OTTAWA — A lack of detail in the federal government’s proposed law to govern artificial intelligence has spooked some of Canada’s leading AI startups, as well as some large firms deploying the technology.

Innovation Minister François-Philippe Champagne introduced Bill C-27 in June 2022 to replace Canada’s two-decade-old private-sector privacy law, after the Liberals’ first proposal failed to pass Parliament. The legislation included a new Artificial Intelligence and Data Act (AIDA), which would set requirements for how the technology could be developed and deployed, and ban uses that could cause “serious harm.”  

Ottawa’s proposal followed headline-grabbing instances of companies and government agencies using AI systems for content targeting, credit scoring, facial recognition and data analysis in ways that negatively affected people, particularly those from vulnerable groups. “I think we’d be one of the first in the world to have this AI framework,” Champagne said at the time, suggesting it would help ensure the technology is “used responsibly.” 

While AIDA will focus on “high-impact” systems, the government will only lay out the criteria for what those are sometime in the future, in regulations that will also outline what steps anyone creating or using the technology would need to take towards being transparent and preventing undesirable outcomes, or making up for them. 

Industry executives say the lack of detail in the legislation makes it hard to assess the impact on the development and use of the technology. 

“The who, what and how are undefined to a large extent,” said Trevor Neiman, director for digital economy at the Business Council of Canada. He said the lobby group has heard concerns about AIDA from across its membership, which consists of the chief executives of the country’s largest corporations in sectors like financial services, natural resources and real estate.

Toronto-based Cohere’s platform lets businesses add AI-generated text to their marketing, products and services. Based on what the federal government has proposed to date, CEO Aidan Gomez said he has “no idea” if AIDA will apply to his startup’s technology. 

The law lays out significant penalties for non-compliance, including prison time for criminal uses of AI. It also lets the government issue fines of up to $25 million, or five per cent of a firm’s revenue; the regulations will include final numbers. That could “scare off investors who feel like their investment in a company that has a [multimillion-dollar] threat hanging over its head is at risk,” said Gomez. Cohere itself has raised US$170 million to date, per PitchBook data, and is reportedly in talks for a new round. But Canada’s broader AI ecosystem is “still super nascent,” Gomez said. “I’m afraid that we’ll stifle something that hasn’t even really begun to flourish.”

It’s not just tech startups expressing concern. “In creating this framework, you’re not regulating an industry,” said Sarah Goldfeder, manager of government relations for General Motors in Canada. “You’re regulating virtually every industry.” Car-safety systems like anti-collision and emergency braking are automated, responding to specific data with programmed actions, she noted. Those features are already regulated under motor vehicle legislation, but “could be construed to come under [AIDA’s] definitions,” even though they don’t currently operate off machine learning algorithms.

The executives with whom The Logic spoke agreed that AI systems can pose real risks, and that the federal government is right to try to address them. And not everyone disagreed with Ottawa’s approach. In making rules, “we have to account for the speed, complexity and scale of AI,” said Philip Dawson, head of global AI policy and advisory at Toronto startup Armilla AI. So “it makes a lot of sense to have skeletal legislation foreshadowing regulations.” 

The alternatives would be worse, said Dawson, who served as a senior advisor for AI and data policy at Innovation, Science and Economic Development Canada (ISED) between August 2021 and May 2022. Ottawa could take the more prescriptive approach of the EU’s AI Act, proposed in April 2021, which lists high-risk systems and sets out requirements and responsibilities for their creators and users. But the bloc is already having to adapt the legislation to address generative AI, Dawson said. Both the AI Act and AIDA predate the November 2022 launch of ChatGPT, San Francisco-based OpenAI’s viral query-answering bot.

Dawson said Ottawa could help address industry demands for more certainty by encouraging firms to assess their AI systems against existing standards in other markets like those developed by the U.S. National Institute of Standards and Technology. “Ideally, the [final AIDA] regulation shouldn’t be a surprise,” but rather “based off of this foundation of industry standards,” he said, noting similar approaches in sectors like aviation and cybersecurity.

Industry groups and executives have raised their concerns about AIDA over the last year with officials in Champagne’s office and ISED, which last month issued a companion document in which it sought to “reassure actors in the AI ecosystem in Canada” that AIDA’s aim “is not to entrap good faith actors or to chill innovation, but to regulate the most powerful uses of this technology that pose the risk of harm.” 

Gomez said the additional details in the document about the factors the government will consider in assessing AI still aren’t specific enough. And Neiman said while it’s “helpful,” it doesn’t provide much certainty because it’s not legally binding on the Liberal government or any successor. The BCC is calling for MPs to split Bill C-27 and advance the consumer-privacy portion of the legislation rapidly through Parliament, while taking the time to consult widely on and refine the AI sections. 

Businesses aren’t the only ones with criticisms of AIDA. Digital-rights groups say it’s too narrow, and should ban outright some uses of algorithmic systems. Artists and other creators have called for the law to require generative AI tools to respect their IP rights. A parliamentary committee studying AIDA could host a more public conversation about the law’s potential benefits and consequences, Goldfeder noted. Right now, “we’re all jousting against invisible opponents on this, rather than actually identifying what the concerns are.”

ISED spokesperson Doreen Flynn said the government is “committed to transparent, broad and inclusive consultations” with industry, academia and civil society “to ensure that the new [AIDA] regulations meet the expectations of Canadians.” Last month’s companion document promises a six-month period of discussions before the new rules are developed, followed by a second three-month set once they’re drafted.

Flynn did not directly address The Logic’s question about whether the government plans to propose any changes to Bill C-27 in response to feedback about AIDA to date, instead noting that amendments may be made during the parliamentary process. The companion document states that the new law will take effect no sooner than two years after the legislation passes, meaning 2025 at the earliest.