Artificial intelligence technology has advanced to the point of self-awareness, allowing AI constructs to systematically and categorically destroy the best poker players on the planet. A headline from a doomsday premonition perhaps? No. It is a real phenomenon – AI technology is now capable of beating the best human poker players, and the technology is only getting better. 

It comes as no surprise that ‘thinking and learning’ computer systems have already mastered the art of 2-player games including No-Limit Texas Hold'em Poker, Chess, Starcraft II, Dota, Heads Up, and Go. But now, this technology has taken a quantum leap forward in the 6-player format of the game, and it's already light years ahead of the competition. 

Over the years, artificial intelligence technology has come on in leaps and bounds. Nothing challenges AI more than games, particularly games with a skill-based element like chess and poker. During the course of a game, certain achievements serve as benchmarks to monitor progress. AI and poker interactions have been rather limited in recent years since the AI constructs have only been applied to 2-player poker games

Everyone knows that it is highly irregular for poker to be played with just 2 players. As such, multiplayer functionality is the norm and AI technology simply had to compete at that level to be relevant. Enter Pluribus – developed by Brown and Sandholm. Believe it or not, Pluribus fits that mold in every way.

This AI construct competed in some 10,000 hands of multi-player poker games, and the results are astonishing. This machine has learned the art of 6-player NLH by competing against 5 x world-class poker aficionados. The creators are not surprised by the results: the AI machine performed considerably better than its human competitors.

Dubbed ‘Superhuman Performance by Artificial Intelligence’, the latest rendition of AI technology now includes new factors, such as imperfect information, multi-player functionality, and randomness. By simulating real-world situations, artificial intelligence software has been programmed to assess a myriad of potential outcomes, probabilities, and analytical assessments. As a result, powerful new algorithms with wide-reaching applications are now available.


Pluribus is a poker bot designed by a team of world-class researchers from Carnegie Mellon University and Facebook Inc’s very own AI laboratory. The AI supercomputer went head-to-head against a dozen exceptional poker players in 2 unique settings. 

10,000 hands of poker were played across 12 days, and in one of the settings, the AI construct competed against 5 human players. In the other setting, one human player competed against 5 different versions of Pluribus. In the latter version, Pluribus constructs were not allowed to collaborate. The interesting part of this story is that the average hourly winnings amounted to $1,000, and the average winnings per hand amounted to $5. The researchers concluded that Pluribus is ‘hands down’ a runaway success. 

Just in case you were wondering what type of opposition Pluribus was coming up against, consider none other than 6-time WSOP champion, Chris Ferguson. In his own words, ‘Pluribus is a very hard opponent to play against. It's really hard to pin him down on any kind of hand.’

It's not only Ferguson who got the short end of the stick – other poker pros like Darren Elias (multi WPT title winner) also got his jacks handed to him by Pluribus. Even Michael "Gags" Gagliano – a multimillionaire poker player found himself on the losing end against the bot. 

Poker presents unique challenges to artificial intelligence technology, particularly when multiple highly-skilled opponents are competing against the AI technology. Many different variables need to be factored into the learning process. Emotional, cognitive, probabilistic, and random elements are continually at play, making it difficult to craft an algorithm capable of self-learning, improvement, and expert-level functionality.

poker AI

‘Predictably unpredictable AI construct – Great for Bluffing’

Several years ago, an AI system mastered the art of poker play in a 2-player game of Texas Hold’em poker. In the years since, dramatic advancements have taken place and now these computers are able to factor in incredibly complex elements. It's interesting how the scientists ‘taught’ these AI constructs to play poker. They teamed them up against one another and allowed them to learn accordingly. 

The training process was a runaway success, and the AI machinery is the safest bet that anyone on the rail can make. It is worth pointing out that it took just 8 days to create Pluribus with 512 GB of RAM and a 64-core server. The scientists cut down on the learning curve by removing virtually limitless possibilities of what players could do during the course of their games, to just 2 or 3 moves ahead.

