Decision Making and Planning by Autonomous Agents; A Generic Architecture for Safety-Critical Applications

Abstract

Patient management in medicine requires flexible techniques for decision making and planning. Classical decision theory has important limitations in this respect, in that it fails to provide a theoretical framework for all phases of decision making; setting objectives; identifying possible decision options; arguing the pros and cons of the options, and committing to new beliefs or plans of action. Symbolic Decision Theory (SDT) addresses these problems, drawing primarily on various forms of qualitative and semi-qualitative reasoning and formal research in nonstandard logics. The basic ideas of SDT are outlined, focusing on the core of the decision procedure, the construction of arguments for alternative hypotheses and actions. The paper also outlines applications of the theory in the RED project ¹, which is using SDT in the design of decision support systems for safety critical applications. RED has incorporated symbolic decision making and plan management functions into a generic architecture for an autonomous agent. For safety-critical applications we propose augmenting the usual methods for develo** knowledge based systems with a range of safety engineering techniques, including conventional software engineering methods and techniques for reasoning explicitly about hazards, constraints and obligations.

This article is based on a fuller paper accepted for publication in Knowledge Engineerin Review, Volume 11 (1996). Extracts reproduced with kind permission of Cambridge University Press. © Cambridge University Press.