2 Minutes Read
A newly developed Artificial Intelligence (AI) AlphaGo by Google DeepMind software recently defeated the best Go players in the world. The researchers behind AlphaGo are not resting though, and believe that it can also beat the best pros at the toughest game to predict – poker!
In May last year, a challenge had been undertaken with four poker pros, including Doug Polk, Bjorn Li, Dong Kim and Jason Les facing off against AI – Claudico. The 8-week and 80,000 hands challenge ended with three of the four pros winning significantly against the software, though the computer experts called it a tie.
Go, is a Chinese game, much like Chess involving two players and determined strategies. The battle to beat the best human minds began 19 years ago, when IBM’s Deep Blue Sea beat Gary Kasparov, 19 years ago. It may seem a long journey in the progress of AI, but then Chess has about 20 moves available, whereas Go has 200 in comparison, much more intricate and complex.
Poker on the other hand is a game comprising “incomplete information’ with no way to determine what the next hand may bring, thus with no fixed strategy. This has made poker extremely difficult to computerize, as against games like Chess or Go.
Johannes Heinrich, a computer research student at UCL, told The Guardian “Games of imperfect information do pose a challenge to deep reinforcement learning, such as used in Go. I think it is an important problem to address, as most real-world applications do require decision making with imperfect information.”
The best way for an AI to tackle poker would be if it had the ability to adapt, grow and intuit strategy, much the way poker players do while “reading’ opponents and hands.
The DeepMind team comprising of David Silver, have ‘trained’ the AI to learn Texas Hold’em and Leduc strategies, by playing matches against itself and evolving on the game. In an almost human-like behavior, the AI learns and adapts from its own mistakes and experience to become better.
It was found that the AI had mastered state-of-the-art skills equivalent to those of humans in poker. In Leduc, the software reached the “Nash Equilibrium”, a game situation, where each player took the best decision, keeping in mind the opponent’s strategy, which remained unchanged.
Speaking about the growth of the AI, Heinrich added, “The key aspect of our result is that the algorithm is very general and learned a game of poker from scratch without having any prior knowledge about the game. This makes it conceivable that it is also applicable to other real-world problems that are strategic in nature.”
The last word is yet to be had though, as another team of researchers from the University of Oxford have joined Google DeepMind and this time the search is on to make bots that will tackle two extremely popular fantasy games – Magic: The Gathering and Hearthstone!