by Ken Wieland

15 July 2019


LONDON — In what was widely hailed as a milestone for AI and machine learning, a robot going by the name of Pluribus beat some of the best poker players in the world.

Developed by scientists at Carnegie Mellon University in collaboration with Facebook AI, Pluribus defeated leading professionals in games of no-limit Texas Hold’em poker.

The victory was especially significant, said AI experts, given the subtle stratagems of bluff and counter bluff that every successful poker player needs. Making things tougher for the bot was that it played five humans in a six-player game, yet it apparently sailed through this test with flying colours.

This is not the first time AI has been used to see off world-class human opponents. In 2016, a machine called AlphaGo – built by Google-owned Deep Mind – beat Lee Sedol, 18-time world champion at Go, an abstract strategy board game for two players. By using machine‑learning algorithms, AlphaGo makers claimed the bot taught itself how to win the Chinese game — soaking in thousands of years of accumulated wisdom — in just 40 days. 


Related: Poker Playing AI Beats Chinese World Series of Poker Veteran


The thinking from big tech companies, which develop advanced AI algorithms to beat humans at games of this sort, is that it will help broaden AI applications in the real world. What those new applications might be is not entirely clear, however.

Flushed with success

Professor Sandholm, who helped develop Pluribus at Carnegie Mellon’s Computer Science Department, was fulsome in his praise for the poker‑playing bot.

“Pluribus achieved superhuman performance at multi-player poker, which is a recognised milestone in artificial intelligence,” he said. “Thus far, superhuman AI milestones in strategic reasoning have been limited to two-party competition. The ability to beat five other players in such a complicated game opens up new opportunities to use AI to solve a wide variety of real-world problems.”

In 10,000 hands of Texas Hold’em, Pluribus competed against five contestants from a pool of 13 professionals, all of whom had won more than $1 million playing poker. Every 100 hands, Pluribus raked in — on average — about $480 from its human competitors. “This is roughly the amount that elite human professionals aspire to beat weaker players by,” said Noam Brown of Facebook AI Research. Brown worked alongside Professor Sandholm in developing Pluribus.