Artificial Superintelligence: A Recursive Self-Improvement Model

EasyChair Preprint 2505

Abstract

Recursive self-improving( RSI ) systems create new software iteratively. The newly created software iteratively generates a greater intelligent system using the current system, then this process leads to a phenomenon referred to as superintelligence. However, many existing studies on RSI systems lack clear mathematical formulation or results. In this paper, we provide a formal definition of RSI systems and then we present a recursive self-improvement model by three different approaches. The first one is to find an optimal program defined by given scores and program generation probabilities using Markov chain. The second one is to model by embedding histories when generating a new program. And the third is to model the programs taking a program as an argument and return a suggested improvement of the given program. We use simulation to show that we achieve logarithmic runtime complexity with respect to the size of the search space and realize good accuracy to a AI model of embedding histories. The results suggest that it is possible to achieve an efficient recursive self-improvement.

Keyphrases: Artificial Intelligence, Markov chain, Recursive Self-Improvement, artificial superintelligence, program embedding, recursion

@booklet{EasyChair:2505,
  author    = {Poondru Prithvinath Reddy},
  title     = {Artificial Superintelligence: A Recursive Self-Improvement Model},
  howpublished = {EasyChair Preprint 2505},
  year      = {EasyChair, 2020}}