Artificial intelligence (AI) last week took 1 more step towards betterment.
DeepNash, an AI agent learned the Stratego game from scratch to a human expert level by playing against itself. According to a post on DeepMind, the DeepNash algorithm uses a novel approach based on game theory & model-free deep reinforcement learning. Due to its Nash equilibrium play style, it is very hard for an opponent to exploit it.
The post explained that due to its hard work, DeepNash had reached the top 3 rankings among human experts on Gravon, the world’s largest Online Stratego platform. Stratego has more value than just gaming, the post claimed. A research paper shows how DeepNash can be used in uncertain situations & successfully balance outcomes to help solve complex problems.
DeepNash is a novel approach that combines game theory & model-free deep reinforcement learning. “Model-free” means DeepNash is not attempting to explicitly model its opponent’s private game-state during the game. In the early stages of the game in particular, when DeepNash knows little about its opponent’s pieces, such modelling would be ineffective, if not impossible.
DeepNash is powered by a new game-theoretic algorithmic idea that is called “Regularised Nash Dynamics” (R-NaD). Working at an unparalleled scale, R-NaD steers DeepNash’s learning behavior towards what’s known as a Nash equilibrium.
Game-playing behaviour that results in a Nash equilibrium is unexploitable over time. If a person or machine played perfectly unexploitable Stratego, the worst win rate they could achieve would be 50%, and only if facing a similarly perfect opponent.
In matches against the best Stratego bots – including several winners of the Computer Stratego World Championship – DeepNash’s win rate topped 97%, & was frequently 100%. Against the top expert human players on the Gravon games platform, DeepNash achieved a win rate of 84%, earning it an all-time top-three ranking.
Those of you interested in learning more, click here.
Image credit: DeepMind
