[From Rick Marken (920327 10:00)
Here is another try at sending this post. Hope it works.
···
--------
Greg Williams (920325-3) says:
Then let's start a new discussion about how organisms are organized
internally -- specifically, about whether certain actions are due to "chains"
or to "loops." By "chains," I mean that INTERNALLY there are no perceptual
signals going to comparators, resulting in error signals which influence the
actions' "trajectories" during those trajectories; generally, I would expect
"chains" to look like precalibrated "responses" to trigger-like "stimuli".
Such "chains" could be "output functions" of loops, and all I think would be
sacrificed in PCT is the idea that control is always continuous.
I think the first step in figuring out how organisms are organized internally
(I mean functional organization -- anatomical organization can be determined
by inspection) is to determine what kind of behavior that organization is
presumed to accomplish. That is why I think that a discussion of how
organisms are organized should begin with testing for controlled variables.
That is, it should start with data. I think we have precious little data
regarding the variables that organisms control. But we do have a lot of
what I would argue is misleading data which appears to show relation-
ships between various input variables and various output variables. If
we know that there are likely to be strong negative feedback relationships
in organismic behavior, then I think it behooves us to test to see whether
any behavioral variable is a component of such a loop or not. This is what
the test for the controlled variable does. If a behavior (such as the now
mythical moth fall) is, indeed, found to be uncontrolled, then a model
such as the one you describe above may be appropriate; and be able to
generate this particular behavior in a realistic environment.
But given the great misconceptinos that can result from ignoring the
POSSIBLE existence (I would say, ubiquitous existence) of strong
negative feedback in the relationship between organisms and their
environments, I would say that it would always be prudent to assume
that organismic outputs are part of a control loop -- and probably
controlled variables themselves. If "the test" reveals no evidence that
a variable is controlled or part of a control loop, then you can take
any observed relationship between this variable and other variables at
face value -- and start modelling the underlying mechanism to account
for this relationship. If, however, the variable is controlled or part
of a control loop then efforts to model the relationship must be informed
by an understanding of how the negative feedback loop affects the
appearance of these relationships.
So the first step in any attempt to build models of the INTERNAL
functions of an organism should begin with a very clear understanding
of the phenomenon to be modelled. Most models of the functional
INTERNALs of organisms are based on the assumption that the S-R
or input output relationships that are observed are just what they
seem -- input-output relationships. But because organisms are locked
in a negative feedback relationship with their environments, these relation-
ships may be (MAY BE -- but I think almost always ARE) the observed
responses to disturbances of controlled variables. If this is the case,
then an S-R type model of the INTERNALs would be wrong because it is
trying to explain a causal chain that doesn't happen to exist in that
case.
And so we come full circle (how approapriate) to "Marken's Law" --
PHENOMENA FIRST; models second. We need the data before we start
arguing about how to account for these data; obviously. The
contribution of PCT is to show that behavioral data must not
be taken at face value -- because the phenomenon you are dealing
with might be (MIGHT BE-- and, I have a strong hunch, almost always
is) the phenomenon of CONTROL.
Regards
Rick
**************************************************************
Richard S. Marken USMail: 10459 Holman Ave
The Aerospace Corporation Los Angeles, CA 90024
Internet:marken@aerospace.aero.org
(310) 336-6214 (day)
(310) 474-0313 (evening)