[From Bruce Gregory 92001.0403.1435)]
Bryan Thalhammer (2001.0403.1246 CDT)
I don't think that the responses are a whole sale "dismissal." What I see
happening here is a collision of philosophical/"big picture" descriptions,
such as this "pattern discovery machine" and another analysis that PCTers
tend to use in analytical/"discrete quantitative" descriptions, or
modeling. I won't go into whether one is better than the other, that is
not my point. Rather, a "pattern discovery machine" is a more poetic,
evocative way of trying to perceive the brain.
Not the way I use the expression. PCTers don't seem to be very interested
in modeling learning. If they were, they would soon discover the need to
use neural networks. Neural networks are not evocative or poetic, at least
as far as I am concerned. They are just as rigorous and quantitative as PCT
models.
But that doesn't go far
enough for those who would prefer to discuss these phenomena at far more
discrete levels. Likewise, Rick's request for "a functional diagram of the
'pattern discovery' process" throws a monkey wrench into David's poetic or
"big picture" attempt to visualize the brain and wonder at it.
Again, I must disagree. I can point you to many neural network models for
pattern discovery. In fact, entire textbooks are devoted to them (cf.
Christopher Bishop _Neural Networks for Pattern Recognition_, Oxford
University Press 1995). Patter recognition is a well-developed domain of
computer science.
Once you identify an example of control, you can build a hierarchical
control model to represent the process, but I am aware of no control models
that addresses even the most obvious pattern-recognition tasks that we
perform effortlessly every day. How do you recognize your spouse? Exactly
what perception(s) are you controlling and where do the reference levels
come from? The PCT literature doesn't seem to have much to say about the
process. Why denigrate disciplines that do? Neural networks learn.
Hierarchical control systems control. Big difference. Conflict only arises
if you are determined to maintain that either model can be eliminated in
favor of the other.
It is what Runkel calls "strife." And Gregory Bateson had a term for this
too, "complementary schismogenesis," a kind of positive feedback loop, or
arms race, where each continues to increase the ante until the conversation
breaks down into a pitched battle. Happens often here, it seems. You
should try to find a bridging vocabulary you both can use, which is in
itself a kind of negative feedback loop. You won't be able to agree, but
you may be able to relax some terminology in a compromising strategy, and
arrive at a commmon talking point.
I don't see the problem. When Rick produces a spreadsheet HPCT model that
performs pattern recognition I'll be the first to congratulate him. Perhaps
he's working on the problem right now. An HPCT model of speech recognition
would make a major impact (to say nothing of a great deal of money for the
developer).
BG