Agent Modelling in Distributed Intelligent Systems

Agents in a distributed intelligent system may construct models of other agents and use them to forecast their behavior and to decide their own actions. This may impose limitations on the coherence of the whole system and affect agent behavior in other ways. An abstract view of such model construction is described. The consequences of this abstract view are explored using simulation experiments. The twin assumptions that this numeric value obeys a law of diminishing returns and that changes in it can be modeled by a nonlinear dynamical system lead to different behavioral regimes for the system in different ranges of values for the precision — especially in regard to the coherence of the system. Finally, the implications for a behavioral model of distributed intelligent systems are discussed.