AI Poker Program Lengpudashi Beats Team Dragon

In January, a Carnegie Mellon University (CMU) created AI software program called Libratus thrashed four poker pros for $1,766,250 virtual dollars over a 20 day, 120,000 hand game of Heads-Up No-Limit Texas Hold’em. This month, six Chinese players subsequently took on another CMU poker program called Lengpudashi in a scaled down version of the original contest, with much the same result, dealing yet another blow to team human.
The latest contest took place between April 6-10 in the Chinese province of Hainan, and lasted a total of 36,000 hands, at the end of which Team Dragon found themselves stuck for $792,000 virtual dollars. Because of its smaller sample size, however, the competition does not hold the scientific merit of the previous Libratus battle, and as the bots’ developer Tuomas Sandholm explains:
“This is an exhibition, not a match, challenge or competition. We are running a relatively small number of hands, so this is not a scientific experiment.”
Nevertheless, the caveat will come as scant consolation for 2016 WSOP bracelet winner Alan Du and his team of poker players who not only failed to get revenge for team human’s earlier defeat, but also missed out on the opportunity to earn $290,000 in the winner-take-all contest.
As far as poker is concerned, one of the key past misunderstandings concerning AI intelligence has been that human’s held an advantage in being able to move their machine opponents off better hand by bluffing. Libratus put paid to that notion earlier this year, and as its co-developer Noam Brown, explains:
“A computer can learn from experience that if it has a weak hand and it bluffs, it can make more money. Its strategies were computed from just the rules of the game [not from analyzing historical data].
If that’s not troubling enough, the actual program itself could be bought for less than $20,000, and within 5 years could be “running on smartphones”, according to Noam Brown.