[Hans Blom, 941130b]
(Tom Bourbon [941129.1633])
Contact! So _that_ was what you were talking about last year. This
construal of the "internal model" makes a lot more sense to me than the
kind of model I _thought_ (erroneously, it would seem) you were talking
about before. Maybe I was too quick, back then, to assume you were
talking about a world model as a representation or recreation of the
world, rather than as the set of parameter values in a control system.
Glad to establish contact!
The preceding section looks like one of the conclusions we reached (more
or less) a year ago: given an environment that contains a set of
adaptive perceptual control systems, all with similar intrinsic reference
signals for certain perceptions and each adapted to the point where it
effectively controls its own perceptions, it is probable that the adapted
parameter settings in the various systems will not be identical.
Assuming adaptation occurs through a random Ecoli reorganization,
different systems will "set themselves up" in different ways, but all
will achieve the same end. If the control systems were human, an
implication of such a state of affairs might be that they disagree about
how the world works, or about the best way to produce a particular result
in that world. Are we on the same wavelength, Hans?
Yes! Yes! Yes!
In the account above, when you say "learning is complete," do you mean
that, in a variable environment, the system's parameters allow it to
maintain perceptual control within the tolerance set by the sensitivity
of its own adaptor process? If so, we are probably in accord. (Looks
like I am hedging a bit, doesn't it?)
I don't know whether I would express myself this way, but let me see
whether this is close enough. "Learning is complete" when the organism (or
the adaptive control system) reached convergence in its estimation of the
correlation between e.g. an action and its effect. This depends upon how
subtle actions can be, how accurate perceptions, and on the details of the
averaging mechanism (does it average over a finite or infinite horizon,
does it "forget", can it be reset by an extreme outlier, etc.). In total,
on the system's built-in characteristics, _not_ necessarily on the quality
of control that is reached. When playing poker, for instance, you learn
what to call given a certain set of cards and given your impression of the
opponents' expectations/knowledge. You learn an "optimal" strategy, i.e.
the strategy with the best odds, not necessarily a strategy that wins in
every case.
Does this make things more clear?
That depends on whether you agree with my interpretions of your post.
Yes, now we _do_ seem to be in contact!
Greetings,
Hans
