A third of the way through a 20-day man vs. machine poker tournament, the artificial intelligence system has the hot hand.
As of Wednesday at 2:30 p.m. ET, the humans and the A.I. system, dubbed Libratus, had already played more than 34,000 hands with about 120,000 hands likely by the end of the tournament.
[To comment on this story, visit Computerworld's Facebook page.]
The "Brains vs. Artificial Intelligence: Upping the Ante" tournament kicked off Jan. 11 at the Rivers Casino in Pittsburgh. During the tournament, poker pros Jason Les, Dong Kim, Daniel McAulay and Jimmy Chou are playing Heads-Up No-Limit Texas Hold'em against Libratus.
Libratus pulled ahead early, leading by a little more than $74,000 on the first day of play and by more than twice amount by Day 2.
"This is quite nice given that in advance of the event the international betting sites considered us a 4:1 or 5:1 underdog," wrote Tuomas Sandholm, a professor at Carnegie Mellon University and lead developer on the Libratus system, in an email to Computerworld.
By the seventh day of the tournament, Libratus had increased its overall lead to $231,329 in chips.
"My assumption would be the longer the game goes on, the more information the A.I. gets and the better it becomes," said Zeus Kerravala, an analyst with ZK Research. "For humans, poker is a combination of skill, intuition and emotion. With the A.I., it's based on learned information and data.... Poker is a good game [to test A.I. against humans] because you play the other players as much as you play the cards. Black Jack is more a pure numbers game, so it's not as good a "human versus machine" test."
According to Sandholm, poker has been an accurate measure of the power of A.I. since the earliest days of the technology because it's seen as even more complex than playing chess or the board game Go.
"Poker poses a far more difficult challenge than these games, as it requires a machine to make extremely complicated decisions based on incomplete information while contending with bluffs, slow play and other ploys," Sandholm said in a statement.
This is the second time that a Carnegie Mellon-built A.I. system has taken on human poker players. In 2015, the university ran the first Brains vs. A.I. contest using a different A.I. system. That first one was called Claudico, which Sandholm built as well.
Claudico did not win that challenge and collected fewer chips than three of the four professionals it played.
Kerravala, though, has high hopes for the A.I. system to come out ahead of this second ternament.
"I'm more surprised the computer has lost some ground but that may be the pattern as it learns how the others are playing," he said. "Over a long period of time, I would expect the A.I. to win. The longer the game goes on, the more data the A.I. has and the scales tip in its favor."