24-12-2019 11:02 am Published by Nederland.ai Leave your thoughts

Artificial intelligence controls multiplayer poker

JASON SOLO / THE JACKY WINTER GROUP

This year, an artificial intelligence (AI) program beat some of the world's best players in the most popular poker version, Texas Hold'em without limit. The distinctive result marks the first time that AI has prevailed in a multiplayer match in which players have only imperfect information about the status of the game.

AI has driven people away in games at a spectacular pace. In 2007 computer scientists developed a program that would certainly not lose at checkers. In 2016, another team developed an AI program that beat the best people at Go, a board game with many more configurations than disks.

Poker offers a greater challenge because players cannot see the cards of their opponents and therefore have limited information. In 2017, computer scientists developed an AI program that was unbeatable with a version for Hold'em for two players, in which each player forms a hand of five cards that are open on the table and two other cards that are private.

Now, AI has outperformed world-class players in the full multiplayer game, as computer scientists at Carnegie Mellon University in Pittsburgh, Pennsylvania announced in August. By playing 1 trillion games against themselves, their program, Pluribus, developed a basic strategy for different types of situations, such as playing for an inside straight. For each specific hand, it might also think about how the cards would probably play. In 20,000 hands with six players it surpassed 15 top players, measured by the average profit per hand.

Pluribus played differently than programs for two-player games. Those programs were looking for a lossless strategy, known as a Nash balance, that guarantees that their opponents would do worse on average, unless they played with the exact same strategy. With multiple opponents there is no such guarantee, so Pluribus easily learned what was most effective in a given situation. The program also taught itself to play while it was running on a single server with 64 processors, while the Go-playing program required more than 1200 processors.

AI developers are not yet ready to play games. In poker there is still room for improvement. Although Pluribus can bluff, it cannot adjust its strategy to exploit the specific weaknesses of an opponent. Some more complex games, such as contract bridge, remain unmastered. Yet the most well-known goal in applying AI to games for the computers has fallen. It may be time that people cash in on their chips.

Adrian Cho

References

  1. Brown and T. Sandholm, Superhuman AI for multiplayer poker , Science , Vol. 365, p. 885, August 30, 2019
  2. Blair and A. Saffidine, AI surpasses people in poker for six players , Science , Vol. 365, p. 885, August 30, 2019

source https://vis.sciencemag.org/breakthrough2019/finalists/#intelligence

Tags: , ,

Leave a Reply

Your email address will not be published. Required fields are marked *

16 − three =

The maximum upload file size: 256 MB. You can upload: image, audio, video, document, spreadsheet, interactive, text, archive, code, other. Links to YouTube, Facebook, Twitter and other services inserted in the comment text will be automatically embedded. Drop file here