The Allais Paradox: What It Became, What It Really Was, What It Now Suggests to Us

Maurice Allais introduced his famous paradox in 1952, not as a descriptive tool to show how people actually behave, but as a normative argument regarding how a rational person ought to behave. The author argues that the paradox was intended to challenge the von Neumann-Morgenstern (VNM) independence condition by appealing to the rationality of the agent as a person, rather than just the consistency of their choices. This historical investigation reveals that the early decision-theoretic community, including figures like Savage and Samuelson, was already engaged in a normative debate, contradicting the common narrative that a 'normative retreat' occurred only after empirical refutations emerged.

By the late 1970s, the Allais paradox was largely repurposed as an empirical refutation of the expected utility (EU) hypothesis. This shift, solidified by Kahneman and Tversky, transformed the paradox into a heuristic for identifying empirical regularities like the 'certainty effect' and 'common consequence effect.' The author contends that this transition, while fruitful for empirical science, impoverished the paradox by stripping away its original normative depth. The paper highlights that early followers of Allais in the 1970s attempted to maintain this normative dimension by incorporating reflective stages into their experiments, a practice that was largely abandoned in later, purely descriptive research.

To move beyond the current divide between normative theory and empirical observation, the author proposes a new experimental approach. Drawing on Allais's concept of the 'experimental definition of rationality,' the paper suggests that experiments should not merely record choices but also elicit the reasons guiding those choices. By using a ternary structure—choices, followed by a presentation of rules of conduct, followed by final choices—researchers can identify subjects who are 'rational' in the sense of aligning their actions with their endorsed rules. This approach aims to bridge the gap between is and ought, allowing experiments to inform which decision rules are normatively sustainable.