A Computer That Bluffs
Ekam Walia | Abhimanyu Agarwal | Shounak Dey
“It’s not about the money, it’s about preserving human dignity and it’s not going well.” says Jason Les who is one among the four professional Poker players engaged in a Heads-Up No-Limit Texas Hold ‘em Poker game against Libratus, an AI computer program created by Tuomas Sandholm and Noam Brown at Carnegie Mellon University.
IBM’s ‘Deep Blue’ became a new benchmark in the field of artificial intelligence when it defeated Garry Kasparov at Chess. AlphaGo, developed by Google as a part of its Deepmind project, upped the ante last year in March by defeating Lee Sedol who has earned a 9-dan ranking at the board game ‘Go’. Now, by giving professional Poker players a run for their money, Libratus has redefined the standard for artificial Intelligence. But why is Poker still a big deal if AI algorithms can play Chess and ‘Go’?
Games such as Chess and ‘Go’ are known as perfect-information games since all the information is presented to the players on the board. However, unlike Chess and ‘Go’, Poker is an imperfect information game which means that it is more random in nature. The only information provided to the player is in the form of his cards and the community cards. In Chess and ‘Go’, due to availability of perfect information, the AI has definitive information to act on. However, this is much harder in games like Poker. The introduction of uncertainty in the mix changes the dynamics of the game and makes it tougher to develop such an algorithm.
To work through the uncertainty and arbitrary nature of the game, Libratus focuses on using information-sets to plan its strategy i.e. it analyzes all possible moves on the basis of known variables (such as its own cards) and unknown variables (such as other player’s cards). This is a challenge because an approximate of 10 to the power of 160 such information-sets are possible. The creators of this AI system merely taught it the rules of Poker. Then, the system perfected its strategy and game-play over one trillion simulations at Pittsburgh Supercomputer Center.
The mathematical undertaking involved in this project is vast. However, this is easier for a computer when compared to humans as the unpredictability of the AI system increases with its every move. It keeps changing its strategies with every hand and learns from its mistakes. By the time the professionals could deduce any of its weak spots, the AI had stopped making those mistakes. Another advantage that the AI has over humans is the lack of a conscience. Owing to this, the AI frequently over-bets and bluffs because over-betting may sound risky to a human but the AI system cannot interpret any such emotion.
One might wonder, “What is the use of investing time and resources in the development of a computer program that can play a game even though the game is highly complicated?”. The answer to that question may be that most problems in reality can be classified as imperfect-information because all requisite information is almost never available. In such a world, a program that can act on indefinite information and surf through a pool of possibilities to figure out the best course of action and – after proper training and development – can prove to be an indispensable asset in solving many problems ranging from cyber-security to treating cancer.
This January, Libratus (like its predecessor Claudico in 2015) played against four highly acclaimed Poker professional at Rivers Casino in a game of 120,000 hands spread over a period of twenty days. Libratus had a clear lead since the first day of the tournament and won a total of $1,766,250 thereby redefining our conceptions of what machines can do.
The ‘Brains Vs AI’ tournament served as an excellent test of the capabilities of AI. However, as some skeptics from around the globe pointed out, this also puts the future of humans in uncertainty. The ethical dilemma involving AI systems is continuously gaining prevalence and by defeating humans at Poker, Libratus has only added fuel to the fire. However, going by the amount of research being carried out in artificial intelligence, we will definitely see more powerful algorithms achieving even greater feats.