TORONTO — The federal government has named a group of notable AI researchers to provide counsel on the field’s safety risks, and is also seeking the input of prominent startups and big business as it tries to push both adoption and regulation of the technology. 

Yoshua Bengio, the decorated computer scientist, is chairing Ottawa’s new Safe and Secure AI Advisory Group. It includes researchers who have led or worked at AI labs at leading academic institutions like the University of Toronto, McGill University and the University of Alberta, as well as tech firms like OpenAI, Meta and Google-owned DeepMind. 

The group will advise federal policymakers and the new Canadian AI Safety Institute (CAISI) on what could go wrong with the technology as well as measures and regulations to address those risks. 

Ottawa has also recast its existing Advisory Council on Artificial Intelligence with a stronger industry focus, adding executives from the likes of startups Cohere and AltaML, retail giant Canadian Tire and consulting firm CGI. The additions will help give input on what’s holding up AI adoption and where it can be encouraged, as well as how to build and best utilize the compute infrastructure to develop and run the technology.

“Canada has such a massive opportunity with AI, and we aren’t capturing it,” said Nicole Janssen, co-CEO of Edmonton-based AltaML. She cited the country’s challenges commercializing AI research, and other sectors’ reluctance to use the technology. 

“The pressure to move quickly to capture this opportunity now will come, likely, from the industry voices,” Janssen said, adding there are now more business representatives on the council, which previously also featured academic researchers. 

The heads of the country’s three AI institutes—Edmonton’s Amii, Montreal’s Mila, and Toronto’s Vector—will remain on the council, as will Elissa Strome, a Canadian Institute for Advanced Research executive director who oversees the Pan-Canadian AI Strategy.

Strome and Bengio will sit on both panels. A Université de Montréal professor and Mila’s scientific director, Bengio is among a number of prominent figures in the field calling for governments to do more to ward off AI’s potentially catastrophic consequences. He is also the lead author for a series of international reports on the capabilities and risks of the most advanced AI models that are meant to inform policymaking. 

Joining Bengio in the safety group is Jeff Clune, a University of British Columbia researcher whose work explores how AI systems can engage in open-ended learning and act in problematic ways to achieve the goals set for them. McGill University professor Joëlle Pineau is another member; she studies AI planning and learning, and also leads Meta’s Fundamental AI Research team. 

The group also includes David Lie and David Duvenaud, two of the directors of the University of Toronto’s Schwartz Reisman Institute, the latter of whom worked at Anthropic on tests to ensure the startup’s AI models aren’t trying to deceive or sabotage users. 

“What do we do if we don’t like the direction that AI is pushing society?” Duvenaud said, adding that researchers in both industry and academia are reluctant to consider the consequences of achieving their goal of artificial general intelligence (AGI), when machines match and then exceed human capabilities. 

“No one actually has a good plan, and it should be alarming,” Duvenaud said, citing challenges like job losses from automation. While he doesn’t yet know what the federal safety group will tackle, he hopes it will look at issues such as AI’s long-term societal impact alongside more immediate regulatory questions.   

Ottawa launched the CAISI last month, tasking it with conducting research on AI risks and testing new models. It joined a network of similar agencies around the world, including the U.K. and U.S. 

The splitting of Ottawa’s AI advisory body into separate industry- and research-focused groups reflects its two concurrent priorities. Innovation Minister François-Philippe Champagne has long argued that Canada needs to balance encouraging the use of AI for economic gains with building the public’s trust in the technology. Startup leaders, meanwhile, have complained that the government is overly focused on AI’s risks—though some researchers have warned that companies are ignoring safety concerns in pursuit of profits. 

“There’s still so much fear around the technology,” Janssen said, adding she hopes that the safety group will help show how AI’s risks can be dealt with rather than stoking new worries.

The proposed Artificial Intelligence and Data Act, Champagne’s flagship effort to regulate the technology, died when the Liberals prorogued Parliament in January. But Ottawa continues to build on a voluntary code of conduct for generative AI, first launched in September 2023. 

On Thursday, the federal innovation department published new guidance for how firms could implement commitments to safety, testing and transparency when using AI. It also announced six new signatories including Intel and CIBC, joining the likes of AltaML, Cohere, Coveo and IBM.