Recently, researchers have developed an artificially intelligent poker playing system called DeepStack. This technological wonder has since become the first algorithm to have ever beaten professionals in poker, playing heads-up no-limit Texas Holdem.

Though this may seem less than impressive considering computers have beaten live contestants in the past in games like Jeopardy and standard trivia, it’s important to note that poker involves asymmetric information, being things like bluffing and body language, which cannot be measured by AI. Programming AI to process asymmetrical information becomes a much more difficult procedure, seeing as probability plays an enormous role in poker, which is largely determined by opponents’ reads on one another.

DeepStack uses a technology known as “counterfactual regret minimization” to reason through the strategies that come with playing cards. Its creators have even developed a program simulating a “gut feeling” for which cards to keep, and which to trade. Through this programming, the algorithm is able to adapt its strategy using recursive reasoning and attempting to predict the actions of each player. During this computing, it completes limited probability scenarios to try and give itself the best possible chances of winning.

While it is a remarkable breakthrough for artificial intelligence, similar programs have been created before. These however, involved imperfect information, mapping out a strategy for the entire game rather than step-by-step decisions. Because of this, these flawed programs could not compute the varying bets, which can cause a staggering amount of different iterations in just one game.

When developing DeepStack, its creators trained it to implement fast, approximate estimates of the current state of the game in which it is playing rather than attempting to predict the final outcome. Amazingly, they used over 10 million different examples of of random poker situations to teach the algorithm, then had 33 professional players from 17 different countries come in and compete against the revolutionary new system. In the end, DeepStack played 44,852 games against these professionals, and averaged an extremely successful win-rate doing so.

This algorithm that has proven to be able to perform successfully even with imperfect information has paved the way for even more breakthroughs in artificial intelligence. DeepStack has shown that AI has not reached its limits yet, and can be useful in many other fields. These may include medical treatment, environmental resources and strategies, and for simpler uses, social media or phone applications. Regardless of where the future of artificial intelligence is headed, innovators and creators are achieving new goals every day.