Introduction to the Special Issue of Group Decision and Negotiation 2002: Theory and Practice of Computational Coordination Mechanisms in Multi-Agent Systems

摘要

Due to a network of dependencies the agents within the system need to interact with one another to simultaneously solve the group conflict(s) as well as their own local problems. Therefore, one of the most central problems of MAS has been the design of coordination mechanisms that have desirable properties, such as selection of an outcome that is not only socially optimal but is also computationally tractable and requires minimal information. Traditionally, the elements of such a coordination mechanism has included specification of: (1) distributed decision making models, (2) protocols (or rules) of interaction and (3) agent communication languages. The goal of this special issue of the Group Decision and Negotiation was to bring together a number of novel papers that addressed these elements from either a computational or a theoretical perspective. To better understand the current status of MAS and the rationale behind the goal of the special issue and the relation of the papers in this issue we must look at the history of MAS. The MAS paradigm derives much of its goals (but not its methodology) from the field of Distributed Artificial Intelligence (Bond and Gasser 1988). Distributed Artificial Intelligence (DAI) itself is a subfield of AI whose goal is to extend the classical knowledge level (Newell 1982) single computational agent paradigm to include representation and reasoning mechanisms for distributed agent systems at the social level. That is, design of a system of agents that can represent, reason and act not only over the possible states of nature but also states, beliefs and actions of other rational agents in the environment.