emergence and parsimony - modeling theory!

[glen e. p. ropella]

Chris Landauer writes:
 > "glen e. p. ropella" <address@hidden> also answered:
 >         If it's *not* a real system, then you've only got two choices: 
 > develop
 >         the micro behavior into a formal system and proceed via standard
 >         proofs, or integrate the system forward in time and measure the macro
 >         behavior.
 > 
 > nonsense - many choices exist, always, whether or not the system is *real*
 > 
[...]
 > the theoretical challenge of emergent behavior is to characterize the
 > kinds of relationships that can occur, and to devise ways to compute
 > macro from micro more quickly than by using simulation, and to suggest
 > choices for or classes of micro from desired or observed macro, to
 > reduce the difficulty of inventing or explaining interesting systems

I don't think it's nonsense to assert that we only have two
choices for exploring the global behavior of a formal system:
theorem proving and integration.  The goal is to "characterize
the *kinds* of relationships that can occur and to *compute*
macro from micro more quickly than by using simulation" [emphasis
mine].  But, in a system where one *has no* data about the macro
behavior of a system, the way this characterization begins is
by using mathematics, integration, or a combination of the two 
in order to find the macro behavior with which the micro behavior
can be related.  If there is no data on the global behavior of
a system, then no relationships can be found.

 > finally, we note that parsimony simply does not apply (Occam's razor
 > is usually the wrong choice for reasoning about biological systems) -
 > one of the well-known facts about biological systems is that there are
 > many "historical artifacts", whose existence depends on how easily is
 > was to change from something that _already_ existed, not on how easy
 > it was or is to invent or describe the concept from first principles

I agree that the principle of parsimony is not always useful
and can sometimes obscure useful solutions; but, I tend to 
agree with Catherine in that it is still a valuable goal.
One thing I think is essential to modelling that Catherine
doesn't mention explicitly is that we are "simulating" systems,
not "emulating" them.  We don't necessarily want to imitate 
a natural system in all it's complexity.  There are some types
of simulation that are trying to do more emulation than 
simulation, namely, those created for the purpose of the study
and measure of the consistency and reliability of a corresponding
physical system.  In these cases, the software should be as
exact an analogue as possible.  And the only real purpose for
the simulation is to cut down on the cost of experiments (time
is not even the real issue).

But, that's not really Swarm's target.  Swarm is, I think,
intended to be a modelling tool.  Which means that a simulation
should avoid as much detail as possible and help extract the 
important pieces of the systems' behavior.  Swarm is a filter,
not a microscope.

glen