[From Bill Powers (960326.0700 MST)]
Martin Taylor 960321 14:00 --
I can see that my reactions to your discussion of redundancy must be
extremely frustrating for you. Just what am I disagreeing about? At one
moment I seem to agree and at the next I don't. Maybe this is the
problem:
You shouldn't confuse the notion of "pattern" with the output of a
pattern recognizer. A pattern is whatever set of values happens to
occur. It's a set of numbers, not a single value out of a
recognizer. The set could be an input to a pattern recognizer, but
never its output.
When I say "pattern" I mean _only_ the output of a pattern recognizer. I
assume that the pattern that is there in the input set depends entirely
on the pattern recognizer that is being used. You appear to be assuming
that the pattern is _really there_ in some sense, that that the pattern
recognizer simply responds to it. That is, if there is a square-
recognizer and there is a signal coming out of it, there must be a
square in the input set. The way I am looking at it, there doesn't need
to be a square at the input: a circle will create some degree of
squareness-signal (not much, but greater than the degree that an
amoeboid shape would generate).
I think you may be speaking out of a tradition of pattern-recognition
theories in which the epistemological position is that the pattern
exists first, and then is recognized.
The statement was exactly that the reproduction is _of the values
at the sensor inputs_. There is no claim that the device reproduces
stripes. The claim is that _so long as the input is stripes_ the
output will reproduce the input _values_ precisely. When the input
is not stripes, the output will be something different from the
input, most probably matching whatever pattern of stripes would
have been "closest" to the actual input.
There are two levels involved here. One is the level at which an
environmental field is represented by intensity sensors as an array of
intensity signals. The second is the level at which the array of
intensity signals is recognized and represented as a pattern of stripes.
A compact pattern-recognizer could economically produce a pattern-signal
based on spatial frequencies, directions, etc.. Using such information,
a subsequent system could produce an idealized pattern simply by
generating the correct spatial frequencies, etc. without requiring more
detailed information. So you see, I do understand what you are talking
about. If this were the only level of perception involved, then for that
level, reproducing the idealized pattern would be the same as
reproducing the actual distribution of intensities, because the pattern-
recognizer can't tell the difference between the actual distribution and
the idealized one.
However, an observer who could see both the idealized pattern and the
actual intensity distribution _could_ tell the difference, because there
would normally be a difference. Such an observer would be a control
system that is controlling the intensity at a given point rather than a
pattern -- a lower-level system. So what is redundant for the purpose of
controlling a pattern is not redundant for the purpose of controlling
_simultaneously and at a lower level_ an intensity.
The point at issue is not really a point about exact reproduction.
It's a point about what perceptual functions make sense to include
in a control hierarchy that works in a real world. The control
hierarchy works because its actions have a somewhat consistent
effect on its perceptions. If those perceptions are values of
functions that relate to frequent patterns that the sensors
provide, control will work better than if the perceptions are
values of functions relating to patterns that almost never occur.
In the latter case, the values of the perceptions will be near
zero, almost regardless of the control actions. The perceptions
won't occur, and reorganization will not allow those perceptions to
be controlled. The perceptual functions that persist are those for
which actions can, and probably will, make a difference. And that
depends on the perceptual functions taking advantage of the
redundancies of their inputs at _all_ levels of the hierarchy.
But doesn't this boil down to saying that we have the perceptual input
functions that we do have? I agree with your point that these are likely
to be the ones that create controllable perceptions, and that such
perceptions probably have more significance (practically speaking) in
the external world than perceptions that hardly ever occur.
But it seems to me that you go too far in attributing the redundancy to
the inputs rather than to the way the inputs are combined in input
functions. It seems to me that you are assuming that there is no _other_
set of input functions that would do equally well -- that somehow the
particular ones we do end up with have a unique significance in the
external world. As you know, I would be extremely interested in finding
that this is true, because it would provide one of the missing links
between subjective experience and reality. But it seems to me that in
your very definition of redundancy you are asserting the consequent;
assuming what I am sure we would both like to prove.
···
-----------------------------------------------------------------------
RE: predictability (Rick Marken and Martin Taylor) --
No need for me to butt in on a successful discussion, but I did have a
possibly useful thought. To say that something is predictable is not to
say that the control system controlling it is doing any predicting (that
is, literally remembering the history and using it to forecast the
future behavior). Of course it's hard to design any system that uses
integration or derivatives without giving it an ability that _can be
interepreted as_ some form of prediction, but if every process is seen
as prediction, prediction ceases to be a useful way to characterize
processes.
Oh, one other thought, sparked by Martin's observation that even with a
lot of noise, _some_ control is possible.
I tend to lose interest in discussions of control when we get to the
point of asking whether some vestige of control can still be seen. This
is because the message I've been trying to get across is concerned
mainly with really GOOD control -- accurate, quick, and consistent.
There is so much of that kind of control going on that it's almost
invisible; you really have to search to find cases where control is weak
and uncertain, but the problem is that you also have to search, in a
different way, to recognize all the fantastically good control that is
going on all the time. Everyone seems to take good control for granted,
assuming that it's just what people "do." It's as if real control isn't
very interesting, so you have to look for difficult situations where
people can just barely control at all in order to find something to
study.
