MODELS OF MULTI-AGENT BEHAVIOR: A SIMULATION AND EXPERT ENVIRONMENT APPROACH (COGNITIVE PSYCHOLOGY, ORGANIZATIONAL LEARNING, DYNAMICS, ARTIFICIAL INTELLIGENCE, TRUST IN TEAMS, ADAPTIVE CONTROL OF SYSTEMS)

The goal of this thesis is to improve our understanding of behavioral phenomena in multi-agent decisionmaking via modeling. A secondary goal is to develop a powerful simulation methodology for analyzing dynamic systems. A research question is the relevance of artificial intelligence techniques.;SEE is used to study the impact of biases, attribution heuristics, and trust on decisionmaking in a team whose members are myopic and altruistic. The theme of this study is trust as a counter-bias. Using experimental modeling and the tools in SEE for exploring parametric solutions, behaviorally substantial results are obtained.;Cognitive biases may cause behavior that is similar to behavior caused by self-interest. There exist qualitatively distinct dynamics of trust, all leading to good performance in the long run. However, success depends in complex ways on the context. When trust is determined by an adaptive process, three heuristics are found to be necessary and sufficient for achieving good performance over a variety of contexts, even with noisy performance observations.;Learning of performance parameters by members may mislead adaptation, and adaptation may cause temporary instabilities in the learning dynamics. Thus an organization may prefer slow-learning members to achieve controllability of behavior. If the members learn fast, if performance is not perfectly observable, and if short run performance is emphasized, a context-sensitive behavioral rule is likely to be superior to adaptive search.;A Simulation and Expert Environment (SEE), developed in LISP, integrates difference equation simulation with object-oriented programming and rule-based reasoning. The object-oriented approach offers a method for managing variants of the models. Ways to integrate rule-based reasoning and simulation are demonstrated, but the former's computational inefficiency limits usefulness. The system provides fast turnaround between defining a model and obtaining results, which increases the productivity of the modeler, and encourages experimental modeling, leading to novel formulations and results.;An integrated simulation environment allows the analysis of the often complex effects of behavioral phenomena on decisionmaking. It requires models to stay on an abstract level, emphasizing qualitiative insights over computational results.