An artificial-intelligence (AI) system has beaten professional poker players at a two-player version of the game called Texas hold'em that has no limits on betting.

Poker is hard for machines to master because it has a huge number of possible outcomes and some information — the identity of an opponent's cards — remains unknown. Michael Bowling and his colleagues at the University of Alberta in Edmonton, Canada, created an algorithm, called DeepStack, that learns how likely it is to win in different scenarios by playing against itself in millions of games. During a game, it uses this 'intuition' to explore only a fraction of possible future moves and outcomes, allowing it to form its strategy in real time. The AI system outperformed 10 out of 11 players by a statistically significant amount over the course of the 3,000 games played by each individual.

Such algorithms may one day be used to aid decision-making in other areas that lack perfect information, such as defence, say the authors.

Science http://doi.org/b2jd (2017) For more on this story, see p.160