Un instrument de mesure des croyances. La Règle de Bayes

Abstract

Jean-Pierre Clero. A measure instrument for beliefs: Bayes' Rule. Bayes's Essay (1763) belongs to a time when the notion of causality began to be criticised. Bayes, contrarily to Bernouilli, does not work on the more and more precise search for a probability, but on the confidence with which it is possible to predict an event from what had happened before. This approach, not from the point of view of the things, but from that of man, is not unique: Buffon and Hume had the same. Bayes also answers Berkeley's criticism against Newton and the « utilitarian humanism » which underpins his demonstrations. If to criticize interest leads to scepticism, Bayes' concern for a formalised method is a safeguard for Science and, there, Bayes belongs to the lineage of Hume and of Nominalism. If neither a scientific law nor the word itself, corresponds to essences but to regularities, the fundamental problem is the confidence which these deserve, and Bayes' Rule makes it possible to deduce it from past events. Furthermore, Bayes' Rule allows the integration of the experience's time dimension, which grounds authority in previous observations, as it is indeed the case for all human institutions. As to Bayes' Rule itself, it must be understood as a measure instrument since no « true value » has a life of its own: what is new in the bayesian approach is this idea, similar to those of contemporary utilitarianism, which purports that it is possible to define an equivalence between the credits mat are deserved by beliefs, Whatever they are. Let us note that bringing all beliefs to a same level may have adverse consequencies for religious beliefs, their privileged status being thus questioned.