Defining rational agents by using hierarchical structured knowledge bases

Summary form only given. Modelling a society of intelligent agents in the multi-agent testbed RATMAN is done by specifying different knowledge bases (KBs), one for each agent. The KBs not only contain the agents world knowledge, but they reflect the fundamental mental abilities of the agents, like the degree of communication facilities, learning skills or self-reflection. The authors describe the prototypical model of such a KB. The different knowledge packages are structured hierarchically from simple propositional knowledge to highly abstract (meta-) knowledge. They are realized with various representation formalisms (i.e different kinds of logics), which are mutually compatible. The agents are modelled by manipulating the prototype KB, i.e by adding specialized knowledge and/or by erasing parts of the knowledge packages