2 Formal Approaches to Agent-based Systems

done in bringing together the two strands of work. This section draws on Luck’s outline [17] of the ways in which some progress has been made with BDI agents, a well-known and effective agent architecture. Rao, in particular, has attempted to unite BDI theory and practice in two ways. First, he provided an abstract agent architecture that serves as an idealization of an imple-mented system and as a means for investigating theoretical properties [28]. Second, he took an alternative approach by starting with an implemented system and then formalizing the operational semantics in an agent language, AgentSpeak(L), which can be viewed as an abstraction of the implemented system, and which allows agent programs to be written and interpreted [27]. In contrast to this approach, some work aims at constructing directly executable formal models. For example, Fisher’s work on Concurrent MetateM [11] has at-tempted to use temporal logic to represent individual agent behaviours where the representations can either be executed directly, verified with respect to a logical re-quirement, or transformed into a more refined representation. Further work aims to use this to produce a full development framework from a single high-level agent to a cooperating multi-agent system. In a similar vein, [25] aims to address the gap be-tween specification and implementation of agent architectures by viewing an agent as a multi-context system in which each architectural component is represented as a separate unit, an encapsulated set of axioms, and an associated deductive mechanism whose interrelationships are specified using bridge rules. Since theorem-provers al-ready exist for multi-context systems, agents specified in this way can also be directly executed. As yet, the body of work aimed at bridging the gap between theory and practice is small. Fortunately, though, there seems to be a general recognition that one of the key roles of theoretical and practical work is to inform the other [8], and while this is made difficult by the almost breakneck pace of progress in the agent field, that recognition bodes well for the future. Some skeptics remain, however, such as Nwana, who followed Russell in warning against premature mathematization, and the danger that lies in wait for agent research [4]