Note: [1] Preliminary results of this research were published in the proceedings of AAMAS 2008.
Abstract: The ability to reliably represent and replicate choices people make is crucial for building accurate models of day-to-day situations. The fact that people are inherently rationally- and computationally-bounded increases the difficulties in designing such simulations. This paper builds on the use of peer-designed agents (PDAs) – computer agents developed by people – to show their effectiveness in generating a variety of strategies and behaviors and in alleviating the simulation and behavior analysis of systems populated by human individuals with diverse strategies. The paper synthesizes the PDA-based simulation components and ideas that appear in recent PDAs literature into a cohesive simulation design, and reports a set of experiments aiming at validating the ability of PDA-based simulations to exhibit realistic behavior both in the individual agent and the system levels. The validation of individuals' ability to reliably capture their strategies into PDAs is quantitative, relying on four games from different domains. The domains vary in aspects such as game complexity and the environment dynamics. The applicability of the PDA-based approach in the system level is evaluated using a large scale experiment involving 34 PDAs, each designed by a different person. All in all, the set of strategies obtained by means of PDAs is substantially richer and more varied in comparison to the limited sets of strategies used in prior multi-agent simulation studies.
Keywords: Peer designed agents, simulation, strategy design
