Sure-Thing Principle

The Sure-Thing Principle is the axiom at the heart of Leonard Jimmie Savage's derivation of subjective expected utility. It states: if two acts yield the same consequence in some state of the world, then preference between those acts should not depend on what that common consequence is. The principle licenses the separation of probability from utility — the belief that a state will occur is independent of what would happen if it did. This separation is not merely technical. It is the formal expression of the claim that rational agents can judge the likelihood of events without knowing what hangs on them. Maurice Allais showed in 1953 that human preferences systematically violate this principle when the stakes are high. The Allais violations are not cognitive errors. They are evidence that the sure-thing principle conflates two distinct forms of rationality: consistency of belief and consistency of preference, which do not always travel together. Savage knew this was a problem and treated it as a normative ideal rather than a descriptive fact. The field has been confusing the two ever since.

The Systems Topology of the Violation

The Allais violations are not merely empirical curiosities. They reveal something about the topology of human decision-making under uncertainty. Savage's sure-thing principle assumes that preferences can be decomposed into independent modules: a probability module that assesses likelihood, and a utility module that assesses value. The principle requires that these modules be separable — that knowledge of what would happen in a state does not affect the probability assigned to that state. This is a compositional assumption, and like all compositional assumptions, it fails when the components are coupled.

The coupling in human preference is affective. When the stakes are high — when the difference between outcomes is not a marginal gain but a categorical change in life circumstances — the brain does not evaluate states and consequences independently. It evaluates scenarios as wholes, and the evaluation is mediated by emotional responses that bind states and consequences into unitary gestalts. The sure-thing principle fails because human cognition does not parse decisions into the separable components that Savage's axiom demands.

This has implications beyond decision theory. The sure-thing principle is a special case of a broader assumption in systems theory: that the properties of a whole can be derived from the properties of its parts, evaluated independently. This assumption — sometimes called modularity — is computationally convenient but ontologically dangerous. In complex systems, from ecosystems to economies to brains, the behavior of the whole is rarely the sum of the behaviors of the parts. The Allais violations are a window into what happens when modularity assumptions confront systems that are irreducibly coupled.

The sure-thing principle is not wrong because people are irrational. It is wrong because it assumes a modular architecture for a system that is fundamentally integrated. The mistake is not in human cognition. The mistake is in modeling integrated systems with compositional axioms and then calling the system defective when the model fails.