[Avery.Andrews 930510]
(Rick Marken 930509.100)
Kinda a lot for the final section of one paper - I think many of these
points could easily fill up a substantial paper all by themselves.
I think it's better do deal with a few topics in substantial depth
than to touch on many briefly. But for some of them, there's a fair
amount of helpful current literature. E.g., in the case of the
`feedback too slow' objection, anyone who says that without careful
qualification is at least five years behind in their reading (trolling
my `feedback too slow' postings should turn up some useful references.
I think the topic of `feedforward' could use some careful thought too.
It certainly has a role to play, but I think is useful under rather
different circumstances than feedback. In riding a bike for example,
it is useful to get your bum off the seat before you hit a bump.
This is clearly a sort of `feedforward', and the reason it works is
because the temporal and quantitative aspects of the required act
aren't very tightly constrained: you want to have your bum off the
seat before the bike hits the bump, but it doesn't matter how far
off, or how much before. It seems to me that it would be useful
to get this kind of `anticipation' (a better term than feedforward,
I think), integrated into the model. E.g. the actual rise is presumably
specified as a change in the reference levels for some perceptions,
but how does this change get triggered by the sight of the upcoming
bump, and how is it integraged with the other aspects of bike-riding.
As far as attractors go, my understanding is that a feedback system
simply is a dynamic system with a limit cycle - the interesting issue
is whether there are other kinds that can do an equally good or better
job at keeping organisms alive, especially keeping mammals alive, with
their highly programmable behavioral repertoire. Kugler and Turvey
have a paper where they appear to demonstrate the regulation of
respiration rate is accomplished by some other than a feedback system,
but an important part of the demo is that the respiration rate is
not actually controlled at a unique level - the system has a whole
string of rates that it will be stable at, and a big enough disturbance
will push it from one basin to another. This would not be a nice
property to have in a programmable motor-control system.
Avery.Andrews@anu.edu.au