Duhemian Themes in Expected Utility Theory

This study aims at reexamining the philosophical problem of underdetermination of scientific theories by experience as well as applying the results thus obtained to the history of expected utility theory. For these purposes, we revisit Duhem's ideas, leaving aside those of Quine, which we do not regard as so relevant, and identify in the former, on the one hand, a thesis of the impossibility of crucial experiments, and on the other, a thesis of underdetermination properly, which will then become the focus of the analysis. We explain the intentionally imprecise, both historical and pragmatic, solution that Duhem offered to the problem created by this thesis for empirical refutation. Our expected utility application concerns the von Neumann-Morgenstern (VNM) theory of choice between lotteries, which we present through its historical development. We emphasize the normative scope of the theory from its beginnings, along with the special difficulty that this raises for empirical assessment, and we reinterpret the 1952 crisis and the Allais paradox in the light of this crucial observation. Then, we highlight the lengthy steps of experimental and theoretical research which the best specialists needed to take before they could conclude that the VNM theory was refuted. Reinterpreting this remarkable episode of the recent history of social sciences, we view it as an illustration of both Duhem's thesis and a resolution according to his ideas; i.e., it became unavoidable to reject the independence condition, among those of VNM theory, because accumulated arguments turned the balance against it. We close our reflective history with a short discussion of two theories that emerged in connection with this rejection, i.e., generalized expected utility theory and rank dependent utility theory. Another conclusion of the study is that theoretical competition did not condition the rejection, but rather benefited from it, contrary to what Lakatos's methodology of research programmes would let one expect. ; Cette étude se propose de ...