A wager against Solomonoff induction

Abstract

The blog post critically examines Solomonoff’s universal prior in the context of Artificial General Intelligence (AGI) and argues against assigning zero probability to non-computable universes. Suggesting an alternative model of universes as sets of “value locations,” it investigates the implications of computable (countable) versus incomputable (uncountably infinite) universes. The author proposes that potential AGIs should assign non-zero credence to incomputable universes to accommodate for uncountably infinite sets of value locations. Furthermore, the concept of “acausal trade” between agents in countable and uncountable universes is explored, indicating possible gains. The author recommends considering the possibility of incomputable universes, cautioning that an AGI using Solomonoff induction might overlook immense amounts of potential value. It suggests a need for closer, careful consideration of AGI’s goals toward incomputable universes. – AI-generated abstract.