Agent-Based Modeling and Simulation

When trying to understand, analyze or optimize a real world system, the transfor­mation of processes and entities into a model has been established as a first step. By modeling systems based on its components, a profound basis for the representation of emergent phenomena is provided.

When using ABM to create systems of agents trying to solve problems, multi­agent systems (MAS) are formed (Ferber 1999). In order to simplify the creation of MAS, a number of software toolkits like NetLogo¹ or RePast^{[1] [2] were developed since the beginning of the nineties. As a result of this even non-informatical sciences, such as social sciences or economics, adopted ABM as inherent part of their set of research methods.}

MAS can be used to generate or reconstruct emergent behavior by executing agent-based models and simulating interactions between the agents. The process of executing models is called computer simulation and provides a series of bene­fits. In case real world experiments would influence or even damage the real system, using computer simulation prevents risk to the real system. Furthermore, experi­ments might be too sophisticated in order to be executed in real systems or real sys­tems are not accessible to researchers. Finally, and this is the most relevant aspect when analyzing long-term dispersal processes, even systems that are not or no longer existent can be examined by using computer simulation.

Fujimoto defines computer simulation as follows:

A computer simulation is a computation that models the behavior of some real or imagined system over time (Fujimoto 2000).

This definition names three relevant components of computer simulation: the sys­tem, the model and the simulation itself (see Fig. 2.6).

A system, also referred to as original system or real world system, is the object of study. It contains a holistic view on relationships between entities and processes. The original system is then being mapped to a model or model system, a formal description of certain effects or phenomena being part of the original system. This is accomplished by the use of domain specific methods for describing the behavior of the system. However, the process of mapping real systems to models is associated with a loss of precision. Models, by definition, are never equivalent to the original system, as the only adequate model of a real system is the system itself. Nevertheless, it meets certain appropriate features or relationships of the original system’s object of study. Simulation can therefore be used for performing experiments with the model. By this means either the real system’s model or artificial scenarios generated by altering the model can be analyzed (Klugl 2001).

In order to use simulation results for predicting the original system, two require­ments need to be met: verification and validation. Verification describes the process

Fig. 2.6 Relation between original system and model system (Timm and Hillebrandt 2006)

of evaluating whether the software itself is designed and programmed in a correct manner, whereas determining the model’s appropriateness for representing the orig­inal system is called validation. However, a detailed consideration of how to verify and validate a computer simulation will not be part of this article as a variety of methods is sufficiently described in literature (Kleijnen 1995).

When providing a valid and verified model of a real world scenario, which depends on the research of domain experts, computer simulation can be used for gaining knowledge from experiments. By applying assumptions to the simulation model, artificial scenarios can be created and analyzed. Furthermore, hypotheses concerning the behavior of the system under certain circumstances can be evalu­ated. Therefore, and due to the condition given in the context of the Out-of-Africa- Hypothesis, we propose the application of agent-based modeling as an innovative methodology for modeling, simulating and analyzing artificial societies for under­standing the dynamics of hominin dispersal processes.

1.4