A team of researchers at Microsoft Asia have coded AI that self-taught the nuances of the game Mahjong. The AI can best even the most experienced players around the world.
Mahjong is a Chinese board game based on strategy chance and bluff. It was developed many years ago and is a common pastime for many Asians, with a good number of passionate players competing online.
Now, even computers have learnt the rules of the game, which is no surprise at all, considering that very complex algorithms can play games like Go and chess. Scientists have taken their skills a notch higher with a new code that can play Mahjong very well.
The researchers named this code Super Phoenix (Suphx for short). It based off complex code that can work through the intricacies of Mahjong. The scientists also feel that, with a little more work, they can develop systems capable of solving complex situations with many random and unknown factors.
Suphx is able to learn and even teach itself new things, which is how it is able to play Mahjong. The Code understands strategy, subtleties, and tactics because it has played against many people on platforms like Tenhou, the platform that millions of Japanese players frequent to play Mahjong online. The platform has more than 300000 members.
With more machine learning, the code rose from a mere novice to expert level after close to 5000 games in a period of four months.
Suphx has developed a unique playing tactic and is able to balance both defense and attack moves by strategically weighing long-term gain against short-term losses. Suphx can come up with quick calculations and decisions with limited information.
Suphx is the first Artificial Intelligence system to take part in Tenhou’s 10th dan ranking. Previously, only 180 people have taken part.