[From Rick Marken (950919.1330)
Jeff Vancouver (950918.1530) --
I am still of the mind that theories are not disconfirmed, but models and
hypothesis are.
I don't know what you have in mind as the distinction between 'theory' and
'model' but you may be thinking of a "theory" as something more generic than
'model'. For example, I think it is reasonable to describe PCT as a closed-
loop as opposed to an 'open-loop' (cause-effect) theory of behavior. There
are many different models that are consistent with this closed-loop theory.
Some of these models will be disconfirmed, others not, while the general
theory (closed-loop) remains in tact. But I think it is possible to
disconfirm the general theory (and, implicitly, disconfirm any model based on
that theory). Bourbon and Powers did this in a paper called 'Models and
Their Worlds'. The results of the experiments described in that paper clearly
disconfirmed the open-loop (cause-effect) theory of behavior and (implicitly)
all models based on it.
Bruce Abbott (950918.1425 EST) --
The problem in 1979, as I have come to understand it, is that researchers in
the field (including both physiologists and psychologists) reified the
standard block diagram of the control model, found evidence that
contradicted that particular reified model, and concluded that "the" control
model was incorrect.
But the model that is always presented as the one preferred to the control
model is not a better control model but, rather some version of a cause-
effect model. This happens all the time. I donUt think that the reason this
happens is because researchers have reified the standard block model; it is
because they have failed to see that control is a real phenomenon. Control
theory is a theory of control. Control is as real and as concrete as linear
acceleration. As long as control is seen as just another Rpoint of viewS on
behavior, control theory will just be another theory of behavior -- and it
not the best theory at that.
Bruce Abbott (950919.1235 EST) --
I really have no quarrel with most of Bolles's analysis.
I sure do. It's not even an analysis, really, since he deals with the
system as though there were only an effect of inputs on outputs; he
completely ignores the simultaneous effect of outputs on inputs.
What seems to be lacking in Bolles's exposition is any recognition that
the analytic tools of control theory apply just as well to the kinds of
systems he describes as they do to systems that physically parallel the
classic diagram.
It's a lot deeper than this, I think. Bolles doesn't know what control is. He
thinks he is dealing with a cause-effect phenomenon. So, naturally, he
rejects control theory in favor of a cause effect model. If he understood
that control is a phenomenon to be explained, he might have rejected the
simple control model in favor of a control model with a Rdead zoneS near the
reference state of the controlled variable or whatever. But he ran off with a
cause-effect hussy faster than you could say "but what about the kids and the
mortgage payments".
Me:
your control system must have a precise measure of its own output
Hans Blom (950919) --
Please read your own (or Bill's code) of the "classical" PCT-controller
with its "leaky integrator". In all of these controllers I find a statement
that starts with u := u + ... This means that the controller, in computing
its new u, uses (and therefore must know!) its previous u.
Actually, we're both wrong. The PCT model doesn't need a precise measure of
its own output and neither does your model. This can be seen by adding a
random noise term to the outputs of both models. Just change the control
model to
u := (u+d[j]) +(r-x)
and change your model to:
dnew := x - b * (u+d[j])
where d[j] is a low pass noise disturbance, different from the disturbance
added to the output to produce the input, x.
The fact that your model works without a precise measure of output leads me
to suspect (along with Bill Powers (950919.0530 MDT)) that your method is
simply a limiting case of a general control method. In other words, itUs a
control of perception model; itUs just not clear yet how it relates to the
PCT algorithm.
Although your model is a control model, it doesn't seem to control (in simple
situations) like a person does. That is, it doesnUt deal with square wave
disturbances the way a human controller (and the PCT model) does.
Best
Rick