DeepMind team behind AlphaGo wins inaugural ‘Nobel Prize for AI’

20 Oct 201781 Shares

Share on FacebookTweet about this on TwitterShare on LinkedInShare on Google+Pin on PinterestShare on RedditEmail this to someone

Image: kc look/Shutterstock

Share on FacebookTweet about this on TwitterShare on LinkedInShare on Google+Pin on PinterestShare on RedditEmail this to someone

For achieving what few thought they could, the DeepMind team behind AlphaGo has been awarded the inaugural ‘Nobel Prize for AI’ by the IJCAI.

After showing its artificial intelligence (AI) could trounce even the greatest human Go player in the world, the DeepMind team behind AlphaGo has won the inaugural Marvin Minsky Medal for outstanding achievements in the field of AI.

Named after the man considered the ‘father of AI’, who died last year, the medal is to be presented by the International Joint Conferences on Artificial Intelligence (IJCAI) at its conference in Stockholm, Sweden next year.

In his career, Minsky was co-founder of MIT’s computer science and AI Laboratory and served on the faculty of MIT for more than half a century, in that time supervising many PhD students who also went on to be leaders in AI and computer science.

“AlphaGo is a wonderful achievement, and a perfect example of what the Minksy Medal was initiated to recognise,” said Prof Michael Wooldridge, chair of the IJCAI awards committee.

“What particularly impressed IJCAI was that AlphaGo achieves what it does through a brilliant combination of classic AI techniques as well as the state-of-the-art machine learning techniques that DeepMind is so closely associated with. It’s a breathtaking demonstration of contemporary AI, and we are delighted to be able to recognise it with this award.”

Aims to aid in scientific research

DeepMind CEO Demis Hassabis added that he and his fellow researchers were honoured with the award.

“We are truly honoured to be awarded the inaugural Marvin Minsky medal for AlphaGo,” he said. “We hope the ideas in AlphaGo are just the beginning and we are excited to now apply those techniques to the next set of grand challenges, such as advancing science and medicine.”

Since AlphaGo’s achievements last year, the DeepMind team recently announced the second iteration of its AI – AlphaGo Zero – that showed it could beat the original AlphaGo 90pc of the time at Go.

What made the achievement even greater from a technical perspective is that the new AI was almost entirely self-taught and required just four AI processors to run, compared with 48 in the original.

Colm Gorey is a journalist with Siliconrepublic.com

editorial@siliconrepublic.com