[From Bill Powers (931108.0900 MST)]
Chuck Tucker (931107) --
I certainly believe it is possible to construct a machine which
has both open and closed looped parts to it.
ALL machines, and people too, are constructed of components which
individually operate open-loop. Think of a muscle: signals in,
contractile forces out. Or a sensory nerve-ending: stimulation
in, perceptual signals out. Even when the components may have
internal control loops (biochemical, for example) that shape
their responses, we treat them as input-output devices at a
certain level of analysis. We break the whole complex interactive
system down until we reach a level of detail at which each
component can be described as a function making an output depend
on one or more inputs in a certain predictable way. Once the
components are all described in this way, we can begin putting
them back together by specifying which inputs come from which
outputs of other functions. This leads to a system of
interrelated components, the interrelations being now a system of
equations. When we solve those equations analytically or by
running models in simulation, and when the resulting system
produces behavior like that of the real system, we can say we
understand how the real system works. And we can predict its
behavior in a lot of new circumstances we would never have
thought of otherwise. We can also deduce what parts of the system
have failed when its overall behavior develops some sort of
problem.
My favorite example recently are the windshield wipers which
appear to be closed looped but are not at least on my car.
Why do the wipers appear closed-loop? When you turn them on, they
go back and forth. They will go back and forth even if the rubber
blade falls out. The shaft on which they are mounted will rotate
left and right even if the whole blade assembly is removed. Such
disturbances have no effect at all on the active part of the
system, the motor and circuits that create the oscillating
movement.
There's an obvious closed-loop aspect to the wipers which is easy
to overlook when you're focusing on the wipers instead of your
purpose in using them. If you turn the stalk or press the button
and the wipers don't go into motion, you get a great big error
signal. You try to turn the stalk farther or press the button
harder, until you figure out that the mechanism is broken. If the
wiper blade drops off, you immediately change your action to
avoid scratching the windshield.
Note how the language changes. When you're talking about the
mechanism itself, you say "it" moves the wiper assembly back and
forth across the windshield. But if the wiper blade drops off and
you don't immediately turn the wiper off, YOU would be scratching
the windshield. The wiper mechanism returns to its proper place
as a means by which you can achieve or avoid going against your
own purposes.
When you're trying to decide between open loop and closed loop
descriptions, remember to think about the effects of disturbances
of the output effect. If the device just keeps on blindly
producing the same action despite a drastic change in the effects
of the action, it's most probably open-loop.
···
--------------------------------------------------------------
Hugh Petrie (931108) --
Fooey. Welcome to the club.
Your protest letter is strong enough to bother a reasonable
person. I think it's very appropriate to protest. In my long
years of experience with referees, my two protest letters have
succeeded in reversing a verdict 1-1/2 times. The half time
resulted in a new review but another rejection. The whole time
resulted in publication (the 1971 rat paper).
The main thing is to keep up the effort and not let the bastards
wear you down.
--------------------------------------------------------------
Rick Marken (phone call yesterday) --
Thanks for bringing me down to earth. In my enthusiasm for my own
ideas I didn't bother to check out my brilliant new conception of
how a control system might switch automatically from closed-loop
to feedforward operation. The one factor I would have seen
immediately if I had actually tried a model (as you did) is that
with nothing but inhibitory reference signals, there is nothing
to cause any output or perceptual signal to appear, except
external disturbances. Some source of positive nonzero outputs
has to exist if this is going to work -- an equivalent of control
of muscle tone in systems that use opposing muscles to control an
arm.
With positive (excitatory) reference signals using opposed one-
way systems, everything works fine: the system can be turned off
just by setting both reference signals to zero. The inhibitory
perceptual feedback signals can't cause any output by themselves.
But unexpected loss of the perceptual signals during operation
does not lead to "nice" behavior.
Somewhere in there may still be a workable solution, but whatever
it is (if it exists) it isn't the one I proposed.
This makes me wonder whether the general concepts others have
proposed for this situation have actually been checked out with
running models. If so, I'd like to see an example. As I can
attest, some ideas just sound so self-evidently workable that one
doesn't bother to try them out. I can also attest that not trying
them out is a mistake.
---------------------------------------------------------------
isaac kurtzer (931107.1530) --
Some computing centers use weekend hours to shut down various
functions for maintenance. That can lead to messages being
delayed until the system comes on line again. It could be your
system, or it could be some other system enroute between the U.
of IL computer and yours -- or even the U. of IL computer itself.
The internet keeps trying until it succeeds, giving up only after
(I think) three days.
I got about 5 messages from CSG-L this weekend.
--------------------------------------------------------------
Best to all,
Bill P.