Poker is a game of imperfect information, and as such has remained one of the most illusive games for AI to master. But Facebook’s AI Pluribus has just bluffed the top poker players into submission.

As a serious poker player myself, and one who has researched its game theory for many years, this breakthrough in AI touches particularly close to my heart. I personally know why it is so difficult for amateurs to make money, and more importantly improve their skills in the game of poker. This very point is also what sets Poker apart from all other games previously beaten by AI, such as Chess, Dota II, StartCraft, Go and so many others, that have already been solved by AI.

In all the above games, taking the correct action will yield a positive result. If you made a great move in Chess, you will improve your chance of winning. If you made a great play in Dota II, you will now be more likely to defeat you opponent. This makes it much easier to train an algorithm, as its rewarded on good actions, and “punished” on bad ones.

But poker is different. In poker the correct decision will often lose the hand, and the incorrect decision will often win the hand. This concept is tough for humans to grasp, and makes it challenging for an AI to train. If you bluff and lose (by say getting called), it could be an incorrect decision by your opponent to call, even though he will win if he does! Likewise, next time you are faced with the same spot, it is often correct to bluff again, even though you lost doing just that the last time.

For an artificial intelligence to win in poker, specifically in a six handed no-limit game, is a huge breakthrough. It is also a bit scary, considering the online poker community could soon be flooded with super-human intelligence bots picking them apart.