Canada’s AI bill needs to catch up to get ahead of the curve

At the Collision tech conference in Toronto earlier this month, Canada’s Minister of Industry François-Philippe Champagne said that he thinks Canada is “ahead of the curve” with its approach to governing artificial intelligence (AI). If only that were so. The Artificial Intelligence and Data Act (AIDA)—one part of Bill C-27, Canada’s Digital Charter Implementation Act—is a necessary first step in regulating AI, but ahead of the curve it is not: it is empty legislation in search of substance. What little substance it does have relegates fundamental human rights as secondary to corporate and commercial interests, when it should be doing the opposite. Regulating AI is a pressing policy issue, and jurisdictions around the world are rising to meet the moment. Last month, after two years in development, the European Union passed a draft of the AI Act. Late in 2022, the United States published a Blueprint for an AI Bill of Rights that while not legislation, provides guidance and a roadmap for America's responsible use of AI. What the EU’s AI Act and the White House’s Blueprint have in common is their focus on civil and human rights, and the consequences that unregulated AI can have on actual people. If Canada is to get ahead of anything, its AI bill first needs to catch up, finding a way to prioritize the public over the private. One of the ways AIDA downplays civil and human rights is in the narrow and restrictive way it characterizes ‘harm.’ AIDA defines harm as “(a) physical or psychological harm to an individual; (b) damage to an individual’s property; or (c)economic loss to an individual.” This neglects group- or society-based harms, like AI’s role in violating of Canadians’ privacy. Privacy is important to Canadians; according to a June poll by the Office of the Privacy Commissioner of Canada, 93 per cent of Canadians expressed some level of concern about the protection of their privacy. Without codifying privacy as a fundamental right worthy of respect and priority, there is little in AIDA to protect people from AI-driven privacy harms like data breaches or mass surveillance. AIDA’s definition of harm also fails to capture the serious concerns experts have about AI being discriminatory. AIs often express serious biases, as we have seen through automated decision-making that discriminates based on race, gender, or socioeconomic status. AIDA’s definition of bias even includes an exception that asks whether an AI-driven decision that “adversely differentiates, directly or indirectly" towards a person based on prohibited grounds of discrimination has adequate “justification.” This is inexcusable; not only does AIDA not define what type of bias they regard as justified, but the government shouldn’t be giving private enterprises that use AI a “Get Out of Discrimination Free” card. Parliament needs to understand that privacy violations and biased decisions are harmful to rights and freedoms, even if they don’t fit within AIDA’s idea of harm as something that only hurts your body, mind, or wallet. Bill C-27 acknowledges the importance of human rights and fundamental freedoms in its preamble, but does little to turn that acknowledgment into responsible legislation. With Bill C-27 going to the House Industry and Technology Committee in the fall, MPs and committee members have the summer to think about how they can better modernize our approach to governing AI. They should start by recognizing the risks AI poses for civil and human rights—something that the Canadian Civil Liberties Association has expressed in our open petition calling for more rights-based AI legislation, and as part of our work with the Right2YourFace Coalition on regulating facial recognition technology. In the meantime, Canada can crib language from our neighbours to the south, whose Blueprint for an AI Bill of Rights lays out, in strong terms, that AI innovation “must not come at the price of civil rights or democratic values.” Canada can also look across the Atlantic to the EU’s AI Act, which outright bans intrusive and discriminatory AI systems like real-time biometric identification systems, on grounds that they violate human rights and the right to privacy. It’s no surprise that the innovation minister thinks Canada’s AI regime is ahead of the curve on innovation and industry, but Canada’s behind the curve where it really counts. Daniel Konikoff is the interim director of the Canadian Civil Liberties Association's Privacy, Technology and Surveillance Program. He is also a PhD candidate at the University of Toronto's Centre for Criminology and Sociolegal Studies, where he is completing his dissertation on the governance of police technology. The Hill Times