Re: Modeling the Emergence of Political Parties

[glen e. p. ropella]

At 11:06 AM 10/22/99 -0700, you wrote:
> I think when some scientists do a derivation or "proof" with
> differential equations, they think they've discovered a result that is
> universal. On the other hand, when they see a simulation model result,
> they consider it one data point for one special set of circumstances.

[...]

> papers are mighty mighty rare.) But somehow its a lot easier to ignore
> the uncertainties in a differential equations model.

I agreed with your entire post.  But, the implication deserves
a little more text, I think.  (If for no other reason than to
prove that I'm still as wordy as I used to be.)

What you seem to be talking about, here, is the difference between
experimentalists and theorists.  Simulations as data points is
a hard pill for the simulant to swallow because of a kind of
parental relationship she has with her baby.  In this, they
are more akin to the way theorists treat their work.  They
come up with an idea, nurse that idea into maturity, then
champion the idea to the world.  The umbilical chord that
grows during this process is often hard to snip.

Experimentalists don't have anywhere near this much of a problem
with their data.  They have the luxury of being more aloof,
abstract, and removed.  One needn't champion the existence or
validity of a data point because data is thought to be somewhat
more *real* than conjectures and ideas.  (Of course, we all
know that this is an illusion because the data was taken
and recorded in a context, which is full of bias.)  Largely,
it's easier to say, "well, the data must be flawed" than it
is to say, "well, my theory must be flawed".

But, what *I'd* like to do, is cast simulation in an
experimentalist light by calling it something more like
what it is: experimental mathematics.  A program is a
formal system.  It runs on a well-defined architecture.
Heck we even have to bend over backwards to get it to
*appear* capable of randomness.  A simulation (or any
program, really) is just a formal system that somehow
came into existence (sometimes through skilled fingers
and sometimes through sheer incompetence... but it doesn't
matter where it came from).  This system should be treated
as an "information source" that generates data.  We then
can go about inducing the structure of this information
source via all the wonderful math we've come up with over
the years to hypothesize about other naturally occurring
systems.

If you treat your computer programs in this way, you will
almost never be caught off guard by an unwanted artifact.

Now, this seems to deny the benefit most people expect from
simulations, namely, to help them tease out (analysis) the
cause and effect relations between sub-systems in the system
they're simulating.  It makes the simulation as hard to understand
as the system being simulated.  But, that's not really the
case.  Using a simulation, you have direct access to the ...
ahem... genetic encoding of the simulation.  The simulation
can help you find the genotype<=>phenotype map in a real
system solely by providing you the potential to find a
genetic encoding (*any* genetic encoding) that leads to
the same phenomenology as the real system.

Contrary to popular belief, the genetic encoding you find
(or dream up.. or read in a textbook) probably has absolutely
*nothing* to do with the real genotype in the real system.
Only the phenotypes match (if you do your validation right).
But, having access to a generator of a system that is
phenomenologically similar to the system you're studying
gives you some knobs to tweak that help you think up
questions you can ask of the real system to find out
where your generators differ from the real system.

So, just as with every other tool that's useful in science,
all simulation does is help you ask more questions, which,
to fold back to your original point, Steve, means that
simulations are really only data generators.

glen


--
glen e. p. ropella =><= Feeding the hamster wheel.  Hail Eris!
Home: http://www.trail.com/~gepr                (505) 424-0448
Work: http://www.swarm.com                      (505) 995-0818


                 ==================================
  Swarm-Modelling is for discussion of Simulation and Modelling techniques
  esp. using Swarm.  For list administration needs (esp. [un]subscribing),
  please send a message to <address@hidden> with "help" in the
  body of the message.
                 ==================================