Re: Simulating Individual Behavior

[Stephen C. Upton]

This is a great thread!!!

First, I'm reminded of a lyric:
IBM,
UBM,
We all BM
for IBM

(I believe this was from the science fiction book "With a Finger in my I"
by D. Hofstader -- if not, then from some other 1970's scifi book on AI :-)

Second, I'll paraphrase Hamming (R. W. Hamming, Numerical Methods for
Scientists and Engineers, 1962):
The purpose of modeling is understanding, not numbers.

Third, I'll add some of my own thoughts:

I am currently attempting to understand warfare.  (For those that find this
offensive, consider the better we understand the origin and causes of
warfare, the better we can prevent it)
There are quite a few models and simulations, as  you may imagine, of
various aspects and levels of warfare, from the individual soldier to the
campaign level, e.g., Desert Storm.  Each abstracts certain features which
are relevant to a particular analysis.  There have been several good points
made that seem to cross problem domain boundaries:
1.  The resolution or the level of abstraction of the agent (actor,
individual, entity) 
2.  The resolution or the level of abstraction of the interaction between
agents
3.  Collecting, gathering, identifying data for 1 and 2
4.  Identifying patterns or structures, emergent or not (modes, or the
resultant behavior of the system as a whole)
5.  The difference in modeling philosophy between ABM and ODE's or PDE's
(The Newtonian legacy)
6.  The relationship of the model to the "real world"
(Maybe this was obvious!)

There have been several attempts to model the campaign level using ABM's.
However, for computational reasons,these were arguably unsuccessful.  But
the attempts bring up some good questions:
At what level do you model your agent or their interactions? Is a Battalion
sufficient or do I need to model individual tanks?
How much detail is required?  (More detail, More detail is the current
rallying cry)
How does that level of detail relate to system characteristics, e.g.,
location of strange attractors, etc.?
How do you know when you have the level of detail and interactions correct,
especially if there isn't a whole lot of data, at either the agent level or
the system level?

Steve Emsley says, "Personally, my interest is in developing a model as a
'surrogate experimental system".  I concur.  I also don't have much of a
choice, except that historical data provides some clues as to what the
important parameters, plus some thinking about the processes in general.
But then I'm not sure if those parameters were a function of some other
interactions, e.g., the evolution of new tactics as a function of technology.

I also sense some "My model is better than your model", i.e., ABM's are
"better" than ODE models, or ODE's are "better" than ABM's.  This is an
easy trap to fall into.  Each is abstracting system characteristics and
behavior differently.  Certainly, thermodynamics is as useful as
statistical mechanics, within their respective regimes.  However, I
currently believe the ODE folks are more guilty of this, but we don't want
the pendulum to swing all the way to other side either.  

Finally, we are all probably guilty of thinking our models represent the
real world in some useful fashion. We should always maintain a certain
amount of sceptism for any model and remember the paraphrase, modeling is
for understanding.

Love to hear your comments.

thanx





*** * **** * ***** ********* ** ****** ***** *** ***** ******** *********
Stephen C. Upton
TSA-5, MS F602
Los Alamos National Laboratory
Los Alamos, NM 87545
505-667-9435    FAX 505-665-2017
address@hidden
******* ********* *** ** *** ******** **** ****** ** ****** **** *** ***


                  ==================================
   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.
                  ==================================