Tar Baby

[From Rick Marken (931118.1530)]

Martin Taylor (931118 12:00) --

The information
contributed to the environment by the output balances the information
contributed to the control system by the disturbance.

OK. But that just seems to be another (and less informative) way of

o ~ r - 1/g (d)

where o is the output, r is the reference, g is the feedback function
and d is the disturbance. What is gained by talking about information?

And, once again, notice that perception (p) does NOT appear in this
equation. I realize that this equation is only a "quasi-static"
approximation (~ rather than =) but it shows that the main deter-
minant of output is the disturbance, not perception. The only other
determinant of output is the reference signal, r. Perception (p) does
not enter the equation even if you decrease the loop gain (thus making
the approximation even poorer). In a control loop, output (the
actual influence on the controlled variable) is not a function of
perceptual input (p); nor is it a function of error (r-p). When the
reference signal is constant, variations in the output of a control
system (the effect of the system on the controlled variable) depend
only on a variations in a variable (or variables) that the system
cannot percieve -- the disturbance.

This is the invisible bridge that Tom Bourbon asks you to step onto.
This is REALLY the way a living control system works. The system
controls input (perception); it is not driven, informed or guided
by it. I think you find this hard to believe and are justifiably
reluctant to step out onto what appears to be an invisible bridge
over an apparently infinite abyss (I think the term "magic" was
bandied about last time). Your position seems to be that some
information about d must be "leaking" through with the perception (p),
providing the basis on which the the system determines the outputs
that correctly counter disturbances. But this is not true. Control
doesn't work this way -- at all. That's the PCT revolution.

Why does this apparenly harmless belief in "information leaks" matter?
After all, it seems like a pretty theoretical issue -- one of those
"why" questions that could only be of concern to mathematicians and
control engineers. I think it does matter, however, because (as I've
said a billion times before in different ways) this "belief" in
"information leaks" points investigations of control behavior (if
one investigates control behavior AT ALL) in the wrong direction.
It is the direction taken in conventional psychology where the search
is still on for the environmental correlates of the information that
guides outputs. I think the belief in "information leaks" might incline
one to look in the environment for things that VARY in order to explain
the "information leak" -based variations observed in the outputs of the
organism. We end up back at input-output research instead of testing for
controlled variables.

Of course, if the belief in "information leaks" doesn't divert one
from testing for controlled variables, then my argument against it is
irrelevant. My main concern is to get people out there doing PCT based
research -- identifying controlled variables like crazy. If people do
the research properly, then I don't care what they believe in -- even
if they want to believe that there's a whole bunch of information
about the disturbance in perception.