Artificial Superintelligence : A Model for Self-Improving / Self-Modifying Programs

EasyChair Preprint 3149

Abstract

Self-Improvement or Self-Modification is any behavior of a system where a program gets better at achieving goals as it receives input. An example of a self-improving program would be a program that gets better at playing chess by playing games against itself. In this paper, we provide a formal definition of self-improvement systems and then we present a self-modification model by two 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 describe a method of Applying Genetic Algorithm on Multilayer Artificial Neural Network  for updating and optimizing the neural network weights. GA creates multiple solutions and evolves them through a number of generations, and each solution holds all weights in all layers to help achieve higher accuracy. The evolutionary algorithm( i.e.GA ) was used as an optimization approach that mimics the concept of natural evolution for creating fitter individuals that have higher chance of survival through natural selection. The GA processes integrated with Artificial Neural Network model and the network improves itself by learning to optimize its own weights. We observe from the test results that the networks were able to self-improve through natural selection with good accuracy and also it is observed that self-modification mechanism for artificial intelligence is convenient.

Keyphrases: Genetic Algorithm, Markov chain, Self-Improving, Self-modifying, artificial superintelligence

@booklet{EasyChair:3149,
  author    = {Poondru Prithvinath Reddy},
  title     = {Artificial Superintelligence : A Model for Self-Improving / Self-Modifying Programs},
  howpublished = {EasyChair Preprint 3149},
  year      = {EasyChair, 2020}}