Prime Minister Mark Carney, with Artificial Intelligence Minister Evan Solomon, announces the federal government's new AI strategy in Toronto on June 4.Cole Burston/Reuters
Helen A. Hayes is a PhD candidate at McGill University, where she researches the social, economic and political impacts of AI in Canada.
Last week, the federal government released Canada’s national AI strategy. It begins with a striking premise: Canada ranks near the bottom among advanced economies on measures of public trust in artificial intelligence.
The strategy is right to identify this as a problem. But, it’s wrong about how that problem should be addressed.
Throughout the strategy, trust is treated largely as a function of familiarity. If Canadians learn more about AI, use it more often, and participate more fully in an AI-enabled economy, then trust will necessarily follow. Accordingly, the strategy places considerable emphasis on literacy, workforce development, skills training, and the acceleration of AI adoption across Canadian sectors.
There are two major problems with this approach.
First, it treats trust as a prerequisite for adoption rather than an outcome of governance. This logic follows from the strategy’s broader economic objectives. The document links AI adