New Poker Bot From Czech University Defeats 10 Poker Pros

March 9, 2017 August 1, 2018 Paul Butcher https://plus.google.com/116893384630351018637
March 9, 2017 by Paul Butcher

As in many other fields, Artificial Intelligence (AI) is gaining ground in the world of poker as well. A computer program DeepStack is reported to have defeated 10 professional poker players in heads-up no-limit hold’em games according to a recent report.

The study results published in Science magazine shows that the computer was able to defeat 10 out of the 11 players in a statistically significant manner after the researchers loaded the bot with deep learning training to develop poker intuition. DeepStack has been created by researchers from the University of Alberta, Czech Technical University and Charles University in Prague.

The research team enlisted 33 players from the International Federation of Poker to play head-to-head against the bot. Eleven of the players chose to play the full 3,000-game match and all of them went down to DeepStack.

SciShow

The comprehensive win by the computer program indicates that humans no longer have an upper hand in imperfect information games. So far computer programs have been able to handle perfect information games like chess or checkers but not games like poker where knowledge of the game state is not fully known.

This is not the first time that an AI bot has beaten human players at poker. Earlier this year, Carnegie Mellon computer scientists built an algorithm called Libratus that was able to beat several professional poker players through a 120,000-hand poker tournament. The bot in this experiment was trained daily using a meta-algorithm that identified and fixed the weaknesses that the pro players had exploited in the day.

DeepStack is said to take a slightly different approach. It treats every hand separately and builds a strategy as the game goes on. The computer is built to evaluate two versions of its neural network, one for the shared cards and another for its own cards. The networks were trained on data from 10 million random poker games.

In a statement Michael Bowling, professor of machine learning and the study author said

In some sense this is probably a lot closer to what humans do. Humans certainly don’t, before they sit down and play, pre-compute how they’re going to play in every situation. And at the same time, humans can’t reason through all the ways the poker game would play out all the way to the end

The study results showcase the increasing ability of artificial intelligence to operate successfully in situations of insufficient information. Researchers hope to use the software for real-world applications in fields like medicine and defence.

About The Author

Author

Paul Butcher is a works as professor during the day and currently contributes to write articles for top10pokerwebsites.net during his time off. Visit Paul’s google+ page here