Developing multi-agent systems
The previous chapter introduced the idea of multi-agent systems and offered some examples of simulations based on this approach. This chapter goes into more detail about designing and building multi-agent systems, outlining a design process that will help with moving from an initial idea to a working system. We shall review some techniques for describing, testing and validating multi-agent systems and conclude by considering how multi-agent simulations can be integrated into research and policy development. Chapter 2 introduced some of these methodological issues, but this chapter will go into more detail and apply them to multi-agent simulations.
The chapter will be illustrated by reference to a quite typical small-scale multi-agent model published in the Journal of Artificial Societies and Social Simulation (Jager et al. 2001). The article reports on a study of conflict in crowds made up of two groups, such as when the supporters of opposing football teams meet in the street. As the authors remark, the outcome can vary from the peaceful mingling of the two groups to the occurrence of fights and riots. The model allows the authors to experiment with different sizes of groups, made up of different proportions of ‘hardcore’ agents, ‘hangerson’ and ‘bystanders’. Eighty simulation runs were conducted with a variety of group sizes and compositions and the article concludes that fights are most common in large groups with a relatively large proportion of hardcore members when the groups are quite different sizes (see Figure 9.1).