[Martin Taylor 970428 10:15]
Rick Marken (970427.2230 PDT)]
Martin Taylor (970427 17:00) --
> What you may not agree with is that I don't see this failure
> [of MCT] as proof that it [finding a version of MCT that doesn't
> fail] couldn't be done.You probably can find a version of MCT that doesn't fail in a particular
case; but if the model has to be changed, post hoc, to
fit every new set of observations then I think you've got a
problem.
That's precisely the Ockham's razor answer. Yes, indeed, you would have
a problem. And you would be justified in discarding the MCT approach if
the model had to be arbitrarily reparameterized for each new situation.
Hans's examples have had no capability for self-adaptation, but when he
originally introduced us to MCT, self-adaptation (Kalman filtering) was
a big part of it. So, without knowing much more about it than what I've
learned by following the discussion, I'd guess that he could probably
produce MCT models that would require less arbitrary changing of parameters
than the versions he has shown as examples. If not, then we'd be completely
justified in dropping consideration of MCT--and even if so, we might be
justified in dropping it if the number of arbitrary parameters in the
self-adapting model turned out to be substantially more (or to require
more accuracy than) the parameters in a self-adapting (reorganizing) PCT
model. That, of course, is assuming we don't go in for dissection and
circuit tracing;-)
Look, I basically agree with Bill on this, that MCT may apply usefully at
the high levels, but is unlikely to do so at low levels. But I have no
data to support what I can best characterize as my _prejudice_. All I
know is that when Bill has said something about an MCT model not controlling
well, Hans produces one that controls ideally, and so on and so forth.
Yes, it looks like a new ad-hoc version to suit each situation, but No,
the versions have not conformed to Hans's original description of MCT
as a system that adapts itself to the circumstances. So I don't consider
the reparameterization as worse than discomfiting.
If one actually took what you say here seriously he or she would simply
reject science as an approach to acquiring knowledge. You seem to be
saying that it is impossible to rule out _any_ explanation of a
phenomenon using systematic observation. I think this is post-modernist
crap.
As with much of what you say, I can't guess where you drag this notion up
from. Ockham's razor has been a prime tool of science since before Newton,
and remains so. If you _read_ my Ockham's razor paper, available on the
Web at <http://www.servtech.com/public/mmt>, you will see that systematic
observation, and more importantly the extension of the range of data, has
to be the primary way of distinguishing among hypotheses. And THAT's what
I'm applying to the MCT-PCT issue, as opposed to applying the prejudice
that I know I have in favour of PCT.
Martin