Microsoft’s AI plays mahjong at the same level as the world’s top players
- ConsenSys and AMD develop blockchain-based cloud computing infrastructure
- Google Drive, Apple iCloud and Dropbox: Which is the best cloud storage?
- Benefits of ERP technology in cloud computing
- 5 reasons why enterprises should use cloud computing
- The combination of cloud computing and virtual private network
- Microsoft arrangements to utilize ARM chips for cloud computing
- Cloud computing - A simple explanation
- Adobe earns big on 'the cloud'
- New Window Server: Breakthrough on cloud security (Part 2)
- New Window Server: Breakthrough on cloud security (Part 1)
To learn, this AI system went live ‘real battle’ with thousands of other players on Tenhou - a global online Mahjong tournament platform with over 300,000 members. The fact that artificial intelligence systems are constantly surpassing people in all fields, from scientific research to playing games, is no longer surprising. After beating humans in Go, Chess or Dota 2, AI recently continued to show their badass in another game – Mahjong.
Microsoft high-tech researchers have developed an intelligent system that can learn to play Mahjong on its own. Remarkably, this AI system has been "trained" to have remarkable skills, comparable with Mahjong players in the world.
Developed by researchers on a series of algorithms that allow reading and making decisions in Mahjong games' unexpected situations, this AI system has quite a name - Super Phoenix (aka Suphx).
A chess game born in the Qing Dynasty (China), Mahjong is not just a game of chance. Players face many unexpected elements, such as the unplayed chess pieces of the opponent. This requires players to calculate and must handle unexpected moves quickly. To win this game, the player must have both the ability to think and calculate the moves, and also the skill to hide emotions.
Therefore, designing an AI system that can play Mahjong well is not a small challenge, if not more complex than AI systems designed for Go or chess.
"Mahjong is an extremely complex game. Playing Mahjong is not only an art but also a science. A good Mahjong player requires observation, intuition, strategy, calculation, and opportunity, and these are huge challenges for an AI system." - said Dr. Hsiao-Wuen Hon, Vice President of Microsoft Research and Development in the Asia-Pacific region.
According to Mashable, Suphx is designed to be able to self-learn Mahjong's strategies, tactics, and subtlety. In order to learn, this AI system directly 'engaged in battle' with thousands of other players on Tenhou - a global online Mahjong tournament platform with over 300,000 members.
By applying Machine Learning technology, Microsoft scientists set goals that the more Suphx plays, the faster and more it learns. The results showed that Suphx became a great player after more than 5,000 matches within 4 months. Currently, Suphx is playing against human opponents in the 10th Division on Tenhou. This is the highest competitive class of this platform, with about 180 people qualified to compete.
After thousands of matches, Suphix now has a style of play of its own, which has a balance between offensive and defensive tactics. This AI system can simultaneously weigh between immediate losses and long-term strategic benefits, as well as make quick calculations and decisions.
According to the researchers, if further developed, these algorithms can be applied to solve problems arising from real-time random factors or events.
By: Joe Cook