A self-learning evolutionary chess program

Abstract

A central challenge of artificial intelligence is to create machines that can learn from their own experience and perform at the level of human experts. Using an evolutionary algorithm, a computer program has learned to play chess by playing games against itself. The program learned to evaluate chessboard configurations by using the positions of pieces, material and positional values, and neural networks to assess specific sections of the chessboard. During evolution, the program improved its play by almost 400 rating points. Testing under simulated tournament conditions against Pocket Fritz 2.0 indicated that the evolved program performs above the master level.