Exploring Ensembles of Alternative Agent Based
Models
Steve Bankes
bankes@evolvinglogic.com
www.evolvinglogic.com
Most current research using Agent Based Models (ABMs)
use these models for hypothesis generation by providing examples of models
with interesting properties. Since few if any of these models can ever
be believed to perfectly replicate the details of important social systems,
more aggressive use of ABMs will require different research strategies.
Individual ABMs can plausibly be asserted to be members of classes of systems
which also include social systems of interest. In order to make assertions
about actual social systems, for either policy analysis or theory, we must
be able to discover invariant properties true for all members of classes
to which such systems belong.
In this talk, I will describe how ensembles of Agent
Based Models can be defined that plausibly spans classes of systems including
social systems of interest. I will show how such ensembles can be systematically
explored in order to discover useful and interesting patterns including invariance.
I will further demonstrate research methodologies for searching and sampling
from such ensembles that can be used to support plausible conclusions about
invariant properties of ensembles of ABMs and hence of the classes of systems
they represent. I will demonstrate this approach using ongoing Agent
Based Modeling research being done by myself and colleagues.