OTTAWA — Digital rights advocates and academics say the federal government’s proposed artificial intelligence law is too narrow to properly tackle the technology’s harms, and would respond to violations with a watered-down enforcement system. 

They’re calling for more public consultation and significant changes as a parliamentary committee studies the legislation. 

Bill C-27,  introduced in June 2022 by Innovation Minister François-Philippe Champagne, is the Liberal government’s second attempt to update Canada’s turn-of-the-century privacy law for the private sector. But the legislation also included a new Artificial Intelligence and Data Act (AIDA) setting requirements for developers of “high-impact” AI systems to “mitigate risks of harm and biased output.” 

The proposed law would cover damage to an individual’s physical or psychological well-being or economic interests. But policy watchers say AIDA neglects the equally important risk of collective harms. 

The proposed law “doesn’t address the human rights implications of algorithmic systems,” said Christelle Tessono, at the time a fellow at Princeton University’s Center for Information Technology Policy and now at the University of Toronto (U of T). In some cases, it may be hard to pinpoint how a specific individual has been hurt, even though an AI tool impacts an entire community.

Studies have shown that facial recognition systems are more likely to misidentify racialized people, while large language models can replicate the biases of the data on which they were trained. AIDA also doesn’t outright ban “inherently discriminatory” uses of AI, as the EU has with live biometric identification, Tessono noted.

The bill is silent, critics add, on concerns ranging from workers’ interests to society-level challenges. “It really fails to engage with a lot of concerns we’re seeing around labour displacement [and] devaluation,” said Blair Attard-Frost, a U of T PhD candidate whose research focuses on AI governance. Nor does the law address the intellectual property issue of developers using artistic works to train generative systems without creators’ permission, Attard-Frost said. This month, Innovation, Science and Economic Development Canada (ISED) launched a consultation on whether to update copyright rules to address AI. 

Some AI systems may damage democracy or the environment, said Sam Andrey, managing director of the Dais, a think tank at Toronto Metropolitan University. Researchers have expressed concern that generative systems could be used to produce disinformation on a vast scale, and that the computing power they require will have harmful climate impacts.

AIDA includes significant penalties for breaking the rules, including fines rising to $10 million or three per cent of a firm’s global revenues. A new AI and data commissioner would enforce the law. But digital rights advocates have expressed concern that the position would be housed within ISED and answerable to its minister rather than to Parliament, noting that both the department and the minister’s office have a mandate to promote Canada’s AI sector and the use of the technology. “It’s not independent,” said Tessono. 

Andrey said Ottawa needs to clarify how members of the public can lodge complaints, and what powers the commissioner will have to proactively investigate.  

Not all regulators are agents of Parliament, noted a senior government official, citing the ISED-linked commissioners of competition and patents as examples. (The government made the official available for a technical briefing on the condition they not be named.) The law ties the new AI commissioner to the department and minister to keep the enforcer close to the ongoing policy work on the technology, the official said. 

Many policy watchers also said Ottawa failed to properly engage with the public on its plan to regulate AI. “There was zero public consultation” before AIDA was introduced, said Ana Brandusescu, a McGill University PhD candidate whose research also focuses on AI governance. Brandusescu said Ottawa should launch and fund awareness campaigns, including offline engagements to allow citizens who may not be online to participate. 

ISED officials have held more than 300 meetings with business leaders, academics and members of civil society on Bill C-27 since its introduction, Champagne told the House of Commons industry committee last month, promising amendments that incorporate the feedback. 

As tabled, the legislation would leave some details—including the definition of high-impact systems—to future regulations. Ottawa has committed to consult on those rules, but startup and corporate executives have said that structure creates uncertainty that could stifle innovation. Champagne’s amendments to the bill would clarify the classes of systems AIDA would cover, including tools used by police, employers, health-care providers and the courts.

But critics say the changes the government is considering and any regulations it writes won’t address all the problems with AIDA. “That still leaves the overall structure and approach baked into the legislation,” said Teresa Scassa, a University of Ottawa law professor and member of the federal AI advisory council. Champagne has said Canada could be the first country with a regulatory framework for AI. But there’s “no prize” for being earliest, Scassa said, proposing instead a “broad national conversation on how to approach AI governance.”   

Some civil society groups say they’re focused on improving AIDA, since many MPs have signalled they’re willing to keep advancing Bill C-27. But Brandusescu and McGill professor Renée Sieber are calling for Ottawa to drop the law and start again, with new legislation that addresses the technology’s power to create monopolies and protects workers’ rights.