[From Rick Marken (970221.1240)]
Bill Powers (970221.0715 MST) --
human being knows before the selection process begins what the
desired outcome is
Peter D. Junger (97022110.5611 EST) --
But this distinction would seem only to apply when the selection
is done by a ``breeder''. If someone were to say, ``that black
kitten is cute, so I'll keep it and you can kill the rest'', that
person would not be selecting (not have as a goal) increasing the
number of black cats in the world, even though if enough people
acted that way that would certainly be the result.
But the person _would_ be selecting for having a cute cat, no? The
person knows, before the selection process, what the desired outcome
is (having one live cute cat and a bunch of dead ugly cats) and acts
to produce that outcome.
Hans Blom (970221c) --
Anticipation is not control. but it can be _used by_ control...An
example from my blood pressure controller, whose core is very much
like a standard PCT controller...Thus the basic controller is
overridden and the flow is shut off immediately on discovery of
large blood pressure drops. In other words, in this case feedback
is simply too slow and "anticipatory" action is required.
It sounds to me like what you have here are two controllers, one trying
to keep blood pressure at a fixed reference and one trying to keep the
rate of decrease in blood pressure _less than_ some value (a decrease
greater than that value indicates that the person is going into shock).
These two controllers are controlling two different perceptual
variables: blood pressure and change in blood pressure. There is also an
implied hierarchical relationship between the controllers since the
blood pressure controller will be "overridden" when the pressure change
controller goes into action. There is nothing here that I would call
"anticipation" or "model based control" for that matter. Where are
do you think these things happening in this system of yours, Hans?
Me:
The essential problem with this argument (as you know) is that
it is based on the idea that events in a control loop occur
sequentially, one after the other.
Martin Taylor (970221 12:00) --
Yes, and I'm almost certain Hans is not at all thinking like that.
Well, here's a chance for you to try looking at actual data for a
change. Try reading [Hans Blom, 970221d] and see if you are still
almost certain that Hans doesn't believe that events in a control
loop occur sequentially.
You presented some data about tracking a sinusoidal disturbance, and
pointed out that the fit...was better if the reference varied
sinusoidally with a phase advance when compared with the disturbance
... What it has to do with feedback being too slow is that you used
what amounted to a predictive model of the disturbance to improve
fit...It's a good demonstration of the value of an internal model
even in the lowest-level type of tracking control that is taken to
exemplify all control in most of these discussions.
OK. I do recall this experiment somewhat, but not the details. I think
I was trying to figure out a way to look at two levels of control
simultaneously. A subject tracked a sinusoidal target; the frequency of
the target was varied. I think I built some models to compare the
tracking accuracy of a single controller (controlling deviation from
the cursor relative to a fixed reference of zero) vs a two level
controller (the higher level system sending a sinusoidal reference to
a the lower order system that was controlling deviation of cursor
perception from varying reference). The two level model did better
than the one level model when the frequency of the track was too
high for the one level model (as long as the reference was an
accurate representation of the target movement).
This experiment is a good example of what is wrong with this whole
discussion of model-based control. What's wrong is that people use the
term "model based control" to mean whatever the hell they want it to
mean when they are arguing about whether "model based control" is
involved in behavior. But Hans has described a specific model (which
controls a perception that is generated by a model of the environmental
feedback function -- the model being continuously updated based on the
actual perceptual result of action) that he called a "model-based"
control The "model" in Hans' "model-based" control system is part
of the connection between the reference signal and output variable
in a SINGLE CONTROL LOOP.
Now you (Martin) are calling the two level pursuit tracking model a
model based control model. But that model is nothing like Hans' model
based control model. It is simply a hierarchical control model. The
sinusoidal reference input to the lower level system is selected
(by me, the modeller) as a "model" of the target movement. In a real
two level model this reference variation would have to be derived from
the difference between the perceptual and reference input to the higher
level system. But the main point is that the "modeling" done by this
two level system is not an aspect of the operation of ANY SINGLE
CONTROL SYSTEM that is a part of the hierarchy. The control systems
themselves are simply perceptual control system.
So, perhaps I can clarify my objection to "model based control"
models by saying that there is no evidence that any of the _individual_
control systems that make up a living organism operate on the basis of
Hans' (or the MCT) version of model - based control.
Bruce Abbott (970221.1355 EST)
As I noted some time ago, a similar notion about the effect of stress
on mutation rate occurred to me independently some years ago when I
first heard about the problem of "punctuated equilibrium." I saw it
as a negative feedback process in which stress (life-critical
variables being pushed into critical regions) led to something of a
breakdown in the machinery of replication
What's really amazing is that, having independently discovered the
principles of feedback control and their relevance to behavior so
long ago, you didn't (and still don't) include anything in your methods
textbook about testing for controlled variables;-)
Best
Rick