[From Bill Powers (930505.1400 MDT)]
Gary Cziko (930505.1319) --
Well done! If the perceptual input sensitivity goes up, so does
the loop gain, because loop gain is the product of ALL
multiplying factors found in one trip around the loop. If a
disturbance is pushing the perceptual signal above the reference
signal, then increasing input sensitivity will reduce the
perceptual signal by some part of the remaining error. Of course
the opposite will happen if the disturbance is pushing the
perceptual signal to a value lower than the reference signal ...
One reason I wrote those demos is that playing with the actual
numbers is the best way to get the feel of the weird and
wonderful relationships in a control loop. I wonder how many
people have actually done this as Gary has done.
ยทยทยท
----------------------------------------------------------------
Tom Anastasio via Gary Cziko --
[I wonder] if Powers himself is the originator of these ideas
as they have been applied to oculomotor neurophysiology. I
would be surprised because I've never seen a reference to
Powers in all the papers I've read in my field concerned with
these topics.
All the basic concepts of HPCT were worked out between 1953 and
1960 by myself and Robert K. Clark, with support from the late
Bob McFarland, at the VA Research Hospital in Chicago. All the
original modeling was done on a Philbrick analog(ue) computer,
but I didn't get heavily into computer simulations until after
1973 when I got my first home computer (first DEC, then homebuilt
PCs, and so on). During the years from 1960 to 1973 I did little
but write drafts of the book and study what I could of
neuroanatomy. I was building low-light-level TV systems for
astromony to make a living. After the book came out I developed
the basic model for the tracking task (odd how hard it was,
considering how simple the result is).
It wasn't until the early 1980s that Wayne Hershberger of our CSG
got me interested in oculomotor control systems. Wayne provided
lots of Xeroxed references for me, include the work of Robinson
who seems to have his head screwed on straight. Wayne's interest
was in the subjective definition of "straight ahead," which
becomes pretty interesting when you get into it. None of the
modelers seemed to approach the control problem as I did, so I
developed a sort of talking model of the combined pursuit and
saccadic systems, which has a place in it for the vestibular
reflex. That's how I came to know a little about your subject.
This model rather foundered when I discovered that the horizontal
recti didn't seem to contain any stretch reflex, which was an
essential part of the model. Given the usual tendon and stretch
reflexes, I could explain ALL of the oculomotor illusions, and
even came up with a nifty way of explaining saccades to preset
positions as resulting from the _suppression_ of the pursuit
tracking system. But I put it all aside because the heart of it,
the stretch and force reflexes, didn't seem to exist where they
were needed. No doubt there is some other way to accomplish the
same end, but ...
I've sent copies of my scribblings to Wolfgang Zocher in
Hannover, Germany, because he is about to start trying some
simulations of a saccadic control system and I thought I could
save him some time. If you're interested I can send copies to
you, too.
What's new to me is the idea of hierarchical control in the
sense of nested control loops. This has very interesting
implications for the organization of the brain as a whole.
Yeah, I've thought so, too.
What I continue to be puzzled by, however, is the apparent
persecution that HPCT adherents seem to suffer.
It only feels like persecution. There's nobody out there
conspiring to Get Bill Powers and his Cohorts. The basic problem
(which took me a couple of decades to figure out) is with HPCT
itself. Even control-system engineers haven't figured out the
basic relationships in HPCT (as far as I know); their way of
analyzing control systems conceals all those simple elegant
relationships because they're interested in the outputs, not the
inputs, and they have habits of analysis that hide the intuitive
relationships -- a Bode or Nyquist diagram really doesn't create
clear pictures in the head of what the system is really doing,
and Laplace transforms just make everything more opaque. When you
try to get mathematically elegant and represent complex control
processes with complex matrices and vectors, you lose sight of
all the little details that actually explain what's happening.
Control engineers really aren't taught a lot about how to get a
feel for control systems. Bang, right into stability analysis.
