Facebook Develops New Poker AI That Beats the World’s Top Players

14.07.2019, 17:00

Researchers from Facebook and a top university in the United States have created an artificial intelligence (AI) that can defeat top human players in poker.

In the artificial intelligence field, researchers view poker as a challenge. The reasoning behind this is because poker, one of the most popular casino games in the world, involves hidden moves by the players. As such, success in the game requires bluffing and many other approaches that are rarely used in different sporting events.

Due to this, poker has remained impervious to AI techniques that have been used to crack other games. Scientists have been able to come up with artificial intelligence programs that can beat one human player at no-limit poker, but up to now, had found the multiplayer version too hard to crack.

However, they have finally managed to develop an AI known as Pluribus. Scientists Noam Brown, who works at tech giant Facebook, and Professor Tuomas Sandholm, developed the AI program.

According to the two researchers, the bot mastered the multiplayer version of poker by competing against previous versions of itself. In this learning method, the bot did not have any access to playing information from humans. The program also did not study games played by other computer systems.

The researchers said that the program starts the game from scratch by playing randomly. It then gradually improves as it analyzes and maps which actions give it a better chance of winning the game.

Tech giant Google has, in the past, exploited the strategy in its artificial intelligence research.

Pluribus defeated top professional Texas hold’em players in both a five AIs plus one human player format, and a one AI plus five human players form. Among the professionals who lost to Pluribus was Chris Ferguson, a World Series of Poker Champion, and Darren Elias, who is the record holder for most poker championship titles.

Even though there was no money at stake during the program’s competition with humans, the researchers said that if one chip were worth one American dollar, the program would have bagged an average of approximately $5 per hand. This means that in a real poker game, the program would have won approximately $1000.

The bot is an advanced version of another bot. In 2017, the earlier version defeated human players in two-player poker games.

However, unlike the previous version, the latest development is fitted with a new online search algorithm that can study its options by thinking a few moves ahead. It also contains algorithms that make it extremely quick. Due to these two advantages, the researchers were able to train the AI using relatively small memory and processing power. According to the scientists, the training only required computing resources worth a paltry $150.

Tech companies are investing heavily in artificial intelligence. The hope is that breakthroughs in gaming will lead to advances in other areas such as energy, healthcare, and science.

For decades, artificial Intelligence scientists have been using games to test their AI agents. Due to advances in computing, better data sets, and more advanced AI techniques, there have been several encouraging developments in the field.