By Lorenzo Tesler-Mabe, April 2017

Poker is an incredibly fascinating game. On the surface, it appears to be a game of luck and gambling. In a single hand, or even a tournament, success really is “in the cards.” If you’re a complete amateur, but get a good starting hand in a crucial spot where your opponent has slightly worse cards, you can have a great amount of success. When we delve deeper, however, poker reveals itself to be a competition of skill. Over the span of hundreds and thousands of tournaments, a poker pro would demolish the amateur. At this scope of play, the luck factor cancels out, and it all comes down to pure skill. At least, that’s how it used to be. The technological revolution has spread into the world of poker, and not only is it changing the game drastically, but it also has the potential to take over completely.

So, how is technology changing poker? You see, there are two sides to the skill of the game: math (calculation) and intuition. As you can probably guess, computers really shine in the former department. Let’s look at a controversial example to explore this. With the emergence of online poker, dozens upon dozens of “poker hand analysis” programs have been designed to give players an edge. These programs allow users to view stats about themselves, as well as other players, such as: frequency of aggression (betting/raising), tendency to make it to showdown, fold to raise percentage, fold to 3-bet percentage, etc. (see glossary below for poker terms) The ethics of using such software is an entire topic in of itself. For our purposes, it should be noted that, depending on its complexity, certain software in this category basically automates the process of playing poker. Save for taking the time to process the data and clicking the mouse once to call or fold, a player who knows such software inside out is essentially making the optimal mathematical decision every single time. At this stage, the natural question becomes, why not just automate the entire process?

            Cue Libratus, the AI bot that shook the poker world. From January 11 to January 31, 2017, the bot played a tournament consisting of 120,000 hands against 4 of the best players in the world. Running on a supercomputer worth almost $10 million, Libratus had three main programs in its arsenal. Firstly, the AI was programmed with the basic rules of poker, but not much more. To learn the game, it played trillions of hands against itself, determined what did and did not work, and used that information to build strategies. Secondly, a piece of software called an “end-game solver” was implemented during the actual tournament. This was the meat and bones of Libratus. With every move made by the poker pros, it would recalculate – with extreme precision – the odds of making certain decisions, in order to give it the best possible chance of winning the hand. Finally, the aspect of the AI that took it to a whole new level was its analysis of all hands played after each day. If it found gaps in the humans’ strategy, it would adapt its own strategy to pressure those gaps more. On the other hand, if it discovered weaknesses in its own play, it would correct those the next day. Even with all these advantages on board, Libratus would have the daunting (and unprecedented) task of attempting to beat the best of the best.

Libratus won, and not just by a little. The tournament was played with imaginary money, but if it had been for real money, the pros would have lost around $1.7 million combined. Needless to say, this was a huge step forward in the progress of AI. Poker deals with imperfect information (there is no way to be exactly sure of your opponent’s hole cards unless you get to showdown), and that’s what makes this victory so remarkable. And scary. We so often correlate poker prowess with the human ability to intuitively “read” someone’s hand and, effectively, their mind, from their behaviours. It appears that AI is quite literally beating us at our own game. Is this the end of human dominance in poker, then? Did Libratus not only match, but also greatly surpass the human aspect of the game?

Perhaps Libratus bypassed it. You see, the misinformation component of poker creates a gap in the process of precise decision-making. Even after much of the gap is closed with calculations and strategies, the onus of the decision always falls back on human intuition: Does it feel right to call this bet on the river for all of my chips? Poker AI, however, doesn’t rely on such intuition. Through calculation, Libratus played the game statistically and mathematically, with absolutely zero emotion. Judging by the results, it seems that the humans’ use of intuition didn’t work out too well.

One way of looking at intuition is as an accumulation of experience. In poker, as familiarity grows with certain game situations, players effectively “feel” that one choice is better than another. One possible explanation is that there are subconscious signals indicating that, based on past experience, a certain action should be taken. In this case, Libratus simply had too large of an experience set (trillions of hands) for the humans to overcome. One could say that the AI system did in fact have an intuition, just in a tremendously precise, and non-emotional, variation.

What are the repercussions of this monstrous defeat at the hands of Libratus? Well, poker players needn’t be too worried, because the AI technology in play here is still far away from being able to succeed at full tables (the tournament against the 4 pros was heads-up only). That being said, online poker rooms need to continue to improve and refine their security systems against technological aids that give players an edge. On a much larger scale, though, consider this: if AI can beat us in card games, why can’t they beat us in financial transactions, or cyber warfare? Combine this possibility with an AI system acting in self-interest, and… let’s just say that losing $1.7 million from poker would be the last of our worries.

Glossary of Poker Terms

Bet: Put money into the pot

Call: Match money put in by other players to stay in the hand

Check: Stay in the hand without having to put in any money

Flop: The first three cards placed face up on the table for all players to use

Fold: Forfeit the hand without having to put in any more money

Hole cards: The first two cards dealt to a player

Raise: Increase the amount of money that other players have to put in to stay in the hand

River: The fifth and last card placed face up on the table for all players to use

Showdown: Occurs after the river – the player with the best 5-card hand wins

Turn: The fourth card placed face up on the table for all players to use

3-Bet: Re-raise after another player has already raised

Sources

Gershgorn, Dave. “How a Poker-playing AI Is Learning to Negotiate Better than Any Human.” Quartz. Quartz, 13 Feb. 2017. Web. 01 Apr. 2017. <https://qz.com/907896/how-poker-playing-ai-libratus-is-learning-to-negotiate-better-than-any-human/>.

Kemeny, Horatio. Personal interview. Mar. 2017.