Le paradoxe d'Allais. Comment lui rendre sa signification perdue ?

Few problems in decision theory have raised more persisting interest than the Allais paradox. It appears that sufficiently many brilliant works have addressed it from within decision theory proper for history and philosophy of science now to enter the stage. In its historical side, the paper recounts the paradox as it arose, i.e., in 1952, at a Paris conference attended by the main decision theorists of the time. They had drawn renewed confidence in expected utility theory (eut) from the way von Neumann and Morgenstern had axiomatized it in 1947, and Allais devised his puzzle precisely to shaken their confidence. The issues between the two camps were normative, but they became lost in the developments of the 1980s that belatedly brought fame to the « Allais paradox ». These works restricted the paradox to be a straightforward empirical refutation, turning it into a stake of also exclusively empirically oriented non-eu theories. In its philosophical vein, the paper tries to evaluate this shift of interpretation. To an extent, decision theorists were right because their experimental work was thus freed from a major complication and amenable to illuminating results: eut was empirically refuted, the independence axiom of von Neumann and Morgenstern was the main culprit, and the next theoretical stage was to modify this axiom appropriately. However, they were also wrong in not addressing an essential feature of their field: i.e., that observed behaviour is informative only if agents are prepared to endorse it reflectingly, i.e., to endow it with some normative value. As reconstructed here, Allais meant to reserve choice experiments to rational subjects, who were either selected at the outset, or identified as such by the experimental results. The paper tries to flesh out Allais?s intuitions by turning to by now little known works of the 1970s, which under his influence provided experimental renderings of rationality, and it eventually suggests that decision theory might diversify its methods by taking inspiration from these original attempts. Classification JEL : B21, B31, B41, C91, D81