A Developer's Perspective on Multi-agent System Design

摘要

This paper draws upon the practical experience gained in the development of software agents for the deployment of intelligent distributed services and information access. We review a set of multi-agent architectures starting from the communication and co-ordination requirements of such systems. The aim is to illustrate the common components in current designs and implementations of MAS which are often based on the communication nature of these systems. Further to this we show some benefits and drawbacks of these systems that are developed form this aspect. Part of the limitations of these systems is due to basing their communication semantic interpretation on the belief desire and intention model (BDI) which is a mental agency. The mental agency is used for the internal reasoning part of the agent and places implicit assumptions on the communication behaviour. We examine this limitation and report on how two MASs overcome some of the constraints. In light of these practical solutions we outline some pragmatic design concepts in reducing potential constraints of the BDI model on the communication layer. The result is a discussion about how to bridge between mental agency dependencies and the role of social agency when developing multi-agent systems.