What CT says about X

[From Bill Powers (920514.0600)]

An essay inspired by bits and pieces that have appeared on the net:

              What does control theory say about X?

Such questions almost always contain hidden propositions: X is a
phenomenon, and control theory should have something to say about it. But
to know what is being asked, you have to ask what theory underlies X. Most
questions about X beg some question, assume a theoretical stance that
existed long before control theory came along.

Control theory is:

not trait psychology ( characteristic --> behavior )
not top-down causality ( cognition --> behavior )
not bottom-up causality ( circumstances --> behavior )
not S-R causality ( events --> behavior )
not intervening variables ( A --> B --> behavior )
not categorical explanation ( things like A cause behaviors like B )
not associationism ( A cooccurs with B )

NOT ANYTHING THAT CAN BE LAID OUT ALONG A STRAIGHT LINE.

A Chomksyite evidently proposes "because I can perceive a certain
structure of relationships in language, that structure produces language."
The question is thus, "what does control theory have to say about the way
structure produces language?"

A Harrisite evidently proposes "because I can perceive sequences of
operators and arguments, operators and arguments cause language." The
question becomes "what does control theory have to say about the way the
occurrance of operators and arguments produces language?"

And others seem to propose "because children I perceive as bright may have
eczema more often than others, brightness causes eczema." The question:
"what does control theory have to say about the effect of brightness on
having eczema?"

The most common mistake in model-based explanation is to put an emergent
phenomenon into the model as a box defined to produce that phenomenon. In
diagrams of control systems one often finds a box labeled "controller." But
control is not something that any one component of a control system can do:
the "controller" box by itself controls nothing; it simply transforms an
input into an output. Control is what emerges when input-output components
are interconnected in a certain way and allowed to interact with an
environment. The term "control" should not appear anywhere in a model of a
control system.

If it seems that certain words act as operators in relation to certain
other words that act as arguments, then a model that explains this
phenomenon should contain neither "operators" nor "arguments." To say that
operators "take" arguments is to put a box into the model that produces the
emergent phenomenon.

If it seems that there is structure in language, then a model that explains
this phenomenon should not contain that structure, but only components that
lead to phenomena which can be seen as having that structure. The apparent
structure should emerge from the underlying processes. Even to say that a
word can "modify" a phrase is to make the modification process part of the
model instead of emergent from it.

If it seems that bright people have eczema more than others do, then a
model that explains this phenomenon should contain neither "brightness" nor
"eczema". It should have an organization from which emerge phenomena that
can be seen as brightness and as eczema.

Every person comes into control theory from some other point of view. As a
result, every person has beliefs about the questions that must be answered
in order to make progress toward understanding some facet of human nature.
But all accepted points of view from the past lead to questions that imply
an answer that can be laid out along a straight line. Control theory can't
answer such questions (or validate the theory that led to them) because
control theory proposes that behavior is NOT produced by processes that can
be laid out along a straight line -- processes in which the outcome can be
considered separately from its antecedents.

The most important question to ask is not what control theory has to say
about X, but what OTHER theory is behind the very question. It is likely to
be one of the kinds of theories listed above. Before you ask such
questions, it would be profitable to ask about the assumptions behind it.
That may tell you the answer without your having to ask the question.

Best

Bill P.

[Avery Andrews 920515.1432]
(Bill Powers (920514.0600)

A Chomksyite evidently proposes "because I can perceive a certain
structure of relationships in language, that structure produces language."
The question is thus, "what does control theory have to say about the way
structure produces language?"

I wouldn't really say this, but that something is producing the structure,
and knowing a reasonable amount about the structure ought to help in
identifying the something. Actually, in the case of grammatical
generations, I suspect that there really are things corresponding to them
(ie., for the generalizations that lead people to postulate noun-phrases,
there is a noun-phrase detector). I suspect this because grammar does
not seem to be real-world interactive in the manner that most things
studied in psychology are, so the structures in grammar can't be
coming from simple interactions with a complex environment. But that
could all be wrong. The real purpose of getting an organized view
of grammatical regularities is to get constraints on possible theories
of the mechanisms that are actually involved.

  Avery.Andrews@anu.edu.au
   (currently andrews@csli.stanford.edu)