How Artificial Intelligence Can Beat Professional Poker Players?

There have been some enormous breakthroughs with artificial intelligence technology over the past couple of years. We’ve seen AI help to encourage social distancing, solve protein folding, and most importantly, learn to have fun. Well, sort of. A team of scientists came together to teach their artificial intelligence machine how to play poker. The results from their experiment have been astounding. We’ll examine them here and see what this means for the future of AI.

Who Made It Happen?

The team behind this breakthrough is DeepStack. This company offers open-source artificial intelligence technology to the world, enabling technology to be pushed further by enquiring minds. Their programs are often used in factory automation, monitoring systems, and smart home installations. One thing that they really wanted to find out though, was if their artificial intelligence model could master a somewhat complex game. Not just well enough to beat other robots, or even the team at DeepStack, but rather to beat poker champions.

Changing the Strategy

It turns out that not only is artificial intelligence capable of learning poker rules incredibly quickly, it can also put them to good use. DeepStack AI were the masters behind this plan and it would be fair to say that not even they predicted the success they would have. Previously AI had been taught to play poker and could achieve some success against a new player. However, difficulties arose when the AI’s strategy was found out, as it had no way to formulate a new one. This was the first challenge for DeepStack’s model. The team taught the AI to use a model to play only the rest of the hand, not the rest of the game. This meant that as the game changed, so could the AI’s strategy, leaving it less open to exploitation.

Learning to Estimate

Some people would assume that having no goal at all in mind for the rest of the game could affect DeepStack’s performance negatively and they’d be right. The team taught DeepStack to use what they called an intuitive local search. Instead of running through every possible situation that could arise in the following hands, the AI made rough estimates of what the value of holding certain cards would be and it played accordingly. Making estimates in this way is an incredibly ‘human’ technique, that has exciting applications outside of the world of poker.

Intuitive Local Search

Perhaps the most exciting thing about this intuitive local search, is that as the AI continues to play poker, it improves its intuition. Just as you and I get capable of making decisions more quickly the more experience we have at playing a game, so too does this AI. the more games it played, the better its outcomes. It was learning on the job and putting poker players to shame. DeepStack played more than 44,000 games and beat all players except one by a margin that was statistically significant. If you want to see it for yourself then there are several of the games available to stream on Twitch. It makes for fascinating viewing, watching a manmade program beat a certified poker pro.