OTTAWA — The federal government wants to have a plan for using artificial intelligence in its own activities by this time next year. Seeking advice on how to put that plan together, Treasury Board President Anita Anand convened think-tankers and academics on Monday in a cavernous conference room in a government complex in Gatineau, Que. 

Here’s some of what the experts had to say.

Definitions: “Generative AI is driving this conversation,” said Stephen Burt, the government’s chief data officer, chairing the first part of the meeting. Machine learning is part of the picture, too. But the government doesn’t want to get bogged down in defining AI for all time, he said.

Trust: Most Canadians don’t trust most public institutions these days, which is a challenge for getting people to accept that their government is going to do occult-seeming AI things with their data and to make decisions that affect them.

“If they trust public institutions, they will trust that public institutions can safely develop AI strategies, in the same way that they trust that public institutions can safeguard us in terms of the foods we consume,” said Cliff van der Linden, the academic director of McMaster University’s Digital Society Lab. If they don’t, they won’t.

Talent: The federal government is short thousands of technology workers, Anand has acknowledged.

“There is talent in Canada. It just might not go and work for the Canadian government,” said Valérie Pisano, CEO of Mila, the Montreal institute built around AI pioneer Yoshua Bengio. If the government wants to use AI well, it will need ways to tap that talent besides hiring it. “We’ll have to be very innovative with how we do this, not only in what we’re doing.”

(The government is making its jobs easier to apply for and plans a new division to train current public servants in cyberskills, Anand said outside the meeting.)

The government also has massive technical debt, such that embedding AI into ancient systems is often not worth it. 

Goals: A strategy should contemplate what the government wants to achieve with AI, said Tony Gaffney, CEO of Toronto’s Vector Institute. 

Improving productivity is one valid possibility, Gaffney said. So is tackling complex social challenges—predicting and mitigating the effects of forest fires might be an example. Other countries are making their own plans for public use of AI. Clarity of purpose is a common theme.

Privacy: What the government can do with data—what information it can feed into what models for what purposes—is partly dictated by the federal Privacy Act. It was passed in 1985, when artificial intelligence was pure science fiction. 

“We have a federal privacy law that is an antique,” said University of Ottawa law professor Teresa Scassa.

Burt agreed. Preliminary work on an overhaul is underway but “it needs to be done in a way that will allow us to be more nimble in future,” he said.