Ken Hacker's !s and ?s

[From Bill Powers (930329.1000 MST)]

Ken Hacker (930328) --

A reasonable set of comments and questions. I'll try to deal with
some of them.

Your list of our criticisms of conventional behavior science is
generally on the mark, but without any reasons being supplied
they sound more extreme than they are.

Consider:

    * All social science is wrong.

    * All behavioral science is wrong.

    * All social scientists and behavioral scientists
             believe what is wrong.

   * All social scientists and behavioral scientists do not
             know what they are doing with research.

In the first place, I always try to put somewhere in the
discussion the proviso "if the PCT model actually explains
behavior...", which make this somewhat less arbitrary and ad
hominum.

In the second place, it makes a difference just what claim of
wrongness is made. Wrong about what? The fundamental disparity
between PCT and conventional theories of behavior is the
treatment of cause and effect. Under PCT, the actions of an
organism are not caused by stimuli with the organism simply
mediating between cause and effect. In the PCT model, but not in
any conventional model, there is a controlled variable between
the apparent cause and its apparent effect as usually observed.
The behavior is directed toward control of the controlled
variable; it is not simply a response to the stimulus or a
dependent variable being determined by an independent variable.
As a result, all theories in the conventional sciences that rely
on some cause-effect or IV-DV paradigm, including methods of data
analysis that assume such a relationship, are fundamentally
contradicted by PCT. It isn't necessary to investigate every
detail of any theory that is proposed; all that is needed is to
see whether the internal validity of the theory depends on
assuming that behavior is a dependent variable and inputs to the
organism are independent variables. If that is true, then no more
need be known: the theory is contradicted by PCT and in terms of
PCT the conclusions are false.

I leave it to you to decide how much of the social and behavioral
sciences would remain intact if it were to prove true that all
behavior is organized around controlled variables, and none is
open-loop.

  * Only PCT has something useful to say about how humans
             regulate their behaviors.

As stated, this is contradictory to PCT on the face of it. Human
beings, under PCT, do not regulate their behaviors at all, as
least as behavior is conventionally understood. PCT says nothing
at all about the regulation of behavior; it is concerned with the
regulation of perceptions, of inputs. Behavior, or action,
becomes what it must become to prevent disturbances from having
an important effect on the controlled inputs. So it is true that
only PCT has something useful to say about the regulation of
input, if that is what is meant by behavior. But that is not what
conventional sciences have meant by behavior.

         * Communication theory is wrong.

Communication theory, as a mathematical treatment of certain ways
of representing variables and signals, is completely right in
terms of its mathematical operations. Where I have difficulties
with it is in the initial definitions and assumptions, which do
not seem to me to have any necessity to them. But that's me, not
PCT.

         * PCT equations (confirmed by PCT theorists) PROVE
             that PCT is correct.

Fortunately, confirmation of the PCT equations by comparison with
experiment is done by public methods easily replicable by anyone,
with high reliability. No faith or special knowledge is required.
The validity of these equations is put at risk every time they
are used to fit a model to real behavior. The equations don't
prove that PCT is correct; as far as they go, they ARE PCT.

  * PCT is real science. Other approaches to human behavior
             are pseudo-scientific or "half-assed."

PCT is real science because it risks everything with every
application to data. It isn't simply assumed to be correct and
twisted to make the data seem to fit it. It can't live with
serious exceptions or counterexamples. Its premises are
themselves testable through experiment, and are tested every time
they are used.

I would accept as real science any other approach to behavior
that had the same characteristics. We're not talking about
control theory here: we're discussing what science is about.

         * People who challenge PCT are misguided, ignorant,
             and not yet fully developed intellectually.

Depends on how they go about challenging it. I see no shame in
being any of the things you mention -- is there anyone who
doesn't fit the description? I object to only one kind of
challenge to PCT: the kind that is made without knowing what PCT
is about, and is based only on a difference in conclusions.

···

----------------------------------------------
The questions:

         a. The scope of PCT. Is it limited to describing and
             explaining physical behaviors of human beings such
             as motor actions?

In that all visible behaviors are physical and entail motor
actions, including debating about quantum physics and women's
rights as well as tracking behavior, yes. One could say, given a
few precautions, that there are "really" only four kinds of
behavior: push, pull, twist, and squeeze. All the rest is
controlled perception, consequences of behavior. And even those
four are controlled perceptions. PCT can be applied to behavior
at any level of organization you please. HPCT is meant to apply
at all levels at once.

         b. If there is no information inside of input, what
             prevents one from asserting behavioristic-like
             internal responses to internal stimuli or a kind
             of reverse behaviorism?

Nothing. You can propose that behavior is caused by the phases of
the moon if you like, but you'll have to defend the proposal. See
my earlier post to Rick on this subject. I'm not going to
generalize about information and input any more until I'm sure
what's being claimed or asserted.

         c. What happens to Ashby's Law of Requisite Variety?

I don't use it, although it's implicit in control theory. Variety
is defined by Ashby as follows:

"The word _variety_, in relation to a set of distinguishable
elements, will be used to mean either (1) the number of distinct
elements, or (2) the logarithm to the base 2 of the number, the
context indicating the sense used." (p.126 of _introduction to
cybernetics_).

The Law of Requisite Variety states that the output of a
regulator must have at least as much variety as the disturbance,
if the result is to be regulation of some variable. That is to
say, the output must have at least as many discriminable states
as the disturbance has.

In terms of control theory, we would say that if a controlled
variable is to be maintained exactly at a reference level, the
output of the system must be at all times quantitatively equal
and opposite to the magnitude of the disturbing variable, both
measured in terms of effect on the controlled variable.

Note that the control-theoretic statement goes much farther than
the LRV goes. It says that not only must the output have AS MANY
discriminable states as the disturbance, but that these states
must have the correct quantitative magnitudes, and they must
occur in pairs: one specific output state for each state of the
disturbing variable. The output and disturbing states must be
quantitatively equal, and of opposite signs. So the LRV, while it
states a weak necessary condition, by no means states a
sufficient condition for control or regulation to exist.

         d. How does PCT account for stored knowledge such as
             schemata (or whatever other term you choose)?

PCT itself doesn't; that's a question of fact for neurophysiology
or neurochemistry. I have proposed some possible relationships of
stored information to the operation of control systems in the
hierarchical model. As Greg Williams puts it, those proposals are
embellishments on the basic model.

"Schemata", as I have heard the term used, could have various
meanings, corresponding to reference signals (when they relate to
goals), to perceptual input functions (when they determine how
lower-level perceptions are interpreted) or memories (in various
circumstances). "Knowledge" has similar usages. Because these
words are used in relation to functionally very different aspects
of a brain model, it's hard to pin them down to any one meaning.
The problem is somewhat similar to the uses of "want." If you say
"I want some ice cream" you could mean to point to a reference
signal, the definition of what it is that you want, or you could
mean to indicate an error signal, emphasizing the fact that you
don't have the ice cream. In my opinion, terms like these are too
loose to be used in a model, although we may use them informally
when context supplies the missing discriminations.

         d. What makes PCT more than intensive descriptions and
             explanations of neural pathways and signals?

The fact that it makes almost no attempt to describe neural
pathways and signals. The primary uses of PCT are in modeling
externally-visible behavior in real environments, showing that
relationships expected under the hypothesis of control do in fact
occur and can be predicted. The models propose that certain
functions are carried out inside the control system: perception,
comparison, and conversion of error to output. Nothing is said
about what neural circuits are involved in implementing these
functions, although in some cases we know quite a bit about
specific neural signals and pathways involved in specific control
systems.

     e. Why must PCT be understood as some sort of revolution
        as opposed to some sort of new and productive thinking?

Because PCT directly contradicts most traditional conceptions of
how behavior itself works. It also predicts phenomena of a kind
that conventional sciences have known nothing about, or that
under conventional assumptions go in a direction opposite to the
direction that control theory would predict and that experiment
supports. For example, under control theory, the prediction (and
the observation) is that doubling the sensitivity of a perceptual
function to stimuli will result in halving the input to the
perceptual function, not doubling its output signal.

Control theory is to conventional theory as Newtonian celestial
mechanics was to Ptolemaic epicycles. Many of the observations
may be the same, but the explanation is radically different.

f. What are essential differences between PCT and cybernetics?

Cybernetics abandoned control theory at about the time of the 5th
Macy conference, when most of the main misconceptions about
control systems were laid in.

      g. What do axioms or propositions from PCT contribute to
          everyday human adapting, living, changing, improving,
          that other forms of analysis do not? Where is the
          proof of human behavioral successes with PCT, as
          opposed to success in the abstract?

