The Role of Context in the Analysis and Design of Agent Programs

Abstract

We discuss the notion of context as applied to the verification of agent programs, and in particular, to the verification of agent programs based on the PRS agent architecture, Georgeff and Lansky [6]. Agent programs are an interesting domain for theories of context for the following reasons: (i) the context of an agent program has both internal (mental state) and external (embedding in the world) aspects, (ii) a logical theory of agent program verification using context-based reasoning must therefore address both syntactic and semantic issues, and (iii) the context of execution of an agent program is dynamic since agents are situated in a dynamically changing environment. We then consider the development of PRS agent programs from the designer’s perspective, and present a logical system of context-based reasoning that enables PRS programs to be proven correct. The methodology involves the program designer constructing contexts for the various procedures used by the agent, so the variety of contexts relevant to the PRS agent is fixed in advance by the programmer and is highly constrained by the PRS agent architecture. The study of context in agent programs thus raises a wide range of general questions that may be considered in the more controlled settings of particular agent architectures and execution environments.