[A post from Mary Powers followed by a post from Bill Powers]
[from Mary Powers 960731]
Greg Wierzbicki (960730)
...Suppose that I suspect someone to be controlling for PCT?
How might I use the test to confirm/disconfirm my suspicion?
Suppose I suspected someone to be controlling for some
alternative set of principles to PCT. How might I
confirm/disconfirm my suspicion?
I think this is a good point. The Test for the controlled
quantity is to apply a disturbance to the presumed controlled
quantity and see what happens. If you resist, it is controlled.
It seems that if I were controlling for PCT, then I would
resist any/all attempts to persuade me otherwise. Likewise,
in the event I were controlling for NOT-PCT, then I just
might resist any/all attempts to enlighten me otherwise.
In either case, how do I know what theory someone else might
be controlling for, and then how do I conduct a conversation
which either helps them see the error of their ways, or come
more judiciously to my own enlightenment/epiphany? Any
thoughts?
I think we know very well what parts of PCT are resisted by
various people. The main one is that what is controlled is
perception rather than action. Some people resist PCT because it
is "mechanistic", i.e. suggests that people are machines in the
pre-control-theory sense. Some people resist it because their
definition of control is coercion. Some, because they didn't
think of it first. Some, because accepting it would mean their
life's work would end up in the trash. There are a lot of people
who have adopted only those parts of control theory that seem to
be compatible with or support their own theories.
I think that perhaps the only way to conduct the enlightening
conversation is to try to take it up a level. IF the PCTer and
the cognitive psychologist or behaviorist share the goal of
understanding behavior and the organization of living systems,
then whatever theory does that best will be the theory they'll
adopt. This does happen from time to time. The rarity of it,
however, indicates that the highest goal of most people in the
field is NOT the finding of the best possible explanation of
behavior and organization. It is the development and defense of
a belief system. This can be as true of PCTers as anyone else.
There is, I think, a problem with the word "theory". It can
mean, as you defined it, a set of principles. Or it can mean
a systems concept - the next level in the HPCT hierarchy - which
sets the reference signal levels for principles. I don't think
that the levels at this level are at all clear, but to me systems
concepts are beliefs, and have a big emotional component,
especially when disturbed. This doesn't detract from PCT or any
other theory - it just seems to be the way we are when whole-
heartedly involved in something - maybe it's what "whole-hearted"
_means_.
Anyway, I don't think you can get far with convincing someone
that PCT is better than what they currently believe in unless
they already have a stronger belief (higher loop gain?) in the
more general concept of the value of understanding.
* * *
Rick Marken (960726) and Bruce Gregory (960726)
Rick: I would say that in real situations it is VERY rare
that people are controlling for getting other people's
opinions about their observed behavior.
Bruce: I agree.
Hey, guys, are either of you watching the Olympic Games??
Mary P.
···
========================================================================
[From Bill Powers (960731.0730 MDT)]
Hans Blom, 960730 --
You, as an analyst external to the system, can do anything you
please. But the system itself doesn't necessarily work in the way you
find most convenient to use as a method of computation.
Bill, I keep running into statements like this from you. What gives
you the privileged information how "the system itself" works?
Nothing -- I didn't say I know how it works, only that it doesn't
NECESSARILY work in a way that mimics any mathematical methods of
computation.
We do not have access to "the system itself", as you well know.
In one sense we do have access. We are both (all) systems of the human
type. We experience the outcomes of our own operation, if not the
mechanisms that lead to such outcomes. Any valid model has to be
constrained by the requirement that it produce experiencable outcomes
like those we actually experience.
I presume that mathematical operations are a learned activity in a human
brain. But it doesn't seem likely (to me) that the processes which
enable us to symbolize and manipulate symbols according to rules are
themselves symbol manipulations done according to rules. Mathematics is
a _result_ of brain operations, not a _cause_ of them.
All we can "know" about the system is expressed in models, be they
words, formulas, diagrams or whatever. You keep reminding us that
we should compare how a model works with "reality", as if that were
possible.
Not compare models with reality, but compare them with other models that
we consider equally or more valid. One important "other model" is called
neurology. I know of no evidence that neurons are capable of performing
matrix inversions, except by the kinds of methods we learn in school --
by manipulating symbols according to rules, a very slow process
involving large parts of the brain. It seems to me that before we rely
too much on these operations in a theory, we need to understand how the
brain creates symbols, and how it learns and applies rules. While we can
describe simple neural operations mathematically, and speak of them as
"computations," they are not computations in the same sense as a
mathematical computation. They are physical processes which take place
without being mediated by symbols. That is my opinion.