You know, E. coli really doesn't control very well; we shjould remember
Dr. Johnson commenting on the dog that walks on its hind legs. It
doesn't do it very well, but it's interesting that it does it at all.
-----------------------------------------------------------------------
Rick Marken (960325.1300) --
I think Glasser's problem is that he assumed that he already knew
control theory when he ran into Bill Powers; therefore, he never
took the trouble to learn control theory.
That's a bit unfair, Rick. After Glasser found B:CP he phoned me, and
while he was writing _Stations of the mind_ he phoned me at least twice
a week for long conversations on control theory -- this went on for over
a year; as I remember it, close to two years. So he didn't just grab the
ideas and run with them; far from it. He put out a lot of effort to
understand.
I've just come back from Chapel Hill, N.C., having given two all-day
seminars to Reality Therapists from a wide region around the South. I
think they all have a pretty good grounding in PCT. Perry Good and Diane
Gossen in particular understand PCT very well and do their best to make
sure that what they write and teach is technically in line, although
they make no pretense of having technical expertise. Most of the people
there learned what they know, initially, from Glasser. They were all
interested and enthusiastic. It was a pleasant two days, and there were
many things I didn't have to explain to them.
What has happened in the Reality Therapy movement is that many people
have gone from Glasser's introduction to control theory to wanting more,
and have started trying to learn PCT more directly and in more depth.
All of the 42 people at the first seminar were there for that reason.
Many of them stayed for the second day, which was billed as being more
technical (and was, slightly). There is no repudiation of Glasser behind
this; just a desire to get a little further into the subject than
Glasser can take them.
These are NOT engineers or physicists or mathematicians; they are
teachers and therapists and consultants who just want to do the most
good that they can. To them, PCT is simply a theory of human nature that
makes sense and clearly applies to the problems they face in their work.
They don't care about information theory or adaptive control or E. coli
or stick-wiggling; they're just looking for something that will bring
some order into the messes they have to handle every day. It's the
_basic_ concept of PCT that they are interested in, not the decorations
or academic nit-picking.
There are some difficulties between PCT and Bill Glasser. We have to
realize that Glasser has worked for many years to build up a large
organization with a network of regional units, a certification program
that makes a lot of money, and a power structure. PCTers, if they were
also building a large organization with similar aspirations, might well
constitute a threat. My appearing in Chapel Hill to talk to RTers could
be seen as an attempt to woo Glasser's empire away from him. I hope it
is realized all around that I wouldn't know what to do with an empire
and don't want one. I think that the people I've met understand that.
But the exposure to PCT has been having effects on people in RT --
they're starting to question the infallibility of their leader and to
demand the right to make their own contributions to RT, particularly
with regard to applying PCT. Control theory is starting to become at
least as important to them as the traditional base of ideas that Glasser
developed when he was starting. It's not so much that there are
breakaway groups developing in RT; what's happening is more like a
grass-roots attempt to expand RT and make it more explicitly based on
control theory. There is still considerable loyalty to Bill Glasser
(deserved, I think) and no wish that I can see to disband the RT
movement. It's just evolving.
I was invited to Chapel Hill, and the meeting was organized, by Fred
Good, whose publishing company, New View, now publishes all our CSG
books (and handles B:CP), as well as publishing many RT books. Fred and
many of the young people who run New View are throwing themselves behind
PCT as the wave of the future. They want to organize our 1997 CSG
meeting. The PCT materials will be moved up to the front of the next
catalog, and the emphasis will be shifted very strongly to control
theory as presented in PCT writings. It's clear that PCT (and not just
"control theory") is going to figure very prominently in the Reality
Therapy of the future. I think it's likely that Reality Therapy will
continue as a movement, and that Bill Glasser will continue to run it,
but that by popular demand it will begin to include PCT as an explicit
part of its structure.
One of these days we're going to have to get together with Glasser and
figure out a _modus vivendi_. All this political crap is a drag.
-----------------------------------------------------------------------
Bruce Gregory 960325.1742 EST --
How about setting your margins a little narrower? Say 75 characters? I'm
getting a lot of alternations between long lines and one-word lines.
----------------------------------------------------------------------
Rick Marken (960325.1830) -- (replying to Bruce Gregory) --
Bruce:
When flying I sometimes try to predict the actions I will have to
take (such as when landing in a crosswind), but I find that this
anticipation doesn't really help all that much.
Rick:
I think that when we have the feeling of "anticipating" actions, we
are actually controlling a high level perceptual aspect of those
actions (like a particular sequence or program of actions to
perform when landing in a crosswind).
It's even simpler than that. You control two variables: (1) you see the
runway as aligned with your line of sight, and (2) the rate of drift
left or right is zero. When you've accomplished this, you find that the
airplane is pointing off to one side, which is really the first you know
about the crosswind. The main thing that's different if you learn about
the crosswind from the radio is that you aren't surprised to see which
way the airplane is aimed.
A crosswind landing _could_ be done by prediction, using information
about the direction and speed of the wind, but it's much easier, and
more reliable, just to control the right perceptions.
-----------------------------------------------------------------------
Best to all,
Bill P.