NetNews Usenet Archive 1992 #30/NN_1992_30.iso/spool/bit/listserv/csgl/1900

Text File | 1992-12-17 | 9.7 KB | 212 lines

Comments: Gated by [e-mail]
Path: sparky!uunet!paladin.american.edu!auvm!VAXF.COLORADO.EDU!POWERS_W
X-Envelope-to: [e-mail]
X-VMS-To: @CSG
MIME-version: 1.0
Content-transfer-encoding: 7BIT
Message-ID: <[e-mail]>
Newsgroups: bit.listserv.csg-l
Date:         Thu, 17 Dec 1992 11:34:17 -0700
Sender:       "Control Systems Group Network (CSGnet)" <[e-mail]>
From:         "William T. Powers" <[e-mail]>
Subject:      Misc up-a-levels
Lines: 197

[From Bill Powers (921217.0930)]

John Gabriel (921216.1154) --

Your paraphrase of the point of PCT is ambiguous enough to
suggest that you may still have a big AHA waiting. If not, it
does no harm to make the point again:

>... people or any other organism, or even state machine, do
>things because of their perceptions, (for state machines, read
>inputs) or sometimes to change their perceptions. Carelessly
>put - In order to change some of their perceptions is the way
>the previous thought should have been stated.

I'd put this in a slightly different way: organized behavior
never occurs except with the aim of controlling some perception.
To speak about "changing" a perception might imply that we can do
an act to change a perception from one state to another, then
relax until it's time to change it again. To say that we
sometimes act to control perceptions could be taken to mean that
we sometimes act for other reasons. In fact controlling
perceptions almost always requires continuous behavior, because
the perception is dependent on the behavior as well as on
independent influences. And behavior never occurs except for the
purpose of controlling a perception.

Behavior -- and perhaps this is the part you meant to edit -- is
not done "because of perception." That's S-R theory. PCT says
that the perceptions that matter most to an organism are as they
are because of behavior, which in turn is as it is because of the
difference between those perceptions and internal reference
signals.

Just keeping the language tidied up.

I don't think I want to debate about the merits, necessity,
justifiability, etc. of the military philosophy any more.
Everything you have said points to a system concept and a set of
perceptions that adequately explain why numbers of people act as
they do with respect to military and related matters. The
relationship of those perceptions and reference levels to
observable actions is interesting from the theoretical PCT point
of view, but the particular content of the reference signals and
the perceptions is of no theoretical interest. Everyone argues
that his or her own reference signals are right. What else? I
would like to be able to do a serious study of these things, but
I don't have the resources and PCT needs a lot of more basic
development before that will be feasible.

More or less the same thing goes for wolves and baboons.
Anecdotal accounts of behavior may suggest certain explanations,
particularly if appeal is made to more conventional theories
instead of PCT. But they don't give us the kind of information we
would need in order to understand the perceptions being
controlled by these animals, or the reference levels associated  with them, or
 the amount and frequency of reorganization
involved. The data you would need to get that kind of information
is never taken because the people doing the observing don't know
control theory and don't know what to look for.

As to your conversation with Tom about advertising: the only
perceptions a person can control are that person's own
perceptions.

>Now, I'm not saying the hierarchy is nonsense, it isn't, but I
>only believe it somewhat more than I believe in the Id and Ego.

Does the following list of perceptions strike you as being of the
same nature as Ego and ID? Intensity, sensation, configuration,
transition, event, relationship, category, program, principle,
system concept. If so, I would be disappointed, because my
intention was to find types of perceptions on which everyone
could agree and which seem obvious and self-evident in ordinary
experience.

>There does seem to me a very convincing explanation of puzzling
>things in the reliability of people in doing jobs like picking
>up glasses of water, or following randomly moving cursors.
>Following the cursor has very clear reference signal - the
>cursor position, and I have no trouble understanding it.

If you think that the target position is an obvious reference
level, how do you explain it when a person intentionally keeps
the pointer two inches to the left of the moving target, or makes
the pointer describe a continuous circle around the moving target
(in two-dimensional "tracking")? Where is that reference
condition to be found in the environment?

