Microsoft claims AI breakthrough in Chinese game of Mahjong

An artificial intelligence (AI) system developed by Microsoft researchers has cracked the Chinese game of Mahjong, making its way into the upper echelons of a global competition.

Dubbed Super Phoenix, or Suphx for short, the system was taught to self-learn Mahjong’s strategies, tactics and subtleties by playing against human players on Tenhou, an online Mahjong competition platform with more than 300,000 members worldwide.

Through constant machine learning, Suphx went from being a novice to an expert after playing more than 5,000 games in four months. The more it played, the more it learned at an ever-increasing pace.

It has now honed its own playing style and can balance attack and defence moves, strategically weighing short-term losses against long-term gains, and making quick calculations and decisions based on unclear information.

This feat has made Suphx the first AI system to compete at Tenhou’s prestigious “10th dan” ranking – achieved by just 180 people to date. Only a handful of professionals now play at a higher level in a private room for human players only.

Mahjong is a complex board game of chance, bluff and strategy invented in China thousands of years ago and remains a pastime for millions of Asians today, with many dedicated competitors playing online.

Apart from Mahjong, computers have learned to play chess and Go, another ancient Chinese game.

But, unlike chess and Go, which are both “perfect information games” in game theory, with players getting to see everything on a board that can affect the outcome, Mahjong is an “imperfect information game” because there are many unknown factors.

For one thing, Mahjong players must account for their opponents’ unseen tiles. This can lead to bluffs and unpredictable outcomes as they decide what to discard, or whether to meld or fold.

“For as long as researchers have studied AI, they have worked to build agents capable of accomplishing game missions,” said Hsiao-Wuen Hon, corporate vice-president at Microsoft Research Asia.

“Mahjong is more complex than other board games, so playing becomes an art as well as a science. Good Mahjong players rely on a combination of observation, intuition, strategy, calculation and chance that presents unique challenges for an AI system.”

But Microsoft scientists see their achievement as more than just a case of technology mastering yet another game. With more work, their AI algorithmic breakthroughs with Suphx could be used to solve problems arising from unknown factors and random events.