[From Rick Marken (970414.0900 PDT)]
Bruce Abbott (970413.1805 EST)
in "e. coli-type reorganization," as I understand it, reorganization
(the equivalent of tumbling) ceases when there is no error in
intrinsic variables.
This is not a necessary feature of the reorganization model of learning.
The model can work if there is some very low level of reorganization at
zero intrinsic error or none at all. I believe, however, that there is
some evidence that reorganization is _always_ going on, at a VERY low
level, even when we can control virtually perfectly (zero intrinsic
error). For example, both Bill and I have experienced "spontaneous
reversals" while we were doing
a tracking task (we suddenly start moving the mouse in the "wrong
direction" relative to deviations of the controlled variable from the
reference). We are both skilled trackers and these reversals occur VERY
rarely. They seem to be the result of the operation of a system (like
the presumed reorganization system) that occasionally changes a
parameter of the system randomly to see if there might be a better
way to control things -- even if control is going very well.
I can understand how a system attempting to gain control over a
given variable could do so by varying the gain (in sign and
magnitude) at random until good control resulted...But what if
this action is not what is needed? There may be literally
thousands of possible error-action relationships from among which
to select. Blind trial-and-error would seem to be an extremely
poor choice of method for finding a workable solution
I agree. There might be more than blind trial and error involved. But
this is an empirical question. We need research that looks at the
reorganization process in detail -- that tracks reorganzation as it is
occuring. A study that makes a first stab at this is reported by
Robertson and Glines (I don't have the reference handy; Dick, if you're
there, could you post it please). They found no evidence of
"non-randomness" during reorganization but that doesn't mean it's not
there in some (possibly many) cases.
Mary Powers (970413) to Jeff Vancouver (970408.1700 EST) --
Whatever point you are making, I don't take too kindly to being
misquoted in the process. Your omission completely changes the
sense of what I said.
I have to come to Jeff's defense on this one. I'm sure Jeff didn't
do this on purpose. His misquote looks like a simple case of someone
desperately trying to reconcile PCT with a cognitive view of behavior.
Best
Rick