I'd say you should ask people who are applying it. Try Ed Ford,
Dag Forssell, and David Goldstein. Or ask people in the CSG who
have been interacting with each other for 9 years with an
understanding of PCT. In a lot of cases, PCT vindicates common-
sense ideas that science has rejected, such as the inportance of
goal-setting, perception, and intention. If you see concepts like
that creeping back into polite scientific society, I think you
can credit PCT with having inspired at least some of those
changes in thinking. Even when an idea is rejected, something of
it sticks.

         h. Most scientists are concerned with not only
             describing and explaining phenomena, but also wish
             to predict and sometimes "control" them. Why does
             PCT deny the desirability of prediction?

PCT does not deny the desirability of prediction. I, personally,
deny the desirability of lousy predictions, particularly when
they're used as if they were good predictions.

          Is prediction not always part of human anticipation,
          whether in ordinary life or in science?

Yes, I think it is. It's explained, more or less, in HPCT as a
phenomenon of imagination, which is part of the story of mental
modeling. People do it, so it belongs in the model. But I can't
say a lot, theoretically, about HOW they do it. Most of my
objections to prediction involve pointing out that it's not being
done very well, and shouldn't be relied on as a method of control
in most situations -- not if good control is important.

       i. Where do reference levels or signal originate? In
             other words, if my control system has error
             signals, what constitutes the sources of the error
             signals?

I guess you haven't read anything about the hierarchical model.
When you say "my control system" I wonder if you have read
anything about the model at all. People have hundreds, thousands
of control systems, all active at the same time and at different
levels. Higher-level systems act not by producing motor outputs
but by adjusting reference signals for lower systems. If you
haven't been aware of this aspect of the model, you must have
found a lot of the conversations on the net pretty confusing.
Next you'll ask "But what about the highest levels of control
system? Where do THEY get their reference signals?" And I will
answer, your guess is as good as mine, but asking the question
shows that you get the picture.

            What is stored that makes me think that I do not
             like or wish to accept certain input?

You're assuming that it's something stored that "makes you think"
that you don't like etc. A reference signal specifies a certain
amount of a given perception, anywhere between the maximum
possible and zero. That becomes the amount that you prefer; when
you say you prefer that state of that perception, control theory
explains this by saying that there is a reference signal in some
control system set to that value, probably as part of controlling
for some higher-level perception. This may or may not involve
"storage" of something. There are lots of proposals on this
subject in BCP.

          How conscious are the processes of matching input to
          the reference signals?

It depends. The same control system can operate consciously or
automatically, at any level. A perceptual signal can occur with
or without consciousness of its presence. The nature of
consciousness is not explained in PCT or HPCT (or as far as I
know, in any T).

         As PCT should not be criticized by me or anyone else
         for not explaining social behavior, social sciences
         which do focus on social behaviors should not be
         excoriated for not explaining individual human control
         mechanisms and processes.

Clark McPhail, Chuck Tucker, and Kent McClelland (the CSG's
sociologists) and Tom Bourbon (who has modeled simple social
interactions) ought to comment on this. My view is that PCT, by
explaining the interactions of individuals with their
surroundings, lays the groundwork for explaining what happens
when groups of independent control systems interact with other
control systems. Social "laws" emerge from the properties of
interacting individuals and the shared environment. While
naturalistic observation is needed to determine the existence of
regularities in social behavior, PCT, I claim, is needed to
explain these regularities.

   a. Communication studies, unlike traditional psychology
and maybe other behavior/social sciences, no longer assumes
that people are simply affected by stimuli in varying
conditions and ways. Since the 1970s, my discipline has said
that each human TAKES from mass media, from interpersonal
interactions, from printed words, from any messages, and
recodes what is decoded. We threw out Shannon and Weaver's
model in the 1970s as anything useful to explain human
communication. Both Aristotelian and electrical engineering
models of message sending and receiving say little of
importance to understanding the complexities of human
communication, mainly because people do not simply take in
messages. Nor do messages do contain meaning.

Good. Give my regards to Klaus. And what is the reason for all
these changes in thought? Is there any theory from which you
could have deduced all these new ideas? Or are they simply ideas
that were proposed, and that others found acceptable for unstated
reasons? Everything in this paragraph could be deduced from PCT,
and has been familiar to PCTers for many years (except for the
remaining buzzwords like "coding" and "recoding"). You're talking
about observable phenomena, and that's what PCT is for: to bring
observable phenomena into a single common framework and make some
sort of sense of them -- not just to accept that they happen, but
to explain, in terms of a unified structure of theory, why they
must happen. The same theory that explains tracking behavior
explains why people give their own meanings to their inputs, and
why that is the only way it can happen.

Hope you had fun with the grading....
--------------------------------------------------------------
Best,

Bill P.