Online poker players sometimes accuse those with uncanny abilities to be bots rather than real people. However, given how devastating Facebook’s Pluribus AI can be at the poker table, these players may want to reassess the validity of their accusations.
Facebook’s newest creation is a program developed in collaboration with Carnegie Mellon University. Quite simply, the two groups have written a code that can consistently beat six-max no-limit hold ’em.
Pluribus 1, humans 0
According to the report from Facebook AI Research Scientist Noam Brown, the code manages an average win of $5 per hand. From that average, Brown extrapolates that Pluribus could extract $1,000 per hour from a game with five human opponents.
That extrapolation comes as the result of several tests that Brown ran against human opponents. The research team tested two different types of situations:
- Five human players and one bot at the table.
- One human player and five bots at the table.
These human players were not just average players either. Pluribus prevailed against some of the very best players in the world. The player pool included:
- Jimmy Chou
- Seth Davies
- Darren Elias
- Chris Ferguson
- Michael Gagliano
- Anthony Gregg
- Dong Kim
- Jason Les
- Linus Loeliger
- Daniel McAulay
- Greg Merson
- Nick Petrangelo
- Sean Ruane
- Trevor Savage
- Jake Toole
Time and again, Pluribus prevailed against a randomly drawn player or players from this pool of killers. The humans lost an average of 2.3 bb/100 in the course of playing against Pluribus.
Lest one think they were soft-playing, Facebook incentivized the players with a $50,000 prize to the top performer. So, Pluribus chewed up the best games of those world-class players and spat them out.
Pluribus’ implications for the future of poker
Of course, Pluribus is not the first poker bot ever created. In fact, the code itself is an improved version of a previous program called Libratus.
However, Libratus could only accommodate heads-up play. Pluribus proved that it could adapt to the variety of styles present in a multi-opponent ring game.
The fact that a code that solves poker exists is shocking, to be sure. But, the price of the code, or lack thereof, is even more surprising.
In fact, Facebook created Pluribus in eight days with $150 and using 512 GB of RAM. For most programs like this, that’s barely enough to plug in the machines. So, in short, Facebook has broken six-max for less than a full buy-in at a $1/$2 game.
Where does that leave the human player? The top professionals raved about the program’s performance, both overall and in terms of nuance.
Chris Ferguson mentioned that Pluribus had a knack for finding thin value on the river. Darren Elias said that he felt gratified to see a computer program mirror the moves of a top professional.
Of course, the true advantage of Pluribus has is consistency. A computer never gets tired, never gets bored, never goes on tilt.
So, Pluribus is a player that brings one’s A game at all times because, frankly, there’s nothing else for it to play. Sean Ruane commented on Pluribus’ “relentless consistency.”
However, God forbid the code falls into the wrong hands. An online poker player without moral compunction could use Pluribus as a money tree.
After all, there’s no need to cheat if one can only play with the highest level of skill possible.