Formal Methods and Agent-Based Systems

As has been discussed elsewhere [17], much agent-related work has tended to focus on either the development of practical applications, or the development of sophisticated logics for reasoning about agents. Our own view is that work on formal models of agent-based systems is valuable inasmuch as they contribute to a fundamental goal of computing to build real agent systems. This is not to trivialise or denigrate the effort of formal approaches, but to direct them towards integration in a broader research programme. In an ongoing project that has been running for several years, we have sought to do exactly that through the development of a formal framework, known as SMART, that provides a conceptual infrastructure for the analysis and modelling of agents and multi-agent systems on the one hand, and enables implemented and deployed systems to be evaluated and compared on the other. In this paper, we describe our research programme, review its achievements to date, and suggest directions for the future. In particular, we consider the role of autonomy in interaction. Autonomy is independence; it is a state that does not rely on any external entity for purposeful existence. In this paper, we use our existing agent framework to address the issues that arise in a consideration of autonomous interaction. We begin by considering several problems that are prevalent in existing models of interaction, and which must be addressed in attempting to construct a model of autonomous interaction. Then we introduce a previously developed agent framework on which the remainder of the paper is based. The next sections describe and specify an autonomous social agent that acts in an environment, the way in which it generates its goals, and finally how it interacts with others in its environment. We discuss how this can be viewed as a process of discovery, and what such a view usefully brings to the problem.