On Sep 25, 2006, at 4:55 PM, GŁnther Greindl wrote:
> I would need it to model cultural diffusion processes, my approach
> stems from viewing human societies as complex adaptive systems
> and special interest will be invested in modeling the cultural
> (either with cognitive agents, or "memetic" data structures or
> in that direction).
> A good starting point seems to me the sugarscape model (I also want
> to include resources and competition between the agents).
> The agents should be adaptive in the long run (read: learning
> ability) -
> although this goes somewhat contrary to the keep it simple paradigm.
I'm just about to release my source code (refreshing it for MASON 11
and doing general cleanup/commenting) and you'll be welcome to look
at or use it. I implemented 70-75% of the rules and outcomes in
Growing Artificial Societies. I believe Repast's Sugarscape has only
2 or 3 rules as a 'proof of concept', but I didn't look closely to
see easily one might grow the source code base with additional
rules. I've not implemented anything in Repast so I can't make any
direct comments about it. Since both MASON and Repast are general-
purpose agent modeling and simulation toolkits, I find it hard to
believe that one would be significantly better than the other for
social phenomena per se. Once the parameter space and number of
agents reach certain levels, performance starts to matter. I was
able to make informed decisions about which MASON data structures and
classes to use based on the Javadoc, tutorials, and consulting the
core design team [a bunch of times].
Any kind of complex cognitive architecture will fly against the
philosophy of Sugarscape, but is interesting that the agents have no
overt memory to use in reasoning/(ir)rational decision making. The
most complex and adaptive cognitive capabilities are the welfare
estimation equations--which are tied to culture/cultural preferences
in one set of rules/outcomes. I think this comes closest to your
desire to model adaptive agents and cultural diffusion. The authors
argue that they were after sufficiency in terms of simple rules
generating complex phenomena. As agents/models become more complex
they will become more difficult to understand and in turn
establishing causality between code and simulation outcomes. Given
the very little replication that has occurred in this field, in
general, there are many open questions.