[Avery.Andrews 920123.1601]
Some thoughts on various issues:
First, the `feedback too slow' issue. While many think that
feedback is too slow to be involved in many activities, just about
everybody agrees that many fast reflexes are modulated by sensory
information of various kinds. But feedback is just modulation
of the intensity of the response w.r.t. the extent to which the
needed result already exists. So these two popular positions are
contractory, as far as the *initiation* of responses is concerned.
And, of course, feedback inhibition can be done with a small
number of synapses, so the time it adds would typically be minimal.
For termination of responses, things might be different. What is required
is that the force delivered by the actuators be approximately zero
when the required result is achieved. It seems plausible to me that
simple linear feedback schemes might indeed be too slow for certain
kinds of behaviors, but then I'm not an engineer, & am perfectly happy
to be shown wrong on this by people who are. Supposing it is, one
can certainly envision that nonlinear feedback schemes might overcome
the problem. E.g. the commands sent to the actuators specify zero
force before the desired result is achieved. Much of the effort in
learning highly skilled activities might plausibly go into tuning
the nonlinear widgets so that they will work right.
At any rate, initiation & termination are different issues, & maybe
nonlinearity is more important than PCT writings tend to suggest.
And, thinking about this issues makes me vividly aware of how little
the typical cognitivist (e.g., myself) knows about actuators, and
what it takes to drive fingers, fins, legs, lips, etc. I wonder
how many psychologists are better off than I am in this respect.
Second: Central Pattern Generators and Planning.
There appears to be widespread confusion between CPGs and Planning, at
least at the level of terminology. A plan is something that you write
out in advance, and then follow. E.g. generate a constituent structure,
and then speak it (how Penni says that speech production doesn't work).
But all a CPG has to do is spit out a waveform. In the general case,
I guess, a CPG is just a nonlinear transducer, perhaps emitting a
sinewave of variable amplituded & frequency, as determined by two
input wires. So the slowing circuits in Arm are simple trajectory
generators, converting a step change in their inputs into a ramp
in their outputs. Evidence for CPGs is thus not evidence against
PCT.
This feeds into the deafferentation issue. In a post of Bill Powers'
some time ago (csg-l (921025.0800)), he pointed out that if the feedback
return is cut, the reference signals will convert directly to error
signals, creating exaggerated movements (since the feedback return is
inhibitory), which is apparently what you get right after
deafferentation (it would be
good to have the references for this). But the nonlinear widgets
will still basically be doing the right sort of thing, even if they
have to be extensively retuned to deliver effective behavior. So the
centrally generated patterns will basically be within the `competence'
of the organism, even if it takes a bit of time for it to be able to
implement them adequately in `performance'.
So it may well be that PCT actually does a *better* job of explaining
the consequences of deafferentation, tho I think there's a fair amount
of substantiatory scholarship to be done on this issue.
Avery.Andrews@anu.edu.auti