An AI called Lengpudashi, the latest Poker machine from Carnegie Mellon University, has comprehensively beaten a six-man team led by World Series champion Alan Du to win $792,327 in virtual chips in 36,000 hands over 5 days. 

Lengpudashi, whose name means "cold poker master", is the latest creation of CMU professor Tuomas Sandholm and Noam Brown his PhD student. It is the successor to Libratus, which as we reported earlier in the year, AI Beats Top Poker Players, took on and beat four world class poker professionals in a 20-day marathon match at Rivers Casino in Pittsburgh in which 120,000 hands were played.

The venue for the Lengpudashi match was a resort conference center on China’s Hainan island and it consisted on 36,000 hands over 5 days. The AI was challenged by Team Dragon, led by Venture capitalist and amateur poker player  Alan “Yue” Du who made history in 2016 when he defeated a field of 863 players to become the first Chinese player to win the WSOP gold bracelet.

Tuomas Sandholm confers with Team Dragon Poker player Kai-Fu Lee head of Sinovation Ventures, which organized the contest.

Noam Brown with Alan Du
Source: Sinovation Ventures

Alan Du, who explains that Poker is a popular game among venture capitalists because:

“every hand you play is like a venture, trying to assess risk and ROI,” 

took a new approach to playing against an AI - essentially trying to play it at its own game. His team included computer scientists and investors who tried to apply their knowledge of machine learning and game theory to their game play - which involves bluffing as well as card counting and other computation..

This sounds like good strategy given Professor Sandholm's assertion that Lengpudashi didn’t learn to bluff from mimicking successful human poker players and analyzing historical data, but from game theory, saying:

“Its strategies were computed from just the rules of the game."

Noam Brown bolstered this opinion saying:

“People think that bluffing is very human -- it turns out that’s not true. A computer can learn from experience that if it has a weak hand and it bluffs, it can make more money.”  

It seems that money is a great motivator for the sort of reinforcement learning that in taking place here.

Strategic Machine, the company founded by Sandholm and Brown, takes home $290,000 from this contest and will use it to continue develop AI technology to targets a broad set of applications: poker and other recreational games, business strategy, negotiation, cybersecurity, physical security, military applications, strategic pricing, finance, auctions, political campaigns, and medical treatment planning. 

Comments

Make a Comment or View Existing Comments Using Disqus

or email your comment to: comments@i-programmer.info