MILTON KEYNES, England — Canadian deep-learning pioneer Yoshua Bengio will lead an international effort to assess the power and perils of frontier AI systems, launched Thursday at the U.K.’s AI Safety Summit. Here’s what you need to know.

The document: Bengio will hold the pen on a “state-of-the-science” report, which will attempt to catalog the abilities of so-called frontier systems—the multi-purpose models that underpin tools like ChatGPT—and their risks. The paper’s findings are meant to inform policymaking, starting with the 29 governments including the U.S., China and Canada that on Wednesday committed to cooperate more closely on AI safety. 

“We must ensure that our shared understanding keeps pace with the rapid deployment and development of AI,” British Prime Minister Rishi Sunak said in a speech Thursday afternoon at the summit at Bletchley Park.

Bengio will be able to draw on the AI expertise of other academics with AI, and will be advised by a panel of delegates from each participating government.

The state-of-the-science initiative is based on the Intergovernmental Panel on Climate Change, the UN body set up in 1988 to produce periodic reports. Those releases always make headlines. But scientists have expressed concern that governments are ignoring their conclusions, while resource sectors and countries that host them have reportedly lobbied for them to be toned down.

The lead author: Université de Montréal professor Bengio—along with University of Toronto peer Geoffrey Hinton, their students and others—helped lay the modern foundations of deep learning, a now widely used AI approach. Bengio is a pillar of Canada’s national strategy for the technology as scientific director of Mila, one of three institutes that support AI research and commercialization.  

The Paris-born scientist has been thinking in public about responsible AI development for some time. He helped steer a December 2018 declaration of ethical principles for the field. But that focus has come to the fore over the last year, as Bengio and Hinton have repeatedly expressed concern that AI could pose catastrophic, even existential, risks to humanity. They’ve called for companies and governments to spend as much on safety as on advancing the technology, and for authorities to step up enforcement. (The third “father of deep learning,” Meta chief AI scientist Yann LeCun, has pushed back against those theories and recommendations.)

Domestically, Bengio has urged Canadian lawmakers to pass Bill C-27, which includes the Liberal government’s proposed Artificial Intelligence and Data Act. He co-chairs Innovation Minister François-Philippe Champagne’s advisory council on the technology.

Speaking to The Logic from the Bletchley summit on Thursday ahead of the announcement, Champagne cited Bengio’s selection as an example of Canada’s leading role on AI governance. “Yoshua played a big role here,” Champagne said, adding, “we’re already front and centre when it comes to thought leadership.”

Next up: While the U.K. government didn’t announce a due date for the report, the paper is supposed to feed into future summits. South Korea will host the next one virtually within six months, and France an in-person one within a year.