Agent complexity

[Gary Polhill]

I'm just trying to canvass some opinions on complexity of agents in
agent-based models, and I would welcome the views of anyone with a
moment to spare.... From reading around a little in agent-based
modelling literature, it is clear that there are a number of views
about how complex (perhaps sophisticated would be a better word...)
our agents should be. On the one hand, there is the KISS view, as
put forward by Axelrod (1997) -- try and get to the heart of a
phenomenon by finding the minimal conditions necessary for it to
emerge; on the other hand (mostly from a Distributed AI direction),
agents are expected to have language, goals, planning and a
production system to manage all their rules. The two views do not
necessarily conflict... For example, if our agents are supposed to
be human there might be agreement that any phenomenon we might
care to observe will require the agents to have a high level of
sophistication. However, I am not sure that all sociological agent-
based models do have such sophisticated agents. Epstein & Axtell's
(1996) sugarscape agents certainly aren't.... their agents have at
most 6 rules in any of the examples they show, and there is no
explicit goal representation so far as I can see. Johnson (1998) couches
the debate in terms of Tractability vs. Verisimilitude -- the drive
for excessive realism hampers research by resulting in a multidimensional
parameter space that cannot be searched in reasonable time.

What are your views on agent sophistication? Are you building models
containing agents with superior cognitive abilities ... what is your
justification? Is it anything more than realism? How will you manage
the parameter space? Or if you have simpler agents, how do you intend
to convince people that your results are relevant/valid?

Does anyone know of two ABMs claiming to show the same phenomenon,
but one having very much simpler agents than the other?

Gary


References:

Axelrod, R (1997) "Advancing the art of simulation in the social sciences" in
Conte, Hegselmann & Terna (eds) "Simulating Social Phenomena" Lecture Notes in
Economics and Mathematical Systems 456, pp.21-40, Springer-Verlag, Germany


Epstein & Axtell (1996) "Growing Artificial Societies: Social Science from the
Bottom Up" MIT Press


Johnson, P (1998) "Adaptive agents versus rational actors: Social science
implications" Paper presented at the annual meeting of the Americal Political
Science Association, Boston, MA 3-6 September 1998. Available online from
http://lark.cc.ukans.edu/~pauljohn/ResearchPapers/APSA98-MTHDS/APSA98-Methods.html


-- 

Macaulay Land Use Research Institute, Craigiebuckler, Aberdeen. AB15 8QH
Tel: +44 (0) 1224 318611               Email: 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.
                  ==================================