[From Rick Marken (960220.1100)]
Shannon Williams (960219.11:30 CST) --
Until you can model reorganization, you cannot say that a thinking creature
uses references the same way that an insect does.
Why not?
Hans Blom (960220) --
It starts to look to me as if there is no black-and-white difference between
a control process and an equilibrium process if the equilibrium takes place
in a strong gravitation-like field.
There is no black and white _indication_ of control. There _is_ a black and
white difference between a control system and a cause-effect system; the
problem is determining, based on evidence that is not black and white,
whether a variable is under control by a control system or not. It's very
much like a signal detection problem where we have to determine whether an
observation was the result of "signal" (control system) or "noise" (a cause-
effect system). One type of observation we make is of the stability of a
variable:
S = 1-sqrt(V.e/V.o)
where S is the stability measure, V.e is the expected and V.o the observed
variance of the putative controlled variable. This measure of control can
range from -infinity (perfect control) to zero (no control) to +infinity
(perfect positive feedback). Any value of S is possible whether the variable
is being controlled by a control system or caused by a cause-effect
(equilibrium) system. But values of S near 0 are far more probable if the
variable is being caused by a cause-effect system; values of S less than,
say, - 10 are more probable if the variable is being controlled by a control
system. So when we do the Test we observe S and decide, based on the value of
S, whether the variable is being controlled by a control system or not. It's
a statistical inference, of course, but the chances of a Hit (concluding
that the variable is controlld when it is) are VERY good and the chances
of a Type I error (concluding that a variable is controlled when it's not)
are astronomically small.
I tend to think now that there is no basic difference [between equilibrium
and control], but the issue is far from resolved for me.
If, after 4 plus years on CSGNet this issue is still far from resolved for
you, I think it's quite unlikely that it will ever be resolved for you.
Me:
Now you should have a much better idea of how to Test for controlled
variables.
Hans:
Actually, I don't. At least not a Test that allows me to discriminate
between an equilibrium system and a control system. How can I test for
the controlled variable if I have no idea what it is?
You do the Test when you have an idea what the controlled variable _might_
be. The variable that _might_ be under control is the one that appears to
vary less than expected when disturbed.
Bill Powers (960215.1200 MST) --
In lineal causality, we have
(1) effect = f(cause)
In circular casuality, we have
(2) effect = f(cause,effect)
Hans:
One of the first things that control engineering students learn is how to
reduce expressions of form (2) into expressions of form (1).
They learn to analyze a closed loop system as though it were open loop?
Why?
So to say that there is a "fundamental difference" between the two "types
of causality" will not be appreciated by many control engineers.
But do _you_ appreciate it?
(My guess: no)
Best
Rick