Annotation and matching of first-class agent interaction protocols

Abstract

Many practitioners view agent interaction protocols as rigid specifications that are defined a priori, and hard-code their agents with a set of protocols known at design time - an unnecessary restriction for intelligent and adaptive agents. To achieve the full potential of multi-agent systems, we believe that it is important that multi-agent interaction protocols are treated as first-class computational entities in systems. That is, they exist at runtime in systems as entities that can be referenced, inspected, composed, invoked and shared, rather than as abstractions that emerge from the behaviour of the participants. Using first-class protocols, a goal-directed agent can assess a library of protocols at runtime to determine which protocols best achieve a particular goal. In this paper, we presented three methods that enable agents to determine if a protocol achieves a specified goal. The two most promising approaches allow an agent to summarise a protocol that it has learned by calculating the outcomes achieved by the protocol, and annotate the protocol with these summaries. The agent can match, via annotations, which protocols in a library achieve a given goal. © 2009 Springer Berlin Heidelberg.