2017 Was the Year AI Defeated Humans at the Poker Table

This 2017 was a pivotal year in the AI (Artificial Intelligence) field. It was the first year that marked an actual clear and decisive defeat for humanity at the hands of AI. The battle was not fought online or with futuristic weapons, but on the battleground of a poker table, with playing cards and chips. Poker sites detailed analysis of the AI poker phenomenon tells us that, in early 2017, the most advanced PokerBot ever created took on four of the world’s top poker stars and defeated them all. The game was a long drawn out affair, done to test the AI named Libratus, and it involved a total of 120,000 hands of poker. The defeat was not subtle and was not, by any means, a close match. Libratus won a total of $1.7+ million and was able to maintain an average win-rate of $14.72 per hand. What is even more amazing is that not only is Libratus already likely the best poker player on Earth, but it also learns as it plays and is only getting better. Here is a look at all you need to know about AI poker-bots and how they have managed to defeat their makers, us mere mortals.

The History Of Poker AI

Before we get into the newer range of advanced poker bots, it is important to note that poker AI technology has been under development for decades. The earliest poker AI software was created in 1984 by Mike Caro and was used to train for the WSOP (World Series of Poker). By 1997, the University of Alberta (UoA) released its first poker-playing AI, a poker bot named Loki, who was still light-years behind what we would consider a threat. The problem was that, up until this point, poker AI was developed on a chess methodology model that failed to take into account the fact that poker was not limited to just strategy. Poker involves bluffing and reading other opponents.

In 2003, developers began to move away from this chess-based model, and the true AI poker bot revolution began. By 2005, the first world series of poker robots was held at the Golden Palace casino in Nevada, and in 2006, the ACPC (Annual Computer Poker Competition) was founded and played. The first real glimpse of poker-bot potential was in 2008 when UoA’s follow up to Loki, a poker-bot named Polaris, played heads up against 6 humans and was able to beat 3 of them, draw with 1 and losing to only 2. Introduced in 2015, UoA’s next generation poker-bot, Cepheus, was another game-changer, as it finally solved Limit Hold’Em and was able to see near-perfect results. Finally, in 2017, Libratus, designed by the AI research and development team of Carnegie Mellon University in Pittsburg, crushed some of the best players in the world.

Libratus Poker Bot

Libratus, unlike its less-advanced predecessors, is a real AI that sets its own strategy and can adjust its gameplay to defeat anyone. Basing its strategy on 15 million core hours of computation on a supercomputer, Libratus is self-taught and has developed its own strategy. Libratus is 30,000 times faster than the average modern desktop computer and has over 274 terabytes of RAM (Random Access Memory). The world’s most advanced poker-bot is not cheap, however. As of right now, it costs around $9.65 million to create.

Libratus is different from earlier AI in a few key ways. First, instead of a fixed, built-in strategy, Libratus relies on an algorithm that computes its own strategy. Secondly, Libratus analyzes its own play every night and corrects any mistakes, playing differently each day. Third, Libratus is capable of randomizing its strategy to make its opponents uncertain of when it is bluffing. Fourth and finally, it is able to identify weaknesses in specific opponents, changing its style in order to exploit those weaknesses. In simple layman’s terms, Libratus is self-taught, its strategy is independent of human play and it evolves to better and better with each interaction. Scary? Maybe a bit.

Advantages AI Has Over Us

When it comes to Poker, Libratus and similar AI have some key advantages over human players. For one, AI does not get tired, need food or nourishment and has no bodily functions that can throw it off its game. Secondly, AI has no feelings and doesn’t have any attachment to money or feel its value. Additionally, it cannot get mad or feel tilt and isn’t afraid of risk. Furthermore, an AI as advanced as Libratus can ram 20,000 human years of work into a single week. This, in itself, means that an AI that can adjust its strategy can simply play more and practice more in a shorter time than a human ever could.

Is Libratus Alone?

Though many would assume Libratus is the only AI poker-bot that is advanced enough to defeat a human, they would be wrong. According to the data, a few different AI research teams have all created advanced AI’s that are capable of beating a human pro. As of 2017, Carnegie Mellon University and the University of Alberta both have multiple poker-bots that are capable of defeating professional players. Aside from Libratus, another highly advanced poker-bot and likely the second-best player in the world is another AI named “DeepStack”.

The Future Possibilities

AI is now clearly the best poker player on the planet, and the truth is, the AI revolution is not just limited to poker. Considering that electronic circuits function a million times faster than biochemical ones, AI is a growing trend that will delve into a plethora of sectors and industries. For example, in the cyber security sector, AI is expected to boost big data, intelligence, and analytics-spending to $96 million by 2021. Technology markets, government, defense sectors and banking have already begun implementing AI and that’s just the tip of the iceberg. According to Libratus’s co-creator Noam Brown, a perfect poker-bot might be able to defeat Libratus even more spectacularly than it was able to beat humans, and we will likely see it within the next decade. Regardless, the first epic battle between AI and humanity is over and we have lost. Even worst, we will continue to be out-paced as AI continues to grow and evolve. 2017 might be the first year of human defeat at the hands of AI, but that will only continue exponentially as AI and technology continue to grow unchecked.