[From Bruce Abbott (951026.1605 EST)]
Bill Powers (951026.1022 MDT) --
Bruce Abbott (951026.0940 EST)
Engineers at the time this was written (and indeed, even today) use
such language to describe how their engineered control systems
work, talking about the system's output rather than its input. But
in a control system, the two are the same, so there is nothing
technically "wrong" about the description.
Bruce, we really have to start taking the subject of disturbances
seriously. The output of a control system -- its action -- is NOT the
same as its input unless no disturbances are acting. I'm discounting
transformations that make the output different from the input even
without disturbances, as when you turn a crank to control the height of
a bucket in a well. Even in a simple proportional environment, there is
no regular relationship between outputs and inputs when disturbances are
acting.
Argh! Moments after that post went out it occurred to me that you would have
just this interpretation, which despite my poor choice of words is not at
all what I had in mind. This apparent gaff came up in the context my
discussing the descriptions engineers gave (and still give) of the standard
ECU, usually illustrated by something like the following diagram:
> disturbance
error V
reference ------>[comparator] -------->[output function] --X---> output
^ |
input | feedback |
···
+-------------------------------------+
This might be a diagram, for example, of an electronic voltage regulator.
In this diagram, the input and the output are the same signal: the system
monitors its own output and adjusts itself to compensate for any
disturbances that would alter the output voltage above or below the
reference value (e.g., -5V).
Here the "output" is identical to the perceptual input of the system
(including any effect of disturbance) and not to the action of the system.
To avoid misunderstanding, I should have noted the difference in usage.
Can't we talk about this and get this point settled? A few simple demos
should clarify the matter. It's the basic difference between
reinforcement theory and control theory.
Yes, I know. But that was reinforcement theory, 1930s style (Ye Olde S-R
theory). More recent elaborations of reinforcement theory have attempted to
resolve this difficulty by appeal to constructs like stimulus control,
generalization, and response classes. Some researchers believe that these
approaches have resolved the problem; I find them unconvincing, and the
control explanation compelling. Here's my understanding of the ECU (I
promise I won't cheat and look it up):
ref sig.---->[comparator]-- error sig.-------+
^ |
percept. sig.| |
[input function] [output function]
^ |
input | | output
+-----------X<------------+
^
>
disturbance
In this way of labeling things, the output is what the output function
produces; the disturbance comes later, and the input is thus output +
disturbance. S-R theory proposed that reinforcement established a
connection between stimulus input and the output function. like this:
input ----->[input function]---percept. sig.--->[output function] --->output
The problem is that this system does nothing to resist the effect of any
disturbances on the output, and thus cannot account for the fact that
organisms like rats, pigeons, and people vary their actions so as to achieve
a constant end. This diffulty was first brought to the attention of Clark
Hull, the preeminant S-R theorist of the day, by Edward Tolman, who showed
that rats trained to successfully negotiate a maze would find their way back
to the goal box on the first trial, without error, even when they were
required to use a completely different pattern of muscular contractions to
do so. Tolman, as you will recall, was the first experimental psychologist
to appreciate the purposiveness of behavior, although he lacked an
understanding of the mechanism by which it was achieved.
So much for Control Systems 101. Did I pass? (;->
Regards,
Bruce