Executable specifications of resource-bounded agents

Abstract

Logical theories of intelligent (or rational) agents have been refined and improved over the past 20 years of research. Such logical theories are used in many ways, one of which is as the basis for executable agent specifications. Here, agents are specified using a logical description and then this description is directly executed in order to implement the agent's behaviour. This provides strong correctness results for the implemented agent, directly corresponding the formal specification with the implementation. With increased application, it has become clear that such specifications of idealised agents are inappropriate in practical situations. An agent may have many resource-bounds to contend with, both in terms of time and memory, and this should be taken into account within the agent's formal specification. Thus, in this paper, we tackle the problem of representing and executing resource-bounded agents. The framework developed here allows the restriction of the amount of reasoning, both temporal and doxastic, that the agent is permitted. We address the formal properties of the framework, present results concerning the execution of such specifications, and consider the practical outcome of such restrictions.