It's astonishing that AI technology is capable of the human art of deception a.k.a. bluffing, and it does so with effortless ease. AI uses bluffing when it is the most opportune decision to make, given the range of outcomes that are possible. Is this the end of human poker prowess as we know it? This question is a nonstarter. 

From a purely scientific perspective, it is invariably true that machines can learn a lot quicker, compute a lot more information, and process probability analysis far more efficiently than any human being. However, humans are capable of learning too. Given that it is human ingenuity that programs the algorithms upon which AI systems like Pluribus function, we definitely owe ourselves some credit. 

It's unlikely that premier poker tournaments like the World Series of Poker (WSOP), the World Poker Tour (WPT), or the Australia New Zealand Poker Tour (ANZPT) will be allowing scientists to deploy the likes of Pluribus at their tables alongside human poker players. Sure, it may happen, but poker is a human game after all – isn’t it? Poker pros readily attest to learning from these poker bots. 

For now, poker players needn't be overly concerned about going head-to-head against AI software like Pluribus. The creators of this poker monster state that it is a static program, with no upgrades or updates implemented after its 8-day training period. That being said, there was never a question about its efficacy, or its relentless ability to consistently beat the best poker players and come out a winner.

Pluribus makes a strong case for advanced poker playing strategies and machine learning capabilities. One of the most notable characteristics to emerge from the use of this type of AI technology against human competition is the prevalence of Donk Betting on the part of the machine.

This phenomenon takes place when a player ends a round of poker with a call and begins the next round with a bet. By mixing up different types of strategies to confuse the competition, Pluribus sets the tone and other players are following suit. The fact that the machine is better suited to random play is interesting, since humans struggle with this aspect of the game.


Anyone looking for expert-level AI technology for Texas Hold'em Poker probably knows that DeepStack is a name not to be toyed with. This technology is capable of reassessing strategic plays after every decision that has been made. Among the many strengths of this algorithm are intuitive local searches, ongoing re-solving of problems, and what is known as ‘sparse lookahead trees’. 

An in-depth poker-playing study was conducted over the years and completed in December 2016. Some 44,000 hands were played and DeepStack’s complex algorithm defeated 11 poker pros. During the course of the games, the algorithm won 49 BBs (big blinds) of every 100. Other important stats to consider include the 4 standard deviations from 0, and had it folded it would have only lost 75 big blinds of every 100.

DeepStack uses what is known as heuristic search methods for games with imperfect information availability. By continually re-solving challenges through situations which arise on-the-fly, DeepStack quickly became a force to be reckoned with. By using intuition, artificial intelligence technology is fully capable of ‘Deep Learning’ through random poker plays. 

Such is the poker prowess of this technology, that of the 11 players completing 3,000 games with this technology, 10 of them were beaten by a statistically-sizeable margin. It’s worth pointing out that bluffing is entirely possible with AI poker bots. The founder of game theory, John von Neumann had this to say about bluffing: ‘… Real life consists of bluffing, of little tactics of deception, ask yourself what is the other man going to think I mean to do.’

The abstraction-based approach used by DeepStack is such that this AI construct works with the current situation and doesn't need to pull data from a repository of infinite possibilities. The low variance technique used to analyze DeepStack’s successes – AIVAT - provides unbiased performance criteria in as little as 3,000 games of poker. Of the 44,852 poker games played by dozens of players from 17 countries, DeepStack consistently outperformed its human competitors.


This is indeed a question worth asking since most of the literature points in the other direction. Poker bots learn from what human players are doing and adjust their gameplay accordingly. However, professional poker players have learned many lessons from poker bots. This is particularly true with respect to plays that human players simply don't make regularly. A classic example of this is ‘bet sizing’. 