How do you explain a person keeping a checkbook balanced without
appeal to some reference perception? I think you may be missing
the main phenomenon that PCT is about -- which is not tracking.

>I've been building feedback systems since 1941, and doing their
>mathematics since 1946 ...

Well, you're way ahead of me there. I've been doing it only since
1953, and my approach began mainly through analogue computing. I
could claim that my Navy electronics experience in 1944-46
counts somewhat, as I learned at least how to troubleshoot
control systems if not to analyze them.

>I think the following things are well founded...

>  1. The mathematical theory of linear feedback systems as put
>forth for example by Bode - Network Analysis and Feedback
>Amplifier Design 1945 Van Nostrand.

Did Bode ever notice that control systems control their own
sensor signals, not their outputs? I read most of that stuff, not
with very deep understanding, but I never noticed any statement
like that.
>2. The idea of classical contact transformations, as first put
>forward by W.R. Hamilton. The classical contact transformation
>is the operator that takes system state from that at time T to
>that at time T+dT ..

When this idea got into the hands of digital computer people it
became an unfortunate idea. The idea that the brain passes from
one discrete "state" to another, with nothing happening between
states, is completely wrong. The brain isn't clocked; neurons are
not in "1" or "0:" states; variables in the brain can't even be
measured at an instant (because frequency is the significant
measure in the context of behavior). And the idea of a whole-
system transfer function is a delusion -- unmeasureable for most
behaviors, and even where measurable, useless.

>3. The idea of the rate of information transmission down a
>discrete channel, being the the upper bound of the number of
>binary decisions a recipent can make in a second, and first put
>forth by Claude Shannon in the two 1949 papers in BSTJ.

But all real neural systems work with continuous, not discrete,
variables. Neurons do not respond to incoming impulse streams by
making "decisions" but by harboring continuously-variable
chemical concentrations and potentials which in turn determine
the frequency of outgoing impulses. The computations done by a
neuron are analogue computations based on continuous internal
electrochemical variables. Certainly information theory could be
applied to these processes. But it doesn't help you model them.

>Although Shannon does not discuss stability, I think stability
>is only an important side issue as compared to information.

Stability is the most difficult consideration in designing a
control system that actually behaves. If you've designed and
built a lot of control systems, it's hard for me to understand
how you could say information is more important. In all control-
engineering texts I have seen, about one chapter is devoted to
the basic principles of control per se, and all the rest is an
exposition of methods for measuring and achieving stability.
Information theory is hardly mentioned. Information content is of
secondary interest in models of behavior, because normal
perceptual signals are far above the noise level and are seldom
ambiguous.

>4. The ideas of conventional Decision Theory and Value Systems
>        are an adequate first approximation to human behaviour
>        in decision making to support qualitative analysis,
>        and if estimates can be made about values, quite good
>        quantitative analysis.

I think that your definition of "quite good quantitative
analysis" may be rather different from mine. Are you talking
about predicting the time-course of behavioral variables in each
individual case with an accuracy of, say, five percent? That's my
definition of "good."
>I think these things are a long way away from your interests,
>which makes fertile ground for misunderstanding. And perhaps we
>don't have the same ultimate objective after all.

It does begin to look that way, doesn't it?

I think you're going through the second stage of learning PCT.
The first stage is the big insight about control of perception.
The second stage begins when you start to see that if PCT is
correct, a lot of things you've believed up to now are probably
either wrong or irrelevant. If you're willing for the second
stage to continue, maybe you'll still be here a year from now.

I hope you will be.
------------------------------------------------------------
Rick Marken (921216.1200)--

I think I have to side with Martin Taylor on this one, Rick. In a
simple reorganizing model, e^2 might be a suitable driving signal
for the rate of reorganization. But that isn't the controlled
variable. It's the error signal.

I have also found in my experiments with reorganization that the
time-rate of change of error signal is important, too. The most
useful measure of error that I have found is the one that Martin
suggested some time ago: e^2 * (de^2/dt).

---------------------------------------------------------------
Best to all,

Bill P.