Playing Iterated Rock-Paper-Scissors with an Evolutionary Algorithm

Rémi Bédard-Couture, Nawwaf Kharma


Evolutionary algorithms are capable offline optimizers, but are usually left out as a good option for a game-playing artificial intelligence. This study tests a genetic algorithm specifically developed to compete in a Rock-Paper-Scissors competition against the latest opponent from each type of algorithm. The challenge is big since the other players have already seen multiple revisions and are now at the top of the leaderboard. Even though the presented algorithm was not able to take the crown (it came second), the results are encouraging enough to think that against a bigger pool of opponents of varying difficulty it would be in the top tier of players since it was compared only to the best. This is no small feat since this is an example of how a carefully designed evolutionary algorithm can act as a rapid adaptive learner, rather than a slow offline optimizer.


Paper Citation