An essay in rational psychology

We examine several formulations of the common practice of jumping to conclusions when actions demand decisions but solid knowledge fails. This practice permeates arti cial intelligence, where systems assume many conclusions automatically as defaults simply because the questions they decide are known to occur frequently, and where other assumptions are formulated and adopted only when ignorance stalls action. After developing the motivations and general nature of these inferences, we introduce a formal basis for describing them. This formulation allows separate introduction of the several ideas involved, and so facilitates characterization of some important combinations and some previous proposals. Initial results are proved about these theories, including the aptness of the formal notions with respect to the intuitive motivations. Bene ts of this formulation include an indication of the ways notions from logic and metamathematics can enter into psychologies without subscribing to all of logic or metamathematics, an indication of the importance of conservation of mental states in the description of psychologies, and formal and intuitive relations between the approach of reasoned assumptions and its popular alternatives, deductivism and Bayesianism. Acknowledgments My debts, obvious and hidden, are many. As always, Joseph Schatz helped more than I can say. Jon Barwise, Johannes Brahms, Jaime Carbonell, Johan de Kleer, Merrick Furst, Clark Glymour, Jussi Ketonen, Robert Laddaga, David McAllester, John McCarthy, Drew McDermott, Marvin Minsky, Robert Moore, Allen Newell, Joann Ordille, Raymond Reiter, Dana Scott, Robert Stalnaker, Gerald Sussman, Peter Szolovits, and Richmond Thomason all helped in di erent ways. Jan Zubkoff graciously did the initial typing. The Computer Science department of Carnegie-Mellon University greatly aided my writing. Note to the Reader To keep this paper to a reasonable length, I assume some mathematical sophistication and acquaintance. Passage of time will, I hope, permit a more relaxed and thorough development. The symbols used are collected and glossed at the end of the paper. The headingsDefinition, Theorem, Corollary, Lemma, and Conjecture serve their usual purposes, and are numbered in a single sequence. Some proofs have been omitted. The heading Question often indicates an issue calling for proper formulation rather than a simple answer. i Analysis of