DeepStack is the ultimate poker opponent. Charles University, the Czech Technical University, and the University of Alberta joined together to create an AI system that could win at an imperfect information game, in this case poker. With a paper released in March 2017, DeepStack secured its place in the world as the first AI to beat professional poker players – 11, to be exact.
There have been AIs that can play games like chess before. But, chess is a perfect information game. Meaning, everything one needs to know about the game is right in front of them, the pieces and moves are visible to the opponent. Poker has more unknown variables than chess and information remains hidden to the opponent. Games like poker were uncharted territory for AI. To conquer the challenges in the world of poker – this was the ultimate goal that the developers at DeepStack had set themselves.
According to their website, Deepstack was created with heuristic search techniques and deep learning. In other words, it was created to be intuitive in its search for solutions and its learning was hierarchical on a brain like network. These techniques and methods resulted in a system with three main traits: DeepStack only looks as far ahead as the current hand, DeepStack has an ‘intuition’ regarding its current hand, and it works at human speeds.
In the end, the team achieved their goal: they had developed an AI system which beat 11 expert poker players from around the world in all but one margin.