[Avery Andrews 921215.1106]
There are alternatives to the `hard line proseletizing' approach to
presenting PCT on the one hand, and watering it down on the other.
It is for example quite obvious
that `cognitive scientists' do not on the whole understand feedback
mechanisms well enough to make an intelligent assessment of what
can actually be done with them. You can therefore simply present
yourself as someone who thinks that the potential of servomechanism
control has not been properly explpoited, and is trying to rectify
this. In academia it is perfectly respectable to push some idea hard
to see what can be gotten out of it - what's not allowed is to claim
that the idea is the solution to everything. & people are
hypersensitive such claims, since they are the hallmark of crackpottery,
and tend to perceive them even where they don't actually exist.
If servo control & its potential were properly understood, Arm-type
programs would be as thick on the ground as SHRDLU's and
unification-based parsers (like, for example, mine), & people long since
would have written programs where Arm-like components were used as
the executive modules of SHRDLU's, Chapman & Agre's game-players,
etc. (and surely Chapman would'nt have allowed Sonja to crash
into walls (`which looks really stupid') if had a real grasp of methods
which would have made it easy to keep her from doing it).
So the idea is just to present reasonably complicated PCT models of
specific phenomena, explain how they work, and let people draw their
own conclusions about the foundations of cognitive science. But,
of course, these models have to be `of' things that people are already
interested in, for some reason, no matter how silly (unfortunately,
tracking seems to be a non-topic, but that's just a fact that has to
be lived with for at least a while), and they have to actually do things
that people haven't done, thinking that they were too hard, and doing
these things has to look easy.
Therefore I see Arm (Penni & agree that `Little Man' has the wrong flavor to
make the right impression on our friends out there in Scienceland) as
a big step forward in presenting PCT principles in an effective way,
at least to the audience that I am thinking about
(even better if it could be integrated into a SHRDLU or Agre &
Horswill style Toast program, something I've been doing some thinking
about, & might try to do if I can get Arm running under Unix/X-windows).
If people find such a simulation interesting and `cool', they will be
more inclined to find out something about the principles whereby it
was constructed.
But I do think that it would be more effective if the accompanying
documentation made a clear distinction between
a) thinks that are well-supported by the neurophysiology (e.g.
the lowest level systems)
b) things that are suggested by the neurophysiology
c) pure speculation
with references provided for (a) and (b).
Avery.Andrews@anu.edu.au