[From Rick Marken (960722.1300)]
The meeting was wonderful. But I was sorry to see so few present from the PCT
science community. Most of the participants at the conference were from the
PCT applications community. I was really happy to see all the educators,
counselors and others who are using PCT in their practice; and the quality of
the applied presentations was the highest it's been in years. Many of the
applied people expressed some dismay, however, at the fact that so few from
the PCT research/modelling community were willing to make it a priority to
get to the only place on earth where they could discuss PCT with others
who understand it and/or want to understand it. I could really think of
nothing to reduce their dismay.
Bruce Abbott (960717.1215 EST)
Either the perceptual signal is a VERY slowly decaying leaky integral of
total shock, or the rat has developed a model of the environment...that
model would predict the delivery of shock at the end of 20 seconds of non-
responding.
I don't think these are the only two possibilities. For example, why not put
the very slowly decaying leaky integrator in the output of the shock control
loop, so that the reference for press rate changes VERY slowly? Also, the
"model based" explanation seems not only unnecessaary by implausible. I find
it hard to believe that a rat has the ability, let alone the poise, to
develop an accurate model of the intershock interval _while it is being
shocked_. And even if the poor rat could develop such a model, which is
basically a timer, it would have to be rather accurate for a long period
in order to account for the persistence of the shock avoiding press rate.
Anyway, I think the best way to answer your question is to write up the
various models and see which fits the data.
Hans Blom (960717b) --
Although, as you correctly note, my world view is different from yours, I
think I now understand yours better. I accept it as honorable.
Jeez, Hans, you mean your original hypothesis was that my position was
dishonorable? The chances for friendship just keep getting smaller and
smaller.
Me:
The mistaken conclusion is...that a control system must have information
about disturbing influences to the controlled variable in order to control
that variable.
Bill Leach (960717.1348) --
I think that your definition of model based control is far too restrictive.
To me using the imagination and then setting references based upon the
"world models" (or whatever) in the imagination is an example of model based
control.
I mind "unrestricting" my definition of model-based control in this way only
because I think it would make things confusing. What you want included as
"model based control" (setting reference signals based upon the world
models in imagination) is a model of something that really happens; we think
and then do. But what Hans calls "model based control" is a model of a non-
existent (certainly non-demonstrated) phenomenon.
Hans Blom (960717e) --
And here you [me;-)] leave an important part out. A better diagram is:
command ---> world model ---> output ---> result ---
^ |
>----------------------------------
See the feedback loop?
I know it's there. I left it out because the point of your model is that it
"controls" while the feedback loop is broken. As I said, your model can
control when the feedback loop is broken, but only when the world model is
designed to predict over long periods of time. And when the world model is
designed this way it controls poorly even when the feedback loop is intact.
When the world model is designed to predict over only a short period of time,
closed loop feedback control is much better -- better than that of the
standard PCT model -- but now, like the PCT model, your model cannot
"control" when the feedback loop is broken (when there is no perceeption of
the controlled variable).
Bill Leach (960718.1335) to Hans --
The basic statement that organisms generate output based upon internal
models is absolutely counter to the most fundamental principles of PCT.
I'd say that this statement is counter to a fundamental _observation_, not
a fundamental principle, of PCT: the observation is that organisms control.
Systems that generate outputs based on an internal model do not control so
they are not relevant to the behavior of organisms.
Martin Taylor (960719 11:10) --
_never_ have I claimed that a control system needs information about the
disturbing _influences_ in order to control the controlled variable.
Maybe the fact that it is _possible_ for the disturbance _waveform_ to be
reconstituted from the perceptual signal, provided control is good, is what
causes Rick to keep making these misstatements. He should note that the
disturbance waveform is not the disturbing influence, that the _possibility_
of reconstruction only shows the existence of an informational link between
the disturbance waveform and the perceptual signal, not the mechanism of
the control system.
Would it be correct, then, to state your position thus:
A control system must have information about disturbing WAVEFORMS in order
to control a variable.
Hans Blom (960718b) --
A hundred line computer program cannot mimick a living system
We use computer models to mimic the _behavior_ of a living system; in
particular, the behavior called _controlling_. A 3 line computer program can
mimic human controlling with an RMS error of less than 2% of the spread of
the data points. You have not shown that your 100 line model based control
system can mimic this behavior any better than the 3 line program. Can you
see why we're not impressed?
Best
Rick