DeepMind is known for its impressive artificial intelligences. The Google subsidiary is particularly fond of the board games Shōgi, Chess and Go. But now the new AI called DeepNash is supposed to go one step further, because it has learned to beat almost all human players at Stratego.
For those who don’t know Stratego, the goal is to find out the identity of your opponent’s 40 pieces, bluff, and ultimately capture the enemy flag.
With such a strategic game, it’s fascinating to see how DeepNash has learned to cope with the game’s imperfect information to eventually become one of the best games in the world.
In this game, the AI must weigh all possible outcomes before making a decision, making Stratego one of the more complex games in AI history. As DeepMind writes, the game Go has ten to the power of 360 possible game states, while Stratego has ten to the power of 535.
To face human opponents, the AI has developed an unpredictable strategy to ensure that the other player is completely in the dark. Even bluffing was learned by DeepNash in this game, as sometimes low-ranking pieces were played as if they were worth much more.
