Algorithmic Probability, Heuristic Programming and AGI

Introduction This paper is about Algorithmic Probability (ALP) and Heuristic Programming and how they can be combined to achieve AGI. It is an update of a 2003 report describing a system of this kind (Sol03). We first describe ALP, giving the most common implementation of it, then the features of ALP relevant to its application to AGI. They are: Completeness, Incomputability, Subjectivity and Diversity. We then show how these features enable us to create a very general, very intelligent problem solving machine. For this we will devise “Training Sequences”— sequences of problems designed to put problem-solving information into the machine. We describe a few kinds of training sequences. The problems are solved by a “generate and test” algorithm, in which the candidate solutions are selected through a “Guiding Probability Distribution”. The use of Levin’s search procedure enables us to efficiently consider the full class of partial recursive functions as possible solutions to our problems. The guiding probability distribution is updated after each problem is solved, so that the next problem can profit from things learned in the previously solved problems. We describe a few updating techniques. Improvements in updating based on heuristic programming is one of the principal directions of current research. Designing training sequences is another important direction. For some of the simpler updating algorithms, it is easy to “merge” the guiding probabilities of machines that have been educated using different training sequences — resulting in a machine that is more intelligent than any of the component machines .