In a feat reminiscent of the controversial victory by supercomputer ‘Deep Blue’ over world chess champion Garry Kasparov, a computer program has managed to beat a string of professional poker players at the game.
DeepStack, as it was called, defeated 10 out of 11 players who took part in a total of 3,000 games as part of a scientific study into artificial intelligence.
The 11th player also lost, but by a margin that the researchers decided was not large enough to be statistically significant.
This is not the first time a computer has won at poker. Libratus, a program developed by Carnegie Mellon University academics, won $1.76m (£1.4m) from professionals in January, for example.
But the researchers said DeepStack's performance represented a "paradigm shift" in AI that could have implications for the defence industry and medicine.
One player who took on the algorithm, Irish professional Dara O’Kearney, said it felt like he had been playing a human who was “a bit better than me, but not massively better”.
He warned there was already “a lot of evidence” that bots were winning money from human players in online poker games.
Writing in the journal Science, the researchers, from Alberta University in Canada, said: “Artificial intelligence has seen several breakthroughs in recent years, with games often serving as milestones.
“A common feature of these games is that players have perfect information. Poker is the quintessential game of imperfect information, and a longstanding challenge problem in artificial intelligence.
“In a study involving 44,000 hands of poker, DeepStack defeated with statistical significance professional poker players in heads-up, no-limit Texas hold’em.”
This type of poker involves just two players, the computer and the human in this case.
The researchers said DeepStack had been able to win despite being given no training from expert human games.
“The implications go beyond being a milestone for artificial intelligence,” the Science paper said.
“DeepStack represents a paradigm shift in approximating solutions to large, sequential imperfect information games.
“With many real world problems involving information asymmetry, DeepStack also has implications for seeing powerful AI applied more in settings that do not fit the perfect information assumption.
“The abstraction paradigm for handling imperfect information has shown promise in applications like defending strategic resources and robust decision making as needed for medical treatment recommendations.
“DeepStack’s continual re-solving paradigm will hopefully open up many more possibilities.”
Dara O'Kearney, an Irish poker professional who completed 456 hands, told The Independent that DeepStack played in a style similar to one used by some human players, based on game theory.
“I would say there wasn’t a massive difference. If I hadn’t been told it was a computer, there was nothing it was doing that would have tipped me off that it was a computer,” he said.
“I felt I did pretty much okay, but … I did feel the computer was a bit better than me, but not massively better.
“Heads up, no limits poker is not my speciality. It’s possible a human who specialises in that might do better.”
However, he suggested things might have gone differently if there had been real money at stake.
“We were playing for play money – that maybe skewed the results slightly. I’m used to playing for money … and I probably don’t play my best game at the end of the day if I know if I make a mistake, it isn’t going to cost me frankly,” Mr O’Kearney said.
He suggested DeepStack’s success was “more significant” than Deep Blue’s because there are more variables in poker than in chess.
“I guess all of these things are one more further proof that AI is better at almost anything than humans,” he said.
Poker, he said, was more complicated for a computer to master than chess because of the greater number of possibilities of different situations.
And playing a more traditional version of the card game with more than just two players would increase the complexity markedly.
“I don’t think that will ever be solved because the number of possible situations is greater than the number of atoms in the universe,” Mr O’Kearney said.
“It [the computer] would not play perfectly – but that’s not to say it wouldn’t play better than the best humans.
“It’s likely that a computer that was dedicated to that sort of thing would play better than 99.9 per cent of human players, but I still suspect the very best human players would remain better.
“But that might be a human fallacy on my part.”
Asked whether computer programs were a problem for the online game, he said: “That’s a massive concern.
“There’s already a lot of evidence that bots have been programmed to play and are already winning money off human players.”