Universal prior


by Eliezer Yudkowsky Jun 20 2016 updated Jun 21 2016

A "universal prior" is a probability distribution containing *all* the hypotheses, for some reasonable meaning of "all". E.g., "every possible computer program that computes probabilities".

A "universal" prior is a probability distribution that assigns positive probability to every thinkable hypothesis, for some reasonable meaning of "every thinkable hypothesis". A central example would be Solomonoff induction, in which the observations are a sequence of bits, the universal prior is "every possible computer program that generates bit sequences", and every such computer program starts with a positive probability.