In this paper we outline a general approach to the study of metareasoning, not in the sense of explicating the semantics of explicitly specified meta-level control policies, but in the sense of providing a basis for selecting and justifying computational actions. This research contributes to a developing attack on the problem of resource-bounded rationality, by providing a means for analysing and generating optimal computational strategies. Because reasoning about a computation without doing it necessarily involves uncertainty as to its outcome, probability and decision theory will be our main tools. We develop a general formula for the utility of computations, this utility being derived directly from the ability of computations to affect an agent's external actions. We address some philosophical difficulties that arise in specifying this formula, given our assumption of limited rationality. We also describe a methodology for applying the theory to particular problem-solving systems, a...

The economic theory of rationality promises to equal mathematical logic in its importance for the mechanization of reasoning. We survey the growing literature on how the basic notions of probability, utility, and rational choice, coupled with practical limitations on information and resources, influence the design and analysis of reasoning and representation systems. 1 Introduction People make judgments of rationality all the time, usually in criticizing someone else's thoughts or deeds as irrational, or in defending their own as rational. Artificial intelligence researchers construct systems and theories to perform or describe rational thought and action, criticizing and defending these systems and theories in terms similar to but more formal than those of the man or woman on the street. Judgments of human rationality commonly involve several different conceptions of rationality, including a logical conception used to judge thoughts, and an economic one used to judge actions or...

Abstract not found

I illustrate the surprising strategic effects a lack of common knowledge can have in a simple example. I analyze a two-person bargaining game in which the seller makes a single take-it-or-leave-it offer. The seller knows everything except possibly what information the buyer has. If it is common knowledge that the buyer knows the value of the object or if it is common knowledge that the buyer does not know the value, then the buyer's expected payoffis zero. However, if the seller does not know what the buyer knows, then the buyer may be able to obtain some of the gains from trade. In the exogenous information case, where there is a fixed probability q that the buyer is informed, if q is far enough from zero and one, the buyer's expected payoff is strictly positive. However, if the seller is almost sure of what the buyer knows, the buyer's expected payoff is zero. Surprisingly, the endogenous information case, where, after seeing the seller's offer, the buyer decides whether or not to become informed, is quite different. If the cost of obtaining information is zero, we get the same outcome as it is common knowledge that the buyer is informed. However, if the cost of obtaining the information is small but strictly positive, the buyer's payoff is strictly positive. As the cost goes to zero, the limit of the buyer's payoff is strictly positive and may even be virtually all of the gains from trade. 2.

plans with perfect vision 126 8.7 The performance profile of the anytime planner 127 8.8 Abstract plans with imperfect vision 128 8.9 The conditional performance profile of the anytime planner 8.10 Compilation of vision and planning 8.11 Intra-frame optimization 8.12 The run-time display 8.13 Architectures for model based diagnosis 8.14 Distribution of solutions over time for S1-4 9.1 Real-time decision making List of Tables 2.1 Approximate relational algebra operations 4.1 The performance distribution profile of the TSP algorithm 5.1 Optimal time allocation in boolean expression evaluation 90 ix Preface Since my first encounter with a computer, I was fascinated by the possibility of writing programs that could really "think." A few weeks after my first BASIC lesson, I remember writing programs that could play and win simple games such as Nim and Mastermind. But those programs did not think. They were designed to follow a winning strategy that could be calculated by a simple formula. The computer's fast information retrieval and speed of computation were a substitute for thinking and left open the question of how to develop a program that could truly reason about its domain.