Microchip v Brain: Why AI is Beating the Brain at Poker

Unlike many casino games out there, Poker is not simply a game of luck. It is not just about spinning a wheel or spinning the reels, there is much more to it.

Man v Robot

When you learn to play poker, you have to master the art of the bluff, knowing when to raise – and when not to; most of all, you have to learn the art of the poker face. The best poker players on the planet perfect the art. They spend years working out the best strategies and how to succeed. However, Artificial Intelligence has changed the face of the game and it’s exposing the weaknesses in the very best players around. This is a very sad day for the human poker playing population – but a great day for the ever growing world of Artificial Intelligence.

More than Just Chess

AI is not new to the scene when it comes to interacting and playing with humans. It has always been a dominant force in games such as chess, and even more recently, Go. However, because of the different levels to poker, it has tended to fall flat. Due to ongoing AI research, this has all changed. For those interested, the Amsterdam Affiliate Conference covers some news about the AI evolution in the gaming industry.

Top AI forces have managed to combine new algorithms alongside deep machine learning – which in some ways copies the workings of the human brain – and this allows the computers and AI machines to actually teach themselves. In fact, Murray Campbell, a senior researcher at IBM in New York and one of the creators of AI Deep Blue says of the step forward: “It’s a scalable approach to dealing with complex information that could quickly make a very good decision even better than people.”

However, chess and Go are very different games to poker. Chess and Go are simply information games. When you play these games, you can see exactly what your opponent has to play with. This helps the AI machine significantly. Poker, on the other hand, is different. The players are dealt two random cards face down and at the start of each new set of public cards, players decide whether to bet, hold or abandon the stake. Because this is such a random game and each player has two unknown cards, players make bets simply by guessing what the other players may or may not do. With chess, you can work out a winning strategy from the playing pieces on the board and all possible potential moves. Poker requires intuition.

Calculating Results

Traditional game playing AIs aim to calculate all the possible outcomes of the game and use this information to rank strategy options using a specific formula that allows it to search data from other winning games. There is a down side to this method though. In order to compress all of the data that it available, the AI algorithms can draw together different strategies that aren’t actually successful. The AI machine, DeepStack manages to avoid this by calculating just a few steps ahead as opposed to the whole game. It constantly recalculates new algorithms with the new data. So, when the AI is required to make its move before its opponent makes his/her bet and it receives new additional information, deep learning comes into play. Here, neural networks – systems that enact the information gained by deep learning – assist in limiting the possible situations that are factored by algorithms because the AI has been trained on all the different behaviours of the game. This, in turn, makes the AI faster and more accurate.

The Results

DeepStack was put to the test against 33 professional poker players. Over four weeks, the players and the AI played 44,852 games of no-limit Texas Hold’em. Having used a formula to completely get rid of instances where luck rather than strategy created the win, it was found that DeepStack won 486 milli-big-blinds per game. A milli-big-blind is 1000th of the bet that is needed to win a game.  Looking at that in layman’s terms, that’s around 10-times of what a pro poker player would consider a sizeable margin.

What are the implications of this? The success of the AI cannot be ignored, in fact, Libratus; an AI machine has won $1.5 million worth of chips from the top poker players in the world in a three weeks challenge. Top betting site, PokerStars, is looking into the emergence of AI, so if you wish to discuss this in detail then they are the people to talk to.


Other Poker news

Leave a Reply

To post comments you need to Login or register your free HighstakesDB account.


No comments have been posted yet.