Can you imagine that there was a machine capable of defeating the best poker players simultaneously? No need to keep imagining, in 2019 a poker player vs AI experiment was carried out in which an artificial intelligence (AI) was able to beat 5 professional poker players in the Texas Hold’em mode without limit. Some of the participants were Darren Elias, a professional player who holds the record for winning titles on the Wold Poker Tour, Chris “Jesus” Ferguson and Jason Les. They were all experienced players with nearly a million euros won through poker games.
This artificial intelligence machine is called Pluribus and is a renewed version of another machine that was created in 2017 for the same purpose. Here we explain everything that has to do with AI in poker and the latest experiments. Let’s get to it!
Poker vs. AI: What is Pluribus and who developed it?
Pluribus is an experiment that was conducted by a team from Carnegie Mellon University (USA) in collaboration with Facebook and the U.S. Army Research Office. This experiment was conducted by researcher Tuomas Sandholm and Noam Brown (Facebook).
The aim of this experiment was to develop an artificial intelligence capable of winning in poker competitions with multiplayer tables. In other words, create an AI that was smart enough to manage scenarios with a lack of information, as is the case with poker, and beat the opponents in spite of it. Something they achieved in 2019, when the research was published in the journal Science by those responsible for this project – and which involved a great deal of rejection and aversion to such research.
A total of 5 players took part in the poker vs. AI (Pluribus) experiment, in addition to the machine – all of them professionals in the world of poker betting, with several records at international level. A total of 5,000 hands were played and of these Pluribus was the main winner, with the most money being raised. The experiment was repeated with 13 other players in which case 10,000 hands were played. Similarly, the result was favorable for the machine, which managed to win by far.
What was proven after that is that the algorithm of this artificial intelligence learned as it played. In poker, artificial intelligence -or Pluribus-, demonstrated that it knew how to fool the players perfectly -according to poker patterns, and that he used strategies such as face betting or ending the rounds with call and starting betting. Several of the poker professionals who competed against her explained that it was impossible for them to find the machine’s weak point and that the machine was bluffing them as a strategy.
Poker: Artificial Intelligence since 2017
Pluribus is not the first AI to compete and win against poker players. In 2017 another poker vs. AI experiment was carried out by the same developers as Pluribus. The machine, in this case, was named Libratus. Unlike its predecessor, Libratus was a machine composed of a supercomputer, while Pluribus only needs a server to run.
Although it participated in different tables and consolidated itself as the main winner, the truth is that Libratus used a much simpler strategy, since it only competed in tables of two players. That is, the probability of success was much higher. The Libratus algorithm, explained Brown and Sandholm had three main phases, although all of them were based on the strategy of zero games, in which the gain of one is the loss of another.
What has happened today with Pluribus?
The truth is that the results of Pluribus were so decisive that the researchers decided not to publish their code. This poker algorithm, they explained, could be very detrimental to online poker houses. It was therefore decided that this poker and artificial intelligence experiment would not be sold so as not to harm anyone. Developers are currently working in the field of innovation and artificial intelligence. Noam Brown is part of Facebook’s Artificial Intelligence department, while Tuomas Sandholm has his own company called Strategy Robot, which focuses on marketing government and security applications with a particular focus on military strategy and tactics, as well as intelligence and cyber security.
On the other hand, artificial intelligence follows another path of development so that it can be applied in situations that facilitate routines, as well as to develop some tasks that people cannot do at certain times.