Solomonoff's theory of inductive inference

Solomonoff's theory of inductive inference is a method for judging scientific models by how briefly they can be described. It holds that the best model of some observed data is the shortest algorithm that would generate that data, and it assumes that the programming language is chosen before the data are seen and that the environment is produced by an unknown algorithm. Introduced by Ray Solomonoff around 1960 and built on probability theory and theoretical computer science, it applies Bayes' rule with a universal prior that gives every computable theory a positive probability, so that shorter theories carry more weight. In this way it formalizes Occam's razor. Solomonoff proved that the induction is incomputable, though he regarded that limit as a benign one, and algorithms derived from AIXI approximate it in practice.