Agents, agency and autonomy : a formal computational model

• Main Author: d'Inverno, Mark
• Published: University College London (University of London) 1998
• Subjects: 006.3
• Online Access: http://ethos.bl.uk/OrderDetails.do?uin=uk.bl.ethos.586863

In recent years, there has been an explosion of interest in agents and multi-agent systems in a variety of areas including artificial intelligence, and software engineering. Agent technology, however, is still relatively young, and there is much debate and discussion over many important concepts and the relevant terminology. In particular, in a great deal of agent research, agents themselves are defined in wildly different ways, if at all, and this makes it extremely difficult to be explicit about their nature and functionality. These problems have arisen in part due to the lack of a common structure and language for describing and reasoning about both single agents and multi-agent systems, which might facilitate a rigorous organisation of the field. In response to this, a four-tiered theoretical formal framework for agent systems is proposed, which we use as a base from which to develop a detailed model of agents and their dimensions, the properties required by agents for effective operation, and the social organisation of multi-agent systems. This framework essentially comprises entities, objects, agents and autonomous agents, and specifies the relationships between them to provide a rigorous and detailed analysis of the structures underlying all such systems. Key to the understanding of this work is our overarching concern as computer scientists, of building computational systems. The development of formal theories and systems as proposed above is inadequate if they are irrelevant to the needs of practitioners. While the construction of any theory or model is unavoidably somewhat removed from the realities of software development, we address this concern by ensuring that the tools used are standard software engineering ones that are accessible and support practical development. In illustration of these ideas, and as a demonstration of the validity of the arguments made, we show how the framework and models developed can both provide a theoretical foundation and be applied directly to existing agent systems and theories: in particular, the Contract Net Protocol, AgentS- peak(L) and Social Dependence Networks.