Coherent Choice and Epistemic Entrechment (Preliminary Report)

1 I n t r o d u c t i o n The notion of the epistemic entrenchment has recently at t racted quite some attention in the AI community. It was introduced, first under the name "epistemic importance", by Peter G~irdenfors [7, 8]. He specified two possible "origins" of epistemic entrenchment: an information-theoretic appoach and a paradigm approach. However, neither of these approaches squares well with the logical constraints which are placed on entrenchment relations in the later and more mature work of Gfixdenfors and Makinson [10]. Although it is now known how epistemic entrenchment can be put to work technically, the problem of providing a systematic justification of the postulates for entrenchment relations is still largely unresolved. The present paper offers a solution to this problem which is faithful to the literal meaning of the term "entrenchment". We follow the philosophical strategy of reducing logical principles to abstract principles of rationality for decisions and actions which we consider to be more fundamental than the former. The standard theory of epistemic entrenchment [10, 21] is interpreted in terms of the general theory of rational choice, and the following questions are being addressed. Are the usual postulates for entrenchment justified by principles of rational choice? Do the latter principles, or the mere interpretation of entrenchments in terms of choice, impose interesting additional constraints on entrenchment relations? Conversely, which principles of rational choice can be retrieved from the entrenchment postulates, and what is their status in the theory of choice?