How to avoid maximizing expected utility

Abstract

Expected utility theory (EUT) implies fanaticism, the requirement that an agent prioritize an arbitrarily small probability of an overwhelmingly large value over any certain, lesser good. This implication poses a critical challenge to the internal consistency of normative decision theories, especially when applied to long-termist ethical frameworks or scenarios involving infinite payoffs. The structural origin of fanaticism lies in the interaction between the Archimedean axiom and unbounded utility functions, which necessitates that no finite benefit is lexically prior to a sufficiently large, though improbable, gain. Common strategies to circumvent this result—such as utility capping, the introduction of thresholds, or the adoption of risk-weighted decision models—often conflict with fundamental axioms of rationality, including transitivity and independence. Consequently, fanaticism emerges as an inescapable feature of any decision-theoretic framework that maintains a linear sensitivity to increasing value. Resolving the conflict between these theoretical dictates and common-sense moral intuitions requires either a radical departure from standard aggregation methods or a recognition that standard rationality necessitates choices that appear intuitively extreme. – AI-generated abstract.