Originating in Ray Solomonoff’s work in the 1960s, Solomonoff induction predicts sequences by combining computable explanations. Each program that generates a pattern contributes probability mass; shorter programs receive more weight. The resulting Bayes mixture keeps multiple explanations in play while favoring simpler ones. section 9.1.2
A shorter program gets more prior weight
With a fair coin assigned to each program bit, a particular 10-bit program has prior weight 2⁻¹⁰, while a 100-bit program has weight 2⁻¹⁰⁰. The longer program receives 2⁻⁹⁰ as much prior weight. section 9.1.2
