Michael Moehring and Elke Schumacher
Abstract
A Multi Agent Approach for Modelling and Simulation in the Social Sciences
We present ongoing research within a project called MASSIF (multi-agent simulation in social science interdisciplinary research). It intends the design and
implementation of an agent-based simulation tool for social science research. One essential issue of the underlying architecture is the structuring in several
hierarchical layers. The objective is twofold: First it provides an orientation scheme for potential users of the simulation tool to find their appropriate level of support,
and second it is conceived as a mechanism to facilitate the formal description of complex social systems. We introduce our multi-layered approach and show its
application with some examples of interaction structures modeling elementary social processes.
Strongly differing needs and skills of users of a simulation tool discourage the development of a unique modeling language. Our experience gained by the
development and usage of the modeling and simulation system MIMOSE shows that users of a simulation tool for the social sciences can be divided into different
groups spanning from simulation and programming experts over social scientists in research and teaching to lay persons and decision makers. To cope with these
heterogeneous requirements we propose to provide several hierarchically ordered user layers stacked on top of each other and being transferable between each
other.
To apply this conception seems to be fruitful especially in the first part of the simulation life cycle where a formal description of the model has to be generated which
can be executed by a software system. This task comprises the definition of the model's dynamic behavior and its structure. Motivated by this, we suggest to split up
the description of a multi-agent model into specifying properties and capabilities of each agent and to define structural relations (i.e. the interaction patterns) between
the agents.
The formal description of structural relations occuring in social systems can be a difficult task as they are quite complex. A popular mechanism in systems theory to
reduce complexity is decomposing into components. In doing so, one of the essential properties of most social systems can be used -- redundancy. That makes it
possible to identify subsystems, repeatedly. Examples of those subsystems are families, groups, topological communities, hierarchies, markets and so on. We
suggest to facilitate the description of complex social systems by decomposing them into building blocks such as relations, elementary interactions and messages and
specify these modules on a distinct level of abstraction. Therefore, our concept provides three layers, each one defining syntactic and semantic issuses.
Agent-based simulation of social systems may require specific kinds of interaction structures. Our concept directs special attention to the link between micro and
macro level and provides mechanisms to explicitly represent aggregated agents. This comprises modeling of interactions between aggregates (e.g. a society) and
their components (the individuals). It also supports the implementation of models whose structure is not fixed but variable, i.e. their description entail the possibility
to change their structure. This model class is requisite to simulate the emergence of social structures such as groups or organizations.