Local Interactions, Explicit Communication and Causal Knowledge in Games and Multi-Agent Systems

In this paper we suggest new structures for modeling games (multi-agent systems). We use prime event structures as the branching runs representing all the possible plays in games. Event structures have been already successfully applied in the theory of distributed systems [18] and several temporal logics have adopted them as frames. We show that a prime event structure extended with a utility function defined on the terminal nodes can be naturally considered as distributed game in extensive form. The notion of passing knowledge introduces a novel concept to the standard definition of games in extensive form. As well, the notion of information sets (corresponding to agent’s knowledge) gets new insight, because now the information sets cannot be defined in an arbitrary way, i.e., they must correspond to causal dependencies in the system.