It is the second time in 2017 an AI program beat competitive poker players.
For the second time in 2017, poker-playing artificial intelligence (AI) crushed human poker players. The poker AI “Lengpudashi” or “cold poker master” beat five Chinese poker players in a “landslide” over 36,000 hands to win the $290,000 winner-take-all prize.
Lengpudashi is an upgraded version of Libratus, which was created at Carnegie Mellon University (CMU) and in January 2017 won 1.7 million play chips against top poker pros. An earlier version of CMU’s poker-playing AI called Claudico lost to humans in 2015.
The human team that beat Claudico included Doug Polk, who is regarded by many as the top heads-up poker player in the world. Lengpudashi, on the other hand, beat a team that consisted of Yue Du, an amateur poker player, engineers, computer scientists and investors who attempted to use game theory and machine intelligence to beat Lengpudashi.
Poker AI Getting Better
We’ve long been fascinated with the idea of computers beating humans at certain games. And over time that’s come to fruition; IBM’s Deep Blue supercomputer beat chess world champion Garry Kasparov in 1997; IBM Watson in 2011 defeated Jeopardy champions Brad Rutter and Ken Jennings; And in 2016 Google’s AlphaGo program beat Lee Sedol in 4 out of 5 Go matches.
Must-Read: Top 5 AI Gaming Machines of All Time
It’s taken a lot longer for poker-playing AI to catch up. Poker is what computer scientists call an “imperfect information game” as unlike chess and Go, not all the playable pieces are visible on the board. This means AI has to rely on complicated betting strategies and bluffing, and be able to spot when opponents are bluffing.
Libratus co-developer Noam Brown now says its clear people misunderstand what computers and humans are each good at. “People think that bluffing is very human — it turns out that’s not true,” he says. “A computer can learn from experience that if it has a weak hand and it bluffs, it can make more money.”
Tuomas Sandholm, a CMU professor of computer science who created Libratus says, the AI didn’t learn to bluff from mimicking successful human poker players, but from game theory. “Its strategies were computed from just the rules of the game,” not from analyzing historical data.
Training AI to beat humans at games like chess and poker helps hone an AI’s reasoning skills and strategic decision making.