Updated December 12, 2025
The rise of AI was one of those deliciously terrifying themes that dominated TV in the Eighties and Nineties, growing up. Many young kids spend hours convincing their mums to let them stay up and watch The Terminator.
Fast forward to today, and as AI moves from science fiction into everyday reality, the idea that machines might one day outgrow their makers is creeping back into the poker spotlight.
But what does a sentient bot that can execute the perfect poker bluff mean for the game?
Let’s delve into the human-verses-AI debate in poker games.
Human vs Machine
The war between humans and machines might never happen, but humans have been keen to test their wits against AI for a long time, nonetheless.
One of the first significant efforts came from IBM in the 1980s. The company developed a computer that defeated some of the world's best chess players.
In 1996, their bot, Deep Blue, battled Russian chess grandmaster Garry Kasparov. Arguably one of the greatest players of all time, Kasparov edged the computer, winning four of six matches and losing two.
It took just one more year of development for Deep Blue to finally get the better of Kasparov, however, winning with 3 wins and 1 draw in 1997.
These days, your mobile phone software is powerful enough to host a sufficiently intelligent bot to beat any chess player – and even challenge the best poker players of all time!
With chess in the books, attention turned to poker. Poker is a game of incomplete information, which makes it much more complicated than chess.
The New York Times noted that in poker, “a computer has to interpret its opponents’ bets despite never being certain what cards they hold,” which isn’t the case in chess, where the board is set at the start of your turn.
Some believed that a computer bot would be unable to overcome human creativity and unpredictability.
Though this belief may have been true in 1997, the surge in computer power has led to the emergence of unbeatable GTO solvers, which are often (ironically) caught because of their superhuman creativity.
And they were right... kind of. Early poker bots were easy to beat and full of glitches. Since non-AI bots were unable to think on their feet, they could be crushed as soon as anyone found a weakness or imbalance in their strategy.
As technology evolved, however, bots became more powerful, and many believed it was just a matter of time before the machines came out on top.
Polaris Sets a Poker Versus Computer Trend
One of the significant breakthroughs in bot research occurred in 2007 at the University of Alberta in Canada. Researchers used several poker bots, pre-programmed with fixed strategies.
Over two days, Polaris battled against poker pros Phil Laak and Ali Eslami at the Hyatt Regency Hotel in Vancouver, British Columbia. The challenge included four duplicate 500-poker hand matches. To help minimise variance, card distributions were identical.
The goal was to reduce the impact of luck, which Laak had attributed to his 2005 victory over Vexbot, an earlier version of Polaris.
When the dust settled, the humans took the W, with two wins, one draw, and one loss. Like with Deep Blue, it took Polaris a year to turn the tide, and in 2008, an improved version competed in the Second Man-Machine Poker Championship in Las Vegas.
Against new human opponents, Polaris won the match overall, with three wins, two losses, and a tie, for a win rate of a little over 3bbs/100 hands.
Though these are pretty solid numbers, and no indication of EV win rate, it’s hard to say for sure if the victory was due to Polaris’ evolution, variance, or the skills of the new human opposition.
Nonetheless, the computer’s success was a sign of things to come, and the rapid development of poker AI continued.
Claudico Takes a Crack
Next up was Claudico, designed by Carnegie Mellon University professor Tuomas Sandholm and a team of graduates.
Sandholm noted that “Poker [became] a benchmark for artificial intelligence research, just as chess once was,” and designed Claudico to learn independently rather than follow reprogrammed strategies.
To do this, Claudico needed a lot of digital brainpower to run; a supercomputer and 16 terabytes of RAM, to be precise. It wasn't quite Skynet, but for 2010, it wasn't far off!
In July 2014, Claudico won a poker tournament against other computers and was ready to take on a team of human players in 2015.
The team was formidable, including some of the best heads-up players in the world at the time: Dong Kim, Jason Les, Bjorn Li, and Doug Polk, eyeing up a share of the $100,000 put up by Rivers Casino and Microsoft.
The televised action saw the team battling Claudico for a gruelling eight hours each day, playing 80,000 hands in total. This sample became the largest human-versus-computer ever recorded.
In the end, the poker players won the match comfortably, with Polk noting that the AI programmers still had work to do based on how strangely the program played.
Interestingly, Polk told PokerNews how Claudico “would sometimes bet a miserly 10 per cent or an over-the-top 1,000 per cent.”
In hindsight, it seems Claudico was on to something. Given the poker probability odds, these kinds of big over-and-under bets have become routine in the modern game.
Libratus Changes the Game
In 2017, the computer science department at Carnegie Mellon University took another shot with an improved bot called Libratus.
Lined up against it were Jason Les, Dong Kim, Daniel McAulay, and Jimmy Chou. With members of the team already beating Claudico, spirits were high, but this confidence was misplaced.
From the outset, Libratus was on another level than its predecessor. Its “brain” was a small army of 600 machines, each with a 28-core processor, all backed by 2.7 petabytes of data (whatever that means!).
Its MO was different, too. Instead of being trained in poker strategies, Libratus was programmed to solve games with incomplete information by using raw computational horsepower to run endless simulations until it figured things out for itself.
Immune to fatigue, mis-clicks or selective memory, Libratus’ raw, tireless power allowed it to study, play and analyse simultaneously.
Libratus mowed down its opponents.
Though it required unbelievable computer power, Libratus was a step forward for AI.
Pluribus and Poker
In 2019, Carnegie Mellon researchers were back at it, with Facebook as a partner. They were hoping for success in multi-way poker games, which are way more complex than the previous bot heads-up games.
The Wall Street Journal noted that, “multi-player poker is considered less a game than an art form that requires a multitude of skills, especially the ability to read human interactions and leverage that knowledge to exploit mistakes and weaknesses.”
The goal was for AI to beat a standard game of poker featuring a table full of the world's top players.
Despite the increase in game complexity, the players couldn't match the machine's and software's data sets and analyses.
“A good AI has a ridiculously unfair advantage against humans: they don’t get tired,” Pluribus player and long-time pro Michael Gagliano told the Journal. “They don’t get hungry. They don’t deal with emotions.”
In the end, Smithsonian noted that Pluribus “won an average of around $5 per hand, or $1,000 per hour, when playing against five human opponents.”
MIT Joins the Poker-Bot Party
Carnegie Mellon and Alberta aren’t the only ones putting poker to work. MIT runs its own annual poker-bot tournament, where small student teams spend a month building an autonomous bot, then unleash it in a digital free-for-all - sort of like a poker cheat sheet.
Due to its complexity, organisers frame poker as the perfect test of decision-making under uncertainty, a skill that is key to various walks of life.
The students mix economics, maths, and computer science to build some formidable bots, but thankfully, you won’t see them at your poker tables. Poker sites ban all bots instantly, and the MIT bots exist purely for research.
It’s a flex of AI, programming, game theory, and a nerdy test to see who can hack together the smartest virtual shark.
The competition gets surprisingly fierce, too, with sponsors offering juicy prize money to attract the most elite entries.
Does AI Have the Advantage?
So, the big question is: Does AI have an edge in poker?
Well, the short answer is, yes. In addition to their immense processing power, computer bots solely focus on making the right decision.
- They don’t tilt
- Have no concept of pressure or stress
- They have perfect memories
- They give off no physical tells
- They don’t get tired.
Humans face real consequences; bots don’t.
But while bots may always win the game, they have no way to gauge their success or even enjoy it. Poker is still a human game. The thrill comes from beating real people and winding them up.
When all is said and done, if you can’t enjoy winning, are you even a winner?
