Abstract. This paper presents a knowledge-based analysis of a number of multi-agent systems. The analysis is based on the knowledge level hypothesis of Newell and Clancey. We claim that the models of problem solving which we obtain constitute descriptions of global coherence in multi-agent systems. We contrast the results of this analysis with the conventional views of coherence.
IntroductionOne objective in the design of multi-agent systems (MAS) is for an agent to coordinate its actions with other agents in a coherent way. we propose a knowledge-based analysis of coherence in distributed systems in which the resulting models can be formally specified.The formal approaches typically combine two intentional attitudes: an information attitude and a pro-attitude (a definition of these terms and an introduction to the various formal approaches can be found in the review chapter of this volume). There are two objectives in the formal specification of multiagent systems. Firstly, formalisation increases our understanding of multi-agent interactions by making intuitive concepts such as common knowledge and group intention more precise and revealing new relationships between them. Secondly, the ability to verify the behaviour of a real multi-agent system requires the existence of a formal model against which it can be compared. Achieving the second objective also entails being able to describe existing MAS in terms of the language and concepts of the formalism, this, however, can be problematic. One example is the concept of common knowledge [11] which can be shown to be prerequisite for agreement among agents, but which defines a situation which cannot be achieved in practice. 3 * This work was supported by DAAD Stipendium number 513 009 0293 at the DFKI, Kaiserslautern, Germany 3 A modified concept of common knowledge can be achieved by real systems, for a fuller discussion see [11]