Introduction

Why dig once again in a field as thoroughly ploughed as expected utility? Only because a mass of experimental work, and some recent theory as well, has pushed the practitioner of expected utility – and the teacher of future practitioners – into an uncomfortable corner. By the practitioner I mean the consultant or “decision analyst,” hired as a professional guide through a tangled thicket by a decider faced with complex choices under uncertainty. Such a practitioner's main tool was and remains the original expected utility rule in its personal probability form. Up to now this rule stands alone in its simplifying power. It lets the practitioner break up the client's (the decider's) complex task of ranking alternative acts into manageable pieces. It permits the practitioner to elicit the client's beliefs and attitudes by asking small questions, to piece the responses together into a ranking of the contemplated complex acts, and to argue that the ranking is inescapable if the client wants to be consistent with personal attitudes and beliefs, consistent with a few basic axioms (constraints on the final ranking), and consistent with the rules of logic.

Yet the experiments of more than two decades have shown that competent persons cheerfully violate one or more of the classic axioms. This has called forth some new theory, notably Machina's elegant “generalized” expected utility theory (Machina 1982), wherein both obeyers and violators of certain classic axioms exhibit the same kind of consistency: Both rank acts – specifically they rank probability distributions on a wealth interval – according to a smooth functional defined on such distributions.