of definition. What I care
about is explaining behavior.
For example, is an economy a living system or not?
If we agree on an answer, what will we know that we don't know
now? We can create categories at the drop of a hat, with any
membership we please. Sure, if you want to include organisms and
interactions among organisms in the same category, an economy is
a living system. If you don't, it isn't. What difference does it
make?
If so (I think this is cleaner), then for example any
economist, whether an ST economist or not, who presumes that
individuals have desires (like the desire for food) and make
economic decisions based on satisfying those desires (like
purchasing food) is ACTUALLY doing PCT.
No, that's too much! PCT is about what it is to have a desire,
about the relationship of desires to actions and their
consequences. It's about how making a decision or having a desire
gets turned into just those actions which will have effects in
the real world that result in an outcome that matches the
decision or satisfies the desire, even if the action required
differs from one instance to another. An economist who says only
what you describe hasn't a clue about how any of these obvious
phenomena come into being: he's simply describing the phenomena
that need an explanation.
The conclusions you can draw from PCT match what anyone can
observe under natural conditions. That says it is a good theory.
It should surprise nobody that an economist who uses common sense
will see that desires relate to what people purchase. That's
commonplace, it's not an insight and it's not a theory. It's just
a description of something ordinary in ordinary terms. That is
where you would START if you wanted to apply PCT. You don't need
PCT to conclude that people desire things and act to satisfy the
desires. What you need PCT for is to explain how they can
possibly do that. Can this economist of whom you speak explain
how it is that when a person decides to purchase Grape-Nuts, the
result is a long train of motor actions that carries the person
from one store to another until the Grape-Nuts are in fact
selected, carried to the checkout counter, and paid for? Of
course not. The economist has no idea how a decision or a desire
gets fulfilled, because the economist doesn't know anything about
PCT. I know of only one economist who does know anything about
it.
... the study of systems of all kinds, NO MATTER HOW THEY'RE
HOOKED UP, is ALSO very interesting (at least to me!), and
THAT'S what ST is about.
I dispute whether ST is about systems of ALL kinds, and whether
it has deduced the properties of ALL systems NO MATTER HOW THEY
ARE HOOKED UP. It is about a certain range of systems that fall
within the definitions of system with which ST begins. It is
unlikely, furthermore, that ST will have deduced everything there
is to say even about systems within that range, because
essentially no time is spent exploring the properties of specific
examples of systems, and looking for unexpected behaviors in
natural examples of those systems (when the systems are
physically realizable). Or put it this way: in general statements
about systems, how come I can so often think of counterexamples?
Everyone is entitled to be interested in whatever seems
interesting. Conflict arises, however, when there is competition
to see whose idea anticipates whose idea. A common strategy, in
and out of science, is for people to go up a level of
abstraction, trying to make true statements that anticipate true
statements that others might make at a lower level. You say,
"It's going to rain tomorrow." I say "There is a chance of rain
tomorrow," thereby seizing the opportunity to prove me wrong and
you right if it doesn't rain tomorrow. And the third guy, looking
for another step up, says "Of course it could snow as well," thus
showing that he has a more general understanding of the situation
than either of us. In this game of who is rightest, the
temptation is strong to rely on more and more remote abstractions
with less and less chance of being contradicted by the facts.
But in my book, it's the guy who says "It's going to rain
tomorrow" who wins in the end. Even if this guy is wrong, he is
going to be less wrong the next time, and finally he will be
right most of the time. The guy at the top level of abstraction
will see to it that he is right all of the time, but that will be
only because he has covered his ass in all possible ways. There
are no prizes for predicting that tomorrow there will be weather,
even if that should prove to be true.
----------------------------------------------------------------
Best,
Bill P.