Evolving a Hex-Playing Agent

Hex is a two-player adversarial board game in which there is always exactly one winner. Although it is known that a winning strategy exists for the first player, such strategies are difficult to find due to the large branching factor of Hex's game trees. A subset of Artificial Intelligence research is devoted to optimizing search algorithms, such as minimax, pursuant to searching these game trees and solving Hex boards for any game position. Our research is not concerned with perfect playing strategies. Instead of minimax approaches, we use Artificial Neural Networks and Genetic Algorithms to test the bounds of how quickly and how effectively Artificial Neural Networks are able to learn to evaluate board-states of a game. We experiment with network topology and evolution strategies and compare different approaches using metrics we developed.