-------------------------------------
Hans Blom, 960730c --
No more handwaving, no more words. Let's try to be as clear as
possible. Assume a perception C (for cause) and a perception E (for
effect). What is it to perceive a "relationship" between C and E?
Will it be another variable R? If so, computed how? Please no words
but something more rigid.
Suppose the relationship is "C --> E" (C implies E). If C is a neural
signal indicating the presence of one perception and E is a neural
signal indicating the presence of another perception, then a simple
logical circuit, describable as "R = not(C and not E)," will, when its
output R is true, indicate that a causal relation exists (by one way of
understanding causal relations). A control system can control for the
truth-value of this perception R. The perception can be made true if E
occurs, or if C and E occur, or if neither C nor E occurs. The only way
it can be false is for C to occur and E NOT to occur.
There are also analog relationships: E = f(C). Without artificial aids
we can detect simple ones, such as E increasing or decreasing in some
proportion to increases or decreases in C. In that example, where E =
kC, the perception of relationship would be a perception of k, the
proportionality itself. We can tell the difference between a brightness
control that yields a large change in brightess for a small angle of
turn, and one that requires a large angle of turn to produce an equal
change in brightess. We say that one control knob is more "sensitive"
than the other. This is a relationship perception in my way of
classifying perceptions.
---------------------------
To say that the change is "unaccountable" means that I experience a
difference between the actual brightness and the brightess I intend
to perceive, but not the cause of the difference.
Call the actual brightness P (for perception) and the intended
brightness R (reference). The difference between both is E (for
error). Now this error E is somehow "unaccountable". That could
mean that there is a reference value for E (let's call it RE) which
is not equal to E. This is again an error between a perception (E)
and its reference (RE). Where is the "unaccountability"?
I could have put my comment better, and will try to do so.
The accountability or unaccountability is not the concern of the control
system that controls brightess. Accounting for a change in brightess
would be, by my reckoning, the business of a system concerned with
accounting for things, not for controlling brightness. This means, first
of all, perceiving relationships. If someone else is seen to reach out
and turn the brightness control of the TV, and if immediately the
brightness on the screen rises or falls, it is likely that we will
perceive a causal relationship: turning the knob caused the brightness
to increase or decrease. We could be mistaken, of course, but we control
appearances, not actualities.
If the brightess changes without anyone turning the knob, we might
perceive it as caused, but the cause itself would not be perceived.
Hence we would not be able to account for the change in brightess by
pointing out the variable that caused it.
In HPCT we don't think of just one overall control process. Any
situation involves many control systems and control systems at different
levels, all operating at the same time.
If, in addition to perceiving this relationship, we have some reason to
want to keep the brightess at a particular level (not so bright as to
wash out the contrast, not so dim as to make the picture hard to see),
then we (meaning all the higher systems involved) will bring the body
into a configuration near the television set where we can turn on a
brightness control system which operates by turning the contrast knob to
control perceived brightness, adjusting the reference signal for this
system until the contrast is acceptable (to one higher control system)
and the picture is clear (to another higher control system).
Can the relationship RE = f (C) be called an expectation?
I don't immediately see how to intepret the statement that the reference
level for an effect is a function of a cause. I could see calling the
reference value itself an expectation, since it specifies a particular
level of a perception, E, which may or may not occur. I could see
calling an expected form of a relationship, f*, an expectation, if it
were a reference signal for a perceived relationship, f. The actually
perceived relationship might or might not match f*.
My impression is that "unaccountable" is close to or identical with
"unpredictable" or "unexpected".
Depends on how you define these terms. How do you define them?
-----------------------
I have sent you all the Artificial Cerebellum source code, .OBJ files,
and .BGI files as one self-extracting zipped file, uuencoded. Hope this
works. Remember to change the paths indicated in the GRUTILS unit for
the location of the .BGI files, and also (in the main programs)
references to the .OBJ files. I am sorry for the messy organization of
the programs -- I just wanted to get them working, and wasn't thinking
too much about making them understandable. You're seeing my dirty
laundry.
-----------------------------------------------------------------------
Best,
Bill P