I am anything but satisfied with my comments in response to Han's message
I am quite sure that what I wrote comes across as weighted heavily
against the idea that model based control is used by living beings except
in quite narrow and restricted conditions or situations.
In a sense, I do believe that this is in fact the case but in another
sense I also believe that a case can be made for extensive use of "Model
While I was trying to think though some of my earlier assertions about
Model control use, it occurred to me that either a model is in use or
random action is in use whenever there is _any_ degree of open loop
conditions between the desired goal and the perceptual input.
Though I am not even pretending that I have an idea as to how this would
occur, it seems to me that what the model control system must do is set
references for lower level systems based upon experience (and possibly
even inference based upon experience) and then provide an "imagined"
perceptual signal (or signals) to the higher levels systems AS LONG as
the lower level perceptions (controlled or otherwise) compare "favorably"
to the expected values.
At the point where this is occurring, that is the lower level perceptions
are "favorably" matched to their expected values, THEN if the model is
indeed correct for the environmental conditions the perception "created"
by the model is also correct. At this point however, the model control
system IS NOT controlling in the normal PCT sense of the term though both
the systems above and below the model are doing so.
The reaching for the soap without looking is, I think, this sort of
situation. If your hand hits the shower wall prior to achieving the
expected "extension" then _overall_ control system failure has occurred.
However, these living control systems are adaptable to the point of
shaming the normal definition for the term and the world model is rapidly
resynced based upon the unexpected perception (in addition to the control
exerted by whatever lower level system was "seeing" to the perceptual
limit on force).
Though willing to be showen that the above is bull****, I think it is
pretty compelling. There seems to be a substantial difference between
the sort of experiences that Bruce and Bill are describing while running
the experiment and the kind of thing that I am thinking of. When running
the BlindTrk Model it seems to me that our experimenters do not really
have an operative model.
As to creating a model to duplicate the behaviour of the experimenter, I
think that we are faced with the problem that we do not have a model for
learning (specifically reorganization). This model would have to
perceive the actions used to control while visible and then duplicate
them when invisible (naturally) but what the model can perceive is by the
nature of our experiments not at all what the human perceives.