next up previous
Next: Future Directions Up: The Open Agent Previous: Multimodal Map

Related Work

Agent-based systems have shown much promise for flexible, fault-tolerant, distributed problem solving. Much of the foundational work on agent technology has focused on interagent communication protocols [[Finin et al. 1997]], patterns of conversation for agent interactions [[FIPA1997]], and basic facilitation capabilities, including agent name servers and other types of registry services (e.g., brokers, matchmakers) [[Sycara et al. 1996]].

Because there is insufficient space here to cover the gamut of work on agent architectures, we restrict ourselves to mentioning several projects that have helped to evolve some notion of facilitation. Genesereth has emphasized the role of a facilitator [[Genesereth and Singh1993], [Genesereth and Katchpel1994]], and in [[Genesereth and Singh1993]] describes a facilitator based on logical reasoning. This facilitator shares our emphasis on content-based routing and the synthesis of complex multistep delegation plans, but does not go as far as OAA in allowing the service requester to influence the strategies used by the facilitator. Similarly, the InfoSleuth system [[Nodine and Unruh1997]] employs matchmaking agents having the ability to reason deductively about whether expressions of requirements (by requesters) match with the advertised capabilities of service providers. KQML [[Labrou and Finin1997], [Finin et al. 1997]] provides ``capability-definition performatives'', such as advertise, and ``facilitation performatives'', such as broker_one and broker_all. While these performatives may be suitable for structuring the basic interactions between the players in a facilitated system, it should be noted that they provide only a communication protocol. That is, the specific strategies employed by a facilitator, and the means of advising a facilitator in selecting a strategy, are beyond the scope of KQML specifications. Sycara et al. delineate the concepts of matchmaking, brokering, and facilitation in a useful way, and explore the tradeoffs inherent in the use of these approaches. Overall, they find that a brokered or facilitated system can exhibit dramatically better performance than one based on matchmaking.

Adam Cheyer
Mon Oct 19 17:14:26 PDT 1998