Terrorized by the AI, this master of the game of Go throws in the towel

After successfully defeating Google's AI for a game, Go's game master finally decided to retire. Handsome player, the latter will play a final gala match against ... an AI.

He had defeated AlphaGo, Google's artificial intelligence system, before finally throwing in the towel. South Korean game master Go Lee Se-dol, who retired from the professional competition last week after gaining worldwide fame in 2016 as the only human player to win against AlphaGo, said that his retirement was mainly motivated by the invincibility of AI Go programs.

Asked by the South Korean news agency Yonhap News, the latter attributed its historic victory of March 13, 2016, to a flaw in the AlphaGo program. "With the beginnings of AI in Go games, I realized that I would not be at the top, even if I become the number one thanks to frenetic efforts", regretted the latter, who sees in the development the IA a major upheaval for his discipline, yet more than 2,500 years old.

Lee Se-dol, a master of Go, is not yet a veteran of the circuit. At just 36, he had won 18 international and 32 national competition victories. He handed over his retirement letter to the Korea Baduk Association (KBA), which oversees the Go professionals in South Korea on November 19, ending its 24-year legendary career.

A final gala match against an AI

"Even if I become the number one, there is an entity that can not be defeated," he said in an interview, referring to the AI ​​system developed by Google.

A handsome player, the latter intends to face another AI Go player next month to commemorate his retirement. Its new competitor will be Handel, a program developed by the South Korean company NHN Entertainment Corp. Developed in 2018, the AI ​​system has already beaten South Korea's top five Go players.

In the first game, Lee will have a two-shot advantage. His advantage in subsequent matches will be determined by the result of the first match.

"Even with a two-shot advantage, I feel like I'm going to lose the first match against Handel, and these days I'm not Go's info anymore." I wanted to play comfortably against HanDol because I already retired, but I will do my best, "said the latter, a bit defeatist on the outcome of this future gala match.

Google is trying out Starcraft 2

After having tried chess and go game, the Deepmind teams announced in 2016 a partnership with Blizzard, the publisher of Starcraft 2. By the end of October, Google's AI branch had achieved the feat of elevating its AlphaStar AI to the rank of "grandmaster" , which means that it is now able to beat 99.8% of all players humans in competition.

To advance AlphaStar, DeepMind explained to have modified its reinforcement learning method by which the program learns by trial and error by multiplying the parts against itself with two opposing algorithms.

Instead of each algorithm trying to maximize its chances of winning, one was programmed to expose the flaws in the other's strategy. A method inspired by the training techniques of Starcraft 2 professional players.

For DeepMind, the progress made by Google in the field of AI suggests the possibility of creating more general algorithms, able to adapt their learning to improve the performance of digital assistants, robots or autonomous cars.