Re nement in Agent

A group of intelligent agents may work together in order to solve a problem or achieve a common goal. If the group fails to achieve a goal, it may be able to adapt its behaviour so that such a goal can be achieved in the future. One of the ways in which the behaviour of the agent group can be changed is by reening the knowledge of individual agents. 1 Motivation Suppose a group of students are working on a team project. They divide up the work involved and agree to meet on a particular day to put their contributions together and submit the project. One student intends to do his portion of the work the night before the team meeting, but is invited to a party, which he goes to instead of doing his work. He does not consider the consequences for the other team members. As a result, the team does not get a good grade for the project. This is an example of a failure to achieve a group goal due to the actions of one of the participants. How can the team ensure that it will succeed if it attempts to achieve a similar goal in the future? A group of intelligent agents should be able to ensure that such a failure is not repeated by adapting their behaviour. One way in which agents can adapt their behaviour is through the use of machine learning techniques to reene their knowledge. Our research is concerned with how they can do this eeectively. Extensive investigation of techniques for reening the knowledge held in a single knowledge base have already been carried out Ourston and Mooney, 1991] Richards and Mooney, 1992] Craw and Sleeman, 1991]. The process of reening multiple related knowledge bases, such as those of a group of agents, presents new challenges. The reene-ment of one agent's knowledge may aaect other agents in the system. The organisation of the agent community , the roles and expectations of individual agents, acquaintance groups, strategies used in cooperation, etc. amounts to a new body of knowledge that does not exist in a single-agent system, but which may also need to be reened. Agents are autonomous or semi-autonomous entities; their knowledge and reasoning processes are not necessarily transparent to other agents. Therefore, agents must volunteer information about their internal processing and cooperate with other agents in order to eeectively reene their …