emergence and parsimony - modeling theory!

[Chris Landauer]

hi, all -

I think we're mostly arguing about apples and bicycles here.  There are lots
of ways to study complex systems that are neither theorem proving nor
integration; i think abstraction, analogy, and scenario generation ("tell me a
story") are such things, and i expect that there are more.  On the other hand,
if by "theorem proving" you mean any and all formal modeling, and by
"integration" you mean any and all computational modeling, then i sort of
agree, but it bothers me, because i think there is something else that is
known to help, and the only way i can characterize it is as "inspiration",
though it does mean something about careful thought.

I was (and am) objecting in part to the notion that it matters at all whether
the system is *real*.  You have the same options in either case: even if a
system is *real*, there is no observational data without a model, which is
perforce NOT *real*, and we get back to the same set of choices.  One can do
much good modeling without any data at all; a theory is often sufficient.

I do agree with both Catherine's and glen's comments about parsimony though.
My point was really only that it is too often applied at the wrong level of
abstraction, models of biological systems being notoriously bad in that regard
(cf.  neural nets), and careful attention to what that level is will reap much
benefit to the modeler.

I also think that a major advantage of mathematical modeling (theorem proving)
over computational modeling (simulation / integration) is one of reliability
and generality of results, not just speed.


more later,
cal