Facebook’s new poker-playing AI could wreck the online poker industry—so it’s not being released | MIT TECHNOLOGY REVIEW


The world of bluffs and tells is the latest game to be dominated by an AI system. Algorithmic victories are well known in chess and GO but until now AI play has been limited to two-parties, methodical rules, and transparent play. A new AI model called Pluribus is winning at Texas Hold’em, among the most popular forms of poker.

Rules are still at the centre of poker but so is concealment and deception. Unlike games where opposing plays are shown, poker players use feints as strategy while keeping their cards to themselves. It’s made more complex with six sets in the mix.

Pluribus is showing poker prowess of its own. It is completely immune to bluffing and has no intention-revealing ‘tells’ for fellow players to weigh. All the while, the MIT TECHNOLOGY REVIEW says Pluribus ‘bluffed like a seasoned pro.’

Pluribus Poker AI: Sample poker hands vs. professionals

Pluribus is the first AI capable of beating human experts in six-player no-limit Hold’em, the most widely played poker format in the world. This is the first time AI has beaten top human players in a complex game with more than two players or two teams.This video shows sample hands from Pluribus’s experiment against professional poker players. (Cards are turned face up to make it easier to see Pluribus’s strategy.)

Posted by Facebook AI on Wednesday, 10 July 2019
Facebook video with sample gameplay by Pluribus (runs 4:36)

Pluribus played 5,000 hands with winning poker pros who had million-dollar-games to their credit. Then it turned 10,000 hands with 13 other players and won again.

The MIT story says practical applications extend far beyond the felt of the gaming table. Future iterations might include strategy-making for pricing or automated control systems, such as in self-driving vehicles.

“Poker is the main benchmark problem for testing imperfect-information game solving capability… Multi-player games present additional challenges not present in two-player zero-sum games.”

Study co-author Dr. Tuomas Sandholm, Professor of Computer Science, Carnegie Mellon University, quoted in Newsweek

OUR TAKE

  • Making good calls with incomplete information lies at the centre of news judgment. It’s a very long path from these developments to machines making editorial choices but you can connect a line from here to there.
  • This is the second AI model released this year labelled to be ‘too dangerous to release’ (the first was the text-generating transformer model GPT-2.) In this case, the concern was for an unfair advantage if playing unsuspecting poker companies online.
  • Cynics may suggest this is a marketing ploy. Others say it is a responsible measure without withholding the scientific significance of the advance. Either way, it signals a higher state of performance from a new cohort of algorithms.

SEE FULL STORY

Facebook’s new poker-playing AI could wreck the online poker industry—so it’s not being released
MIT TECHNOLOGY REVIEW | July 11, 2019 | by Will Knight

SEE RELATED ITEMS

Leave a Reply

This site uses Akismet to reduce spam. Learn how your comment data is processed.