Google’s Go-playing AI retires after beating world's best human player three times

AlphaGo, the AI trained by Google to play Go that became the first piece of software to defeat a human professional at the abstract game, is going out on a high.

AlphaGo was originally conceived in 2014 as a way of testing deep learning, utilising Go’s complexity over other games such as chess.

It cemented its fame in late 2015 by beating a real-life pro – the first time the feat had been achieved by a virtual opponent – before defeating South Korean Go icon Lee Sedol 4-1 in a $1 million five-game match.

This month, AlphaGo took on the world’s highest-ranked player, Ke Jie, at the Future of Go summit and emerged the uncontested victor, claiming a hat-trick victory three-nil.

With no human player seemingly worthy of taking on the artificial champion, the developers at Google behind the programme have announced that AlphaGo will now be retired, having reached “the highest possible pinnacle as a competitive program”.

The research accomplished using AlphaGo’s learning will be applied to other areas such as seeking cures for diseases, improving energy usage and creating new materials, the team added.

Information gathered from 50 of AlphaGo’s matches has been made available for the Go community to study, with a teaching tool based on the AI and assisted by Ke Jie in the works for the future. However, AlphaGo itself will not be released to the public.

“We have always believed in the potential for AI to help society discover new knowledge and benefit from it, and AlphaGo has given us an early glimpse that this may indeed be possible,” said Demis Hassabis, head of Google’s DeepMind division.

“More than a competitor, AlphaGo has been a tool to inspire Go players to try new strategies and uncover new ideas in this 3,000 year-old game.”