Toward Rational Planning and Replanning Rational Reason Maintenance, Reasoning Economies, and Qualitative Preferences

Efficiency dictates that plans for large-scale distributed activities be revised incrementally, with parts of plans being revised only if the expected utility of identifying and revising the subplans improves on the expected utility of using the original plan. The problems of identifying and reconsidering the subplans affected by changed circumstances or goals are closely related to the problems of revising beliefs as new or changed information is gained. But traditional techniques of reason maintenance---the standard method for belief revision--choose revisions arbitrarily and enforce global notions of consistency and groundedness which may mean reconsidering all beliefs or plan elements at each step. To address these problems, we developed (1) revision methods aimed at revising only those beliefs and plans worth revising, and tolerating incoherence and ungroundedness when these are judged less detrimental than a costly revision effort, (2) an artificial market economy in planning and revision tasks for arriving at overall judgments of worth, and (3) a representation for qualitative preferences that permits capture of common forms of dominance information. Planning and replanning We view the activities of intelligent agents as stemming from interleaved or simultaneous planning, replanning, execution, and observation subactivities. In this model of the plan construction process, the agents continually evaluate and revise their plans in light of what happens in the world. Planning is necessary for the organization of largescale activities because decisions about actions to be taken in the future have direct impact on what should be done in the shorter term. But even if wellconstructed, the value of a plan decays as changing circumstances, resources, information, or objectives render the original course of action inappropriate. When changes occur before or during execution of the plan, it may be necessary to construct a new plan by starting from scratch or by revising a previous plan. To replan effectively in demanding situations, replanning must be incremental, so that it modifies only the portions of the plan actually affected by thc changes. Given the information accrued during plan execution, which remaining parts of the original plan should be salvaged and in what ways should other parts be changed? Incremental replanning first involves localizing the potential changes or conflicts byidentifying the subset of the extant beliefs and plans in which they occur. It then involves choosing which of the identified beliefs and plans to keep and which to change. For greatest efficiency, the choices of what portion of the plan to revise and how to revise it should be based on coherent expectations about and preferences among the consequences of different alternatives so as to be rational in the sense of decision theory (Savage 1972). Our work toward mechanizing rational planning and replanning has focussed on four main issues: ̄ Identifying formal notions of rationality appropriate to agents of limited mental resources and methods for evaluating specific reasoning architectures with respect to these notions, ̄ Developing methods for incremental revision of plans based on a new approach to reason maintenance, ̄ Exploring a market-based approach to allocation of planning resources that simultaneously balances mental resources against each other and against physical resources, and ̄ Developing logics of preferences that make acquisition and specification of planning preferences convenient. This paper focusses on the latter three issues; for our approach to the first, see (Doyle 1988; 1992). Reason maintenance for replanning Replanning in an incremental and local manner requires that the planning procedures routinely identify the assumptions made during planning and connect plan elements with these assumptions, so that. replanning may seek to change only those portions of a plan dependent upon assumptions brought into question by new information. Consequently, the problem of revising plans to account for changed conditions has much 130 ARPI From: ARPI 1996 Proceedings. Copyright © 1996, AAAI (www.aaai.org). All rights reserved.