[What?] learning proposal

CSGnet Archive CSG1996
1

[From Chris Cherpas (961127.1835 PST) ]
  [re: Bruce Abbott (961127.2010 EST)]

Bruce Abbott (961127.2010 EST)

I do have in the back of my mind the idea that the failure of the old act to
"produce" would lead to the repetition of previous acts that were successful
in the past (and now being recalled from memory), but I don't have much of
an idea yet how to construct the mechanism that might have this result, just
a short list of its requirements.

Chris Cherpas (961127.1835 PST)
Bruce, could you give me a pointer to where the list of requirements
is stated and/or synthesize what might have been spread out over
several posts into an updated proposal/list?

Bruce Abbott (961127.2010 EST)

Because the rat already knows how to make any movements
it wants to, all it needs in order to obtain food is some reason to want to
approach and press the lever -- it need not learn any new transfer
functions...

Chris Cherpas (961127.1835 PST)
I'm assuming the rat spends a lot less time/energy reorganizing than
"directly controlling" only after a fair amount of experience. But
for _any_ given interval, in this case let's say the few seconds
before the pellet is delivered, I imagine there is _some_ time spent
reorganizing around at least the lowest/intensity level perceptions
because one never is in exactly the same position/orientation/whatever
as perceived at some previous time. My view is that living in the
same Skinner-box with the same contingencies in effect would lead
to a linneage of reorganizations which progressively decrease the
amount of time spent reorganization at higher levels. The relative
amount of reorganizing would be pushed lower and lower, since only
the minutia of control would need to be reorganized. The limit is
then zero reorganization (not considering for the moment changes in
other variables like water deprivation or whatever).

Maybe this is related to Bruce's proposal, but I don't see why
error has to reach catastrophic levels for a little more reorganization
to occupy the organism's busy life. Maybe I'm just confusing "normal"
HPCT control and reorganization/learning, but it seems that _where_
what I would consider some amount of on-going reorganization in the
hierarchy is concentrated or distributed shifts around continuously;
and gets "pushed downward" as the organism "learns the task." Sorry
about all the quotes for the moment, but I'd appreciate it if somebody
would be willing to interpret comment on this perspective.

Regards,
cc

Regards,
cc

2

[From Bill Powers (961128.0100 MST)]

Chris Cherpas (961127.1835 PST)--

My view is that living in the
same Skinner-box with the same contingencies in effect would lead
to a linneage of reorganizations which progressively decrease the
amount of time spent reorganization at higher levels. The relative
amount of reorganizing would be pushed lower and lower, since only
the minutia of control would need to be reorganized. The limit is
then zero reorganization (not considering for the moment changes in
other variables like water deprivation or whatever).

This would be the case if organisms were capable of producing _exactly_ the
same outputs under the same conditions every time, and sensing _exactly_ the
same state of affairs every time. But perceptions adapt and muscles fatigue,
and there is both noise and computational inaccuracy in the nervous system
itself. The result is that there is always a background of self-generated
disturbances, which require continual corrective shifts in actions.

And don't forget that great big fat source of intrinsic error signals called
"80% of free-feeding body weight" or "24-hour water deprivation" and so
forth. I would expect a rather high continuing level of reorganization to be
going on as long as this condition is maintained. It could be that a large
part of the variability we see in the behavior is due to continuing
reorganization. Who knows?

BEst,

Bill P.

···

Maybe this is related to Bruce's proposal, but I don't see why
error has to reach catastrophic levels for a little more reorganization
to occupy the organism's busy life. Maybe I'm just confusing "normal"
HPCT control and reorganization/learning, but it seems that _where_
what I would consider some amount of on-going reorganization in the
hierarchy is concentrated or distributed shifts around continuously;
and gets "pushed downward" as the organism "learns the task." Sorry
about all the quotes for the moment, but I'd appreciate it if somebody
would be willing to interpret comment on this perspective.

Regards,
cc

Regards,
cc