Game Changing Poker Bot from Facebook Beats Libratus

August 3rd, 2020 | by Jason Reynolds

Facebook’s poker bot has raised the game and may not only pose a threat to humans but the top artificial intelligence (AI) player out there, Pluribus.

Facebook’s latest AI innovation, ReBeL, has proved once again that bots are superior to humans at No Limit Hold’em. (Image: Singularity Hub)

The innovation, known as Recursive Belief-based Learning (ReBeL), has been hailed as another breakthrough for AI.

As with other bots, poker was used because it’s a game of incomplete information. ReBeL has managed to master the nuances of heads-up No Limit Hold’em using a new concept known as game state.

The major difference between ReBeL and other poker bots is that it creates two simultaneous models: a value network and a policy network. The former considers information such as cards, bet sizing, and hand ranges. The latter models each player’s (agent’s) beliefs.

Facebook Shakes Things Up with ReBeL

The system runs through millions of scenarios, learning which moves are optimal in each. In tandem, the two models give ReBeL the ability to not only play a game theory perfect style of poker but react in a similar way a human would.

This makes ReBeL adept at solving imperfect information games such as No Limit Texas Hold’em. In fact, it’s become so adept that it’s already able to outperform previously impressive bots like Libratus.

In a 7,500-hand match against heads-up specialist Dong Kim, ReBeL achieved a win rate of 165 thousandths of a big blind per hand. Libratus “only” had a win rate of 147 thousandths of a big blind per hand.

At this stage, it’s not quite as accomplished a Carnegie Mellon University’s most recent creation, Pluribus. However, the goal isn’t necessarily to create the AI equivalent of Phil Hellmuth.

Poker Bot Will Change Many Games

Even though researchers at Facebook will refine the system, the long-term goal is to use ReBeL in real-life scenarios. By having a system that can devise optimal strategies in a vacuum and in the moment, the applications are potentially limitless.

Cyber security, self-driving cars, and trading are just three uses for the new technology. Of course, there could also be implications for online poker players.

Operators are more proactive than ever when it comes to identifying bots. In 2019, Partypoker led the way with a new security team tasked with catching and eradicating cheats from the network.

However, as is often the case, it’s a game of cat and mouse. As technology improves, it becomes harder to spot those using bots. Then, if security teams can catch the cheats, they typically find a new way to game the system.

ReBeL is another example of how AI is becoming ever more sophisticated. If this technology could be package and sold to players, it would pose a serious threat to the integrity of online poker.

For now, however, the skills of ReBeL are more likely to change the game in industries outside of poker.