With respect to the life sciences, HPCT is simply at odds with
all the fundamental conventional assumptions about what behavior
is and how it works. In order to understand and use HPCT, just
about every person brought up under traditional approaches to
behavior would have to scrap everything theoretical and start
over. Control theory is full of unexpected relationships; cause
and effect often seem to run backward. Nothing happens the way a
person trained in ordinary cause-effect thinking would expect it
to happen. This makes it almost impossible for a conventional
referee or editor to see anything in a PCT paper except a series
of obvious mistakes and incorrect inferences. Not understanding
the PCT explanation of a particular behavior in an experiment,
they simply fall back on explanations that they understand, and
complain that the PCT researcher doesn't grasp the accepted
explanation that everyone (who reads that journal) already knows.
So we get all these complaints that we haven't taken so-and-so's
work into account, the complainer never realizing that the PCT
approach offers a far simpler and more effective explanation of
the same phenomena, and shows that so-and-so are about a mile off
the right track.
PCT and HPCT are far more revolutionary ideas that I realized
during my first 20 years of working on them. A lot of people on
this Net will agree that when you start to understand the basic
concepts of PCT, the foundations of conventional behavioral
science start looking extremely shaky. I think this is dimly
sensed by those who resist the ideas of PCT and HPCT. This is not
just another theory -- it's a monkey wrench in the machinery.
Can it be true that scientists who want to test HPCT are
actively restrained from doing so?
There are all kinds of restraint, the most powerful coming from
inside. The one that PCTers have run into is the simple fact that
you can't get support for work on a new idea when you're the only
one who sees anything worthwhile in it. Even just publishing what
you are doing becomes extremely difficult. Our people are very
scattered; most members of the CSG are the only ones in their
departments or even their universities who have gone goggle-eyed
at this idea. It takes a lot of dedication to keep pressing even
while your colleagues are looking at you as if you're crazy. PCT
does not lead you onto the tenure track.
In my opinion, the case for HPCT could be made much more
strongly by citing evidence in its favor. This could involve
the use of HPCT to explain previously mysterious data, or the
results of actual tests designed to directly test the theory.
All of our experiment work (with human subjects) is of exactly
this kind. Ask Gary for a bibliography, and for copies of my Demo
programs. Every time we apply PCT we're testing it. We've come up
with all sorts of interesting explanations for data that would be
mysterious under conventional paradigms. But most of it has been
published only as Xeroxed papers passed around among us; this
work makes no sense to conventional editors and referees and
researchers.
What will really convince scientists is hard data, the more
neurophysiological the better, that begins to flesh out HPCT.
First you have to get through the door. A few years back, Gary
Cziko and Clark McPhail arranged for me to give a talk at the
Beckman Institute; I showed the Little Man (version 1) demo,
which involves pointing at a moving target with a 3 d.f. arm,
using a two-level array of eight control systems. I emphasized
that this model involved no trajectory planning and no special
calibration of the output functions; its behavior was determined
almost completely by the perceptual functions and feedback
effects. The model ran in real time on a portable PC. The
interest was polite; the host was more interested in showing me
how strange attractors could produce stability (of a sort). I
haven't heard a word since.
Last year, Greg Williams and I worked up version 2.0 of the arm
model, with the stretch and tendon reflexes explicitly modeled,
with two levels of kinesthetic control and one of visual control,
with a reasonably accurate muscle model, and with a 3 degree-of-
freedom arm modeled using the correct dynamical model in 3-space.
A total of nine control systems was involved. We also included
data showing that the model arm behaved very similarly to human
subjects in experiments by Atkeson at MIT. We sent it to Science,
and it was rejected without review.
Various members of the PCT crew have had similar experiences.
We're accumulating a collection of weird reviews from rejections,
and maybe some day we'll publish them.
This sort of thing does get you down after a while.
I suspect this is about the normal state of affairs for a true
scientific revolution. You going to join up and suffer with us?
--------------------------------------------------------------
Best,
Bill P.