Apps Assumptions, Model-based control

[From Rick Marken (950831.1330)]

Avery Andrews (950831.1930) --

There's at least one important feature that all the PCT apps that have been
discussed have in common, which I don't think has gotten quite as much
attention as it deserves. This is that they presuppose that people's higher
level goals are basically the same

I think this is only the case if everyone is controlling for the same higher
level perceptual variables. For example, if everyone is controlling for a
perception that is the same function, f() of the same lower level variables,
s1,s2,...sN (p = f(s1,s2...sN)) then there will be conflict if everyone has a
different goal for that perception (a different reference setting, p* , for
p). Conflict can only be avoided if everyone has the same goal for the
perception.

I think it is highly unlikely that groups of people actually end up
controlling precisely (or even approximately) the same high level perceptual
variables. So the chances of conflict due to people in the group adopting
different high level goals is, I think, extremely small.

I think the far more common cause of conflict between people in a group is
the control of perceptions that are based, at least in part, on perceptions
of the behavior of other people. PCT apps should be aimed, in part, at
showing how to reduce the conflicts that results from controlling such
perceptions.

Hans Blom (950831) --

And are you sure that it was an ATTITUDE that nearly killed?

The attitude I was referring to was reflected in your statement:

I haven't discovered internal inconsistencies (conflicts) in this model.

This is a non-scientific attitude toward evaluating the merits of an
explanation (model). The attitude is: if the explanation makes sense (is
internally consistent) then it is a good explanation. You consider your
model-based control system a good model of blood pressure control because it
is internally consistent. That can lead to problems.

And, pretty please, tell me how I can get to know "external reality",
so that I can verify that my theory is consistent with it.

You do experiments and see if the behavior of the model matches the behavior
observed in the experiment. If the behavior of the model continues to match
the behavior observed in many experiemnts we consider the model our
(currently) best estimate of what constitutes "external reality".

Bill Powers (950831.0730 MDT) --

Where did you get the idea that Hans' system nearly killed a patient?

From Hans' description of the event on the net.

Did you get a copy of Hans' paper on this control system?

No.

Your own comments support the idea of model-based control at the higher
levels of organization. You form a picture of some person you have never
met called Hans Blom...

My comments might support the idea of model-based control but it is not the
kind of model-based control that occurs in Hans' model (as you know). The
model in Hans' model-based controller is a model of the function that
connects the outputs of a controller to the results produced by that
controller; it is also a model of the predictable effects of other variables
(distrubances) on the results produced by the controller. The kind of model
based control you are talking about is control of imagined (modelled)
perception; Hans' model is not a model-based controller in this sense.

This is how we invent heros and villains...

we do not see what is before our eyes; we see our mental models.

I don't agree with many of Hans ideas about control but I certainly don't
have a mental model of Hans as a villain. He seems like a pretty nice guy to
me (though I think he should find a nice girl and settle down already;-)) .

I was not accusing Hans of attempted man-slaughter, by the way. Hans had said
that he evaluates a model in terms of internal consistency. I was saying
that this approach (attitude) to evaluating models was not scientific; if
internal consistency really is the main reason why the model-based
architecture was selected for the blood pressure controller, then problems
are to be expected - - and a problem of the kind that might be expected
did, indeed, arise.

I didn't know -- but I assumed -- that the blood pressure controller had been
carefully tested before use. I have no doubt that Hans was extremely
responsible and cautious in the use of the device. I did not mean to imply
that the problems that did arise were the result of negligence. But I do
think that these problems were, in part, the result of sticking with a
perfectly plausible control architecture (model based) for the wrong reasons
(internal consistency).

As I recall, the model-based blood pressure controller is essentially an open
loop system that administers a drug on the basis of an algorithm that
continuously converts measures of blood pressure into drug dosage; if the
system controls anything it is an output (drug dosage); not an input
(measured blood pressure). The architecture of this control system was
apparently selected on the basis of a conviction that control is best
achieved by a model- based organization. This conviction, in turn, was based
on Hans' evaluation of the internal consistency of the model-based control
organization. This, I believe, is where the problem lies. The particular
(human-corrected) failure of the blood presssure controller that Hans
described, though the result of unforseen and unlikely events, is the kind of
failure that is attributable to the choice of the model-based "control"
organization.

Best

Rick