Can AI Beat Poker Pros?

New developments in the field of artificial intelligence and machine learning continue to shake the world. In 2017, a technology made in the Carnegie Mellon University beat poker pros at Limit Texas Hold’Em thanks to its innovative principles of work. This was not only an important move for poker bots – technological advancements tested in the game can be applied to many situations that require effective decision-making.

Improvements of Poker Programs

The history of AI poker software dates back to 1984 when Orac was created for WSOP competitions. Until 2003, developers stuck to the chess methodology model, which couldn’t be that much effective as chess which, unlike poker, is a game with full information. During the last two decades, many poker bots were introduced at various competitions or even at casinos. Computer science departments from the leading universities have been trying to design a perfect solution, incomparable to human players. But it wasn’t until 2017 for a program to actually beat every professional poker player competing with it. AI called Libratus implemented the strategy dependent on previous plays and human behavior at the table – this became possible thanks to the deep neural network implemented in a program like this for the first time. The probability of people playing better than Libratus is estimated at around 0.0001%.

Strategy of Libratus

The reason why strategies based on chess had their vulnerabilities when applied to poker lies in bluffing. This is a purely human activity which, as it seems, can’t be copied and automatized. Libratus proves this impression wrong. The program is capable of analyzing the flow of the game by breaking it into smaller manageable parts and adjusting decisions in accordance with opponents’ moves.

This AI uses abstractions to make the process easier. Combinations that have a slight difference or similar bet sizes are grouped together. As for the human part, Libratus can compute solutions after analyzing the opponent’s move. Developers Sandholm and Brown call this “nested subgame solving.” While many people wonder how to win at poker using math smarts and psychological tricks, improved technology can already handle any situation thanks to deep learning.

So What?

Poker programs have lots of valuable benefits over people: they don’t get tired or emotional, they are indifferent to money and not afraid to take risks. With AI knowing how to read the opponents and identify their weaknesses, it’s possible to enter the era when no human brain can compete with technology.

It doesn’t say much about poker itself, though. The pleasure of this game is of social nature. Players enjoy sitting at the table, bluffing, and trying to predict the opponents’ behavior. Poker pros can be intimidated by the fact that some computer program outperformed the skills they’ve been developing for years. However, the invention of such smart bots will never stop the industry from growing. Poker events will still attract many visitors, as this is one of the most intriguing games which requires an array of talents.

What AI programs developments applied to poker can do is improve other domains that work in a similar manner. This technology can be used not only in imperfect-information games but in any situation where continual learning and decision-making are needed. Enhanced planning techniques or financial markets – the number of opportunities is huge. With an ability to bluff, such programs can help in a business negotiation in many ways. Just imagine the app able to negotiate the best price for a product – this is a reality that is likely to come in next years.