The scientists behind the creation of Pluribus also created another poker prodigy in the form of Libratus. This bot decisively took down 4 poker professionals over the course of 120,000 hands of NLH in a 2-player version of the game. In the case of Pluribus, the constructs work with proven strategies that allow it to outplay its opponents time and again. By intentionally being unpredictable, poker bots engage in obfuscation techniques which humans can learn from. 

Programmers discovered that the algorithm requires 5 continuation strategies for each player to develop an overall strategy for playing them. By determining how it acts on every hand, given the strength of its hand at any given time, strategies are developed for all possibilities.

Humans can certainly commit these lessons to memory and employ them with increasing success rates over time. There is no doubt that machines are devoid of emotion associated with winning and losing, attachments to money, fear of making specific calls, or the excitement that may otherwise cloud one's judgment. 

A poker bot looks only at the current state of the game and how it can make the best decisions in order to consistently win. In terms of learning processes, Libratus required 15 million core hours to fine-tune its strategies with 1400 CPU cores. But the upgraded version, Pluribus required just a fraction of that to become the best multi-player No Limit Texas Hold'em Poker playing bot.


Different poker professionals have different opinions about poker bots and AI. One such poker professional – Daniel Negreanu – has taken an optimistic position on AI and the game of poker. His perspective is that AI gives you a zero-risk opportunity to learn the game of poker. Previously, poker players learned by making mistakes. 

Now technological advances in poker software make it easier for players to improve their game through AI. This all comes full circle to what artificial intelligence really is all about. In a sense, AI is about giving machines and software copious amounts of data and then using that data for problem-solving purposes.

Many poker pros are of the opinion that the psychological component of poker makes it difficult for any machine to understand the nuances of deception, chicanery, posturing, body language, and all behavioral and psychological elements of the game. Even then, machine learning has narrowed the gap and it is possible for AI technology to learn how players act under certain conditions. 

Patterns repeat themselves, and poker bots can easily identify different types of poker-playing styles. The technology has advanced to the point where there is no longer any question about the efficacy of poker bots. Now, players are learning from poker bots and their behavior to improve their own strategies.


It's really easy to play poker against a computer. The games play out as if you were competing against another player online. The Texas Hold'em game begins like any other, with a pair of hole cards dealt to you and the same to the poker bot (computer). Naturally, you can't see the cards that have been dealt to the computer. 

The option to check, raise, and bet is available at each stage of the game, with the Flop, Turn, and River. When playing poker against a computer, the games play out quickly, requiring attention to detail, particularly when playing for real money. Plenty of hands can play out every hour, and meticulous bankroll management is needed to keep things in check.


From a practical perspective, there are many more applications with poker bot-style software than meets the eye. Machines, particularly AI supercomputers are now being used in situations where there are competing interests and it is unclear how win/lose conditions play out. Most experts agree that the best way to test AI technology is the gaming arena. That's where humans are competing against machines in a battle royale for dominance. 

If machines beat humans, they are regarded as superhuman. While AI poker bots tend to focus on specific problem-solving elements such as chess or poker, the conceptual abilities of human beings extend beyond that to real world applications.

For now, the stage has been set for the next chapter in poker gaming: the rise of the machines. It's doubtful that the WSOP will be allowing poker bots to compete against thousands of human players for a berth among the poker gods, but never say never…

About the Author

With digital marketing strategies in his blood, Louis Wheeler has traveled around the world, exploring gambling cultures and gaining experience in casino games from 2003. If you are in a casino anywhere around the planet, you may find him right next to you, playing blackjack, roulette or texas hold'em. 

Related Content
How Can a Small Stack Player Win Big?

How Can a Small Stack Player Win Big?

6 Types Of Poker Games

Best 6 Types Of Poker Games

Poker VS Bridge

Poker VS Bridge

Irish Poker Rules

How To Play Irish Poker

It's Strip Poker Playtime

How to Play Strip Poker

Poker Buy-In: Everything You Need to Know

The Poker Buy-In: Everything You Need to Know