Is AI Calling Poker's Bluff?
Alex Hudson
Executive Editor | Technology journalist and correspondent | AI expert | strategy leader | Web3 innovator | Newsweek | BBC | Metro
Poker, the game James Bond seems to be so good at, relies on “imperfect information” and a combination of psychology, math, game theory, and luck. But new AI software is looking to “solve” a game that used to revolve around looking into somebody’s eyes to tell if they were bluffing.??
“There’s so much luck in a poker tournament,” the new World Series of Poker ( WSOP ) Main Event champion Daniel Weinman said. “I thought I played very well, but there were so many hands that I got incredibly lucky.”
The most famous poker tournament with a $10,000 buy-in had a record 10,043 entries and a record prize pool of $93,399,900. And Weinman shouldn’t have won, in all probability (then again, nobody should, more on that later.) He’s over $12m richer than he was a few days ago.?
In the most viral hand of the whole tournament, he had an 8% chance to win with two cards to come. In layman's terms, the worst hand when all the money went in won. PokerNews called it “one of the most pivotal suck-outs in poker history.”
This gets to the age-old argument of whether poker is a game of luck or skill. Most experts believe it is both but over a long enough sample size (poker pros start believing results after 10,000 hands and cement theories after 50,000 then 100,000) the more skillful players will win.?
But now, with artificial intelligence, machine learning, and a new “solution” to poker, are humans who play more like machines better at the game, just like chess, or is there still a large dollop of human intuition required to win??
For reference, the poker we’ll be talking about is no-limit hold-em (NLHE), the most popular type and the one played in the WSOP Main event as it makes the maths easier. There are many different forms of poker with many different strategies attached. Limit heads-up hold-em, for example, is considered a solved game.?
“Can computers play poker better than humans? Yes, but I'm not sure that matters all that much, especially for live poker,” PokerCoaching.com founder Jonathan Little tells Newsweek.?
“What the best players try to do with a solver is not to remember or replicate exactly what the solver does. They try to figure out general rules that can be taken from the solver and then they use that whenever they are playing.”
This “solver” is known as a “Game Theory Optimal” strategy, or GTO for short. It’s a computer program that provides a chart for every single poker situation imaginable. And now, with AI involved, it’s learning more about the tiny nuances of the game. If you say “GTO” to a poker player, they’ll likely either roll their eyes or tell you an increasingly technical and boring story about all the charts they had to learn.?
If you’re playing Tic-Tac-Toe or Connect Four, the GTO strategy can be memorized quickly. It is impossible to beat a machine at Tic-Tac-Toe, for example. With poker, particularly when bet sizes can vary dramatically, the possibilities are close to infinite. It also becomes more difficult to solve because there are often more than two players and not all the information (the other players’ cards) is available.?
“Every good online player for the most part has spent a decent amount of time playing around with solvers and studying them a lot,” Little says. “That's almost a prerequisite in today's game. It’s just what everyone does if you want to be anywhere near decent at poker.”
The idea is that every possible situation in poker has a “correct” solution that means, over the long term, it is impossible to lose money. The detailed explanation of how this works is best left to poker sites, but it has led to people wondering if NLHE poker has become a “solved” game.?
The shortest odds on the best poker players in the world were +6,000 (or 60/1) with bookmakers and even that, people I’ve spoken to say, was much too generous to the best players. While nobody would be drawn on specific numbers, +50,000 (or 500/1) or higher in a field of 10,000 was seen as more sensible (though poker players love discussing odds. Disagree? Get in touch.)
As a guide, men’s marathon world record holder Kenyan Eliud Kipchoge was - 200 (or 1/2) to win the 2023 Boston Marathon before the race started. The clear difference is that there’s no gamble in marathon running. No matter how good you are at poker, the cards can still conspire against you. Jonathan Little is one of the top 200 on poker’s all-time money list but hadn’t cashed (generally tournaments pay the top ~15%) in the 95-tournament WSOP until the very last event.?
领英推荐
“The nice thing about live poker is that sometimes you can just look and tell someone's bluffing if they are not a very serious player and they don’t have a good poker face,” Little says. Outside the U.S., online poker dominates live poker in the amount of money won and lost (and while important, we’re not getting into gambling addiction. A problem gambling helpline is here.) But because online poker is banned in most states, live poker still dominates proceedings.
Here still, the gap between the “maths” and study of GTO online poker has shrunk. In a recently streamed high-stakes battle between two of the all-time great players— Daniel Negranu and Phil Hellmuth, charts were printed out on hand to help, personalized for their opponent. One player at a recent World Poker Tour event was heavily criticized for a GTO solver at the table as he played.?
That said, live poker — except for a couple of high-profile alleged examples — is generally free of “cheating” where AI is used to “solve” every situation. Online, it is more prevalent, and big poker sites are looking to crack down on any player deemed to have an unfair advantage.?
“If you play too close to perfectly, they will ban you,” Little says. “[The banned players will say] they’re getting banned because they’re good. That’s not what it amounts to. Poker is a super-duper difficult game, and if you play too close to perfectly, you have to be cheating.”?
Ali Imsirovic was considered one of the best poker players in the world until he was banned from several sites. Amongst the numerous allegations, he allegedly used real-time assistance from artificial intelligence and GTO solvers to give himself an advantage, claims he denies.?
He recently admitted that some of the allegations about him were true, but the GTO use was “impossible” in the circumstances he found himself in. Imsirovic has not responded to a request for a response to these allegations.??
So can the top players be beaten by an AI and/or a GTO solver? In 2019, an AI called Pluribus defeated games of six-player poker, defeating 13 pros, all of who’ve won more than $1 million.
But Little says that winning at poker is much simpler than all of these solvers make out:?
“To win at poker, all you’ve got to do is three things,” he says. “One, find a game you can beat. Two, Play it a lot. Three, keep a proper bankroll. And if you do that... you'll win a ton of money whether or not you're good at poker in general.”
And this comes down to understanding who the best poker player in the world is.?
“Is the best player [the one] who knows GTO strategy the best? Is it the player who wins the most money? Is it the player who wins the most tournaments? It's a question that does not have a clear answer.”?
For now, Daniel Weinman will take his $12.1 million winnings as evidence that he’s both lucky and good.?
The questions we still don’t have answers to after researching this article: (all musings/questions/opinions welcome to [email protected] if you think you know the answers)?