Photo/Illutration Demis Hassabis, co-developer of the AlphaFold AI model, speaks with reporters in Tokyo on Nov. 21. (Tatsuya Shimada)

This year’s Nobel Prizes were a “watershed moment” for artificial intelligence, according to Demis Hassabis, a co-winner of this year’s prize in chemistry.

Hassabis, 48, CEO and co-founder of Google DeepMind, developed the AI model AlphaFold to predict protein structure with company co-founder John Jumper.

Proteins are made up of amino acids linked together in long strings and folded into a three-dimensional structure.

Scientists have spent years trying to unravel the structure of specific proteins in order to develop new drugs. The AlphaFold model can predict protein structures in a matter of minutes at almost no cost.

Speaking at a news conference in Tokyo on Nov. 21, Hassabis said AI “will be one of the most beneficial technologies we ever invent” but added it had “a lot of unknowns” regarding risks that needed to be further researched.

He said that, rather than rushing to develop AI applications, researchers should “use the scientific method to really understand what we’re actually building and the consequences of it before you deploy it everywhere.”

Hassabis was in Tokyo to receive this year’s Keio Medical Science Prize.

He said the Nobel Prizes were a way for the Nobel Committee to say, “AI now can be recognized as a proper science.”

The Nobel Prize in Physics also went to two scientists for their work in AI-related research.

Hassabis added that it had been his dream for 30 years to “use AI to advance our understanding of the world around us and accelerate scientific discovery.”

The Nobel Prize was a “kind of testament that we have reached that level with this technology,” he said.

Hassabis also acknowledged the risks associated with AI.

“A bad actor or rogue nation can repurpose the same technology that we’re trying to use for good to do harm,” Hassabis said. “One big question for society is, how do we enable all the good users whilst restricting access to the technology for people who want to use it for harm.”

While he said some AI experts were pessimistic about regulating its use, Hassabis himself expressed optimism that the issue could be resolved.

He added that, with the advances in AI in recent years, achieving artificial general intelligence was now a viable goal for his lifetime.

Hassabis also visited the Nihon Ki-in, headquarters of the Japanese game go, in Tokyo’s Chiyoda Ward on Nov. 21.

Hassabis, a chess master, learned go when he was an undergraduate student and used that knowledge to develop AlphaGo, an AI model that became famous for defeating the best human go players.

After a ceremonial match against the highly decorated go player Yuta Iyama, Hassabis said that AlphaGo laid the groundwork for later development of the AlphaFold model.

(This article was written by Takahiro Takenouchi and Kenta Noguchi.)

Related News