Facebook developed a new poker bot
Facebook Corporation introduced a new poker bot called ReBel. The main focus of this bot is on self-learning and finding solutions for the games with incomplete information.
The developer company provided a report with the basic principles of bot operation and its comparison with other artificial intelligence programs.
The main difference of a new bot is that it analyzes information about the game, as well as takes into account the opponent’s opinion in the process of hand play. Therefore, ReBel can “think” about opponent's reaction, like all successful professional players do.
ReBeL takes into account not only information about the visible state of play (for example: known cards, bets’ size and even possible range of opponent’s hands). In addition, it also takes into account the players’ opinion concerning their state, i.e. how the person can realize whether the opponent thinks he is ahead or behind in a hand.
Rebel's creators believe their bot is stronger than its predecessor Libratus. Rebel is two seconds faster in playing hand and it needs only 5 seconds to make a decision.
Who did ReBel play with?
So far, Rebel has played against one real opponent – Dong Kim. He also played against Libratus. In total, Dong and Rebel played 7 500 hands, and bot won 0.165BB against Dong in one hand.
Of course, this is a small sample to estimate the strength of a new bot. In the future, we will see other games of bot against famous poker players.
In any case, the creators of bot assure that their bot won’t negatively affect online poker. Their goal is to help people organize the work of complex systems with an incomplete amount of information, rather than getting illegal earnings.
According to creators’ team, the most immediate risk associated with this job is its potential for swindle in entertaining games such as poker. Partly for this reason, they decided not to release code for poker.
However, they released an open source implementation for Liar's Dice in order to help in the future research. The developers believe that ReBeL can help in the development of more efficient algorithms of finding general equilibrium for applications in auctions, negotiations, cybersecurity and self-driving vehicles.