Advances in artificial intelligence applied to poker have allowed the development of computer tools capable of handling variables in a poker game and many players are still skeptical about its effectiveness or the scope of its use in a game.

Although AI in poker is still under development and is expected to expand its scope of use with new functionalities, in certain games artificial intelligence and poker are limited in their capabilities by the restrictions on their use.

Can AI compete with professional players?

Year after year, technological advances touch various areas of computing, and the development of new forms of artificial intelligence has made it possible to optimise a programme capable of competing head-to-head with an expert player. There are several artificial intelligence programmes for poker that have proven to be competent and capable of pitting a professional poker player against the AI. But what are these programs and what makes them so competitive?


If we talk about a pioneering poker AI in terms of competitiveness and functionality, we have to mention Libratus. It was developed by Tuomas Sandholm and was one of the first programs to make duels of expert poker players vs AI viable. It was, of course, developed for one-on-one (i.e. two-player) games, but it laid the foundation for what would later become a reality: that technology can be used to develop intelligent behaviour capable of beating a human in a mathematical game.

Sandholm spent more than two decades dedicated to creating a computer tool that could take on a human, and so Libratus was born, eventually becoming an advanced poker and AI solution that earned Sandholm international recognition. Thanks to this push, he was awarded numerous prizes and in 2017 he set up Strategic Robot, his own computer development company with competencies in the field of military strategy and cybersecurity.


Pluribus is an artificial intelligence that has been optimized to match the most experienced poker players on a level playing field, and has already proven that it can beat them in a game without cheating or loopholes. While many players see Pluribus as a very competitive poker simulator, the AI is actually designed not to tailor the games to the player’s needs, but to beat them, which makes this program extremely competitive.

It was developed at Carnegie Melon University in the US, and differs from Libratus in that it can compete in five- or six-player games. During the optimization process, its creators pitted it against replicas of itself, allowing the AI to develop an action plan that it then improved autonomously when pitted against real players. Scary, isn’t it?

By creating games of real poker players vs. the AI, the results began to blossom: in most cases, Pluribus beat professional Texas Hold’em players over 10,000 card hands in 12 days, establishing itself as the most advanced technological tool of the moment.

Differences between Libratus and Pluribus

Although both programs have been designed with the same premise and are able to function correctly in a poker game, the AI of one and the other actually has different capabilities, which means that, for example, Libratus can only perform in 1 vs 1 games, while Pluribus can play with up to five other players at the same table.

While Libratus is good for simulating one-on-one confrontations or for getting an idea of the possibilities during a showdown with only two survivors, Pluribus handles many more variables. It’s a program that takes into account luck, bluffing and player psychology, allowing you to gradually improve as you play and beat more players. And this is a symptom of the fact that human behavioral patterns are very well defined and an AI developed by us is able to see what we do not yet perceive. Disturbing to say the least.

The key thing that makes Pluribus so special is that its strategy is not based on analyzing all possible moves, but on simplifying all options as much as possible. This allows the system to focus on potentially winning plays, increasing its probability of success. Pluribus has a limitation that has proven to be key to its tremendous advantage during the games: its algorithm has been designed so that it is not possible to focus on future moves, making a step-by-step reading of the game. And the most surprising thing comes when you analyse the Pluribus patterns: among all the variants that players handle in each hand, Pluribus only needs five to achieve victory. That’s tremendous.

Another point in its favour that makes Pluribus so special is that it has demonstrated an enormous capacity for abstraction. It is an AI that handles blocks of information, not isolated concrete data, which makes it work with more limited variables. It applies this to plays and strategies, grouping all possibilities into blocks of data to simplify the play and increase its chances of success.

Not to mention that an AI has an advantage per se, and that is that, as it has no physical presence, it is impossible to make a reading of the possible variables it is handling. This, in the words of players who have faced this AI, is one of its greatest assets: it is practically impossible to find a pattern, being one of the most infallible and unpredictable opponents that any poker player can face today.