Facebook AI that learns how to play Poker on the fly

Artificial Intelligence has left no stone unturned when it comes to beating humans poker hand sequence. Be it applications, software, chatbots and now this revolutionary technology is setting a firm foot in the gaming industry as well. Well, Facebook has come up with its latest poker bot, which has been making noise in the gaming industry ever since its launch. Facebook AI that learns poker is designed by a joint collaboration of Facebook’s AI department and Carnegie Mellon University. The reason behind the hype of Facebook AI that learns poker is the fact that it has outperformed some of the best poker players in the world in a series of six-player Texas Hold’em poker after you know poker hand sequence. Let’s see what this Facebook AI algorithm is all about and what its scope is!

ReBel is believed to be the most popular Facebook AI that learns poker. 

Some time back, poker was seen as a major challenge in the field of AI. The reason behind this is, the game involves the use of hidden tricks and information — you can never know what your opponent’s cards are and what strategy he is using — means in order to win, one must use bluffing and other advanced tactics that are specifically designed for poker. 

This problem has kept poker away from AI techniques that have already worked great for other games once you know poker hand sequence. But this was before the launch of Facebook AI that learns poker – ReBel. 

Yes, ReBel is the second poker AI developed by Facebook. It launched exactly one year after the release of Pluribus, Facebook’s first poker bot. What makes ReBel stand out is the fact that it scores way more higher than Pluribus!

ReBel — The Facebook AI That Learns Poker  

If anyone remembers Pluribus, it gained a total of $1,000 on per hour basis when competing against the human professionals. But this new poker AI is more advanced and efficient than the previous one. It has amassed a higher score than Pluribus. 

How Rebel AI Works?

The ReBel AI is based on the reinforcement learning model that allows it to learn a game very quickly. What makes it stand out from the other poker AIs is, it comes with not one but two models in play. One represents the value and the other represents value. This results in the creation of a “public belief state” that allows it to coin probabilities on the basis of all the relevant information and tactics in the play. These include all the possible outcomes and overall pot of the game. Because ReBel uses a public belief state and not the “specific domain knowledge” used by the other AIs, it works in a more generalized way. 

The most fascinating thing about ReBel is not that it is capable of beating the world’s best poker players once you know the poker hand sequence. Instead, it is that this AI was originally developed to solve poker game’s applications off the gaming industry such as for cybersecurity measures and self-driving cars. Don’t be surprised to know that one poker bot was even hired for effective military planning!

ReBel Vs Poker Pros 

During the trials, the researchers used ReBel on a series of heads-up no-limit Texas hold ’em poker, turn endgame hold’em and Liar’s Dice, where both players call or check for the four initial rounds. To generate the game data, the team used 128 PCs with eight graphics cards each. Further, they randomized the stack and bet sizes between 5,000 to 25,000 chips, during the training. ReBel was fully trained and had $20 thousand to bet against the opponents. 

The researchers reported that while playing against the #1 player in the world, Dong Kim, ReBel played 2 seconds faster per hand for up to 7,500 hands. Also, it never needed more than 5 seconds for making a decision. 

Further, in aggregate, ReBel scored 165 thousandths per game. Against human players, it left behind Facebook’s previous poker system, Libratus, which wrapped up at 147 thousandths. 

Another interesting news is, Facebook AI team has decided to launch ReBel as an open-sourced implementation, instead of releasing it as a codebase poker for the fear of cheating. This implementation is easy to implement and understand. 

Should Poker Players Be Concerned About The Arrival Of Sophisticated AIs? 

Facebook revealed that it will not release the ReBeL codebase anytime soon. So, players can sigh in relief that they will not have to face this gigantic AI bot in the real poker world. 

However, this can lead us to another issue that is encountering AI copies of similar poker bots. Many players are wondering if hackers can replicate this game-breaking Facebook AI that learns poker. Because if it is done, it can soon become a matter of concern to the players. 

To develop state-of-the-art AIs like ReBel and Pluribus, extensive research and analysis needs to be done. This is the reason for which Pluribus was developed by collaborating with the researchers of Carnegie Mellon University. These pioneers and knowledgeable researchers have worked hard to develop a groundbreaking tool like that. 

Another thing about the use of intelligent poker bots is, there are many poker websites and software that have their entire teams dedicated to identifying and eliminating bots. So, even if the most advanced poker AIs come into the market, these poker sites and rooms will have them shut as soon as they find them. So, as of now, players don’t need to worry about  being cornered by the advanced poker bots. 


Related Articles

Leave a Comment