Moving multi-agent systems from research to practice

The state of the art in multi-agent research and engineering is insufficiently reflected in the state of the practice in complex distributed systems because the community has yet to demonstrate the significant benefits of using agent-oriented approaches to solve complex problems. The practitioner's view that multi-agent approaches are not technically superior to traditional approaches is understandable; for every successful multi-agent system, it is possible to envision a non-agent approach that is equally suited for the task. Agent-oriented software engineering lies directly at the heart of this problem. In order to be accepted, the agent community needs to demonstrate that they can build reliable complex, distributed systems using agent-oriented approaches that are repeatable and sound. This paper identifies three obstacles that hamper progress towards such a demonstration: the lack of a common understanding of key multi-agent concepts, the lack of a common set of notations and models, and the lack of flexible, industrial strength methods and techniques for developing multi-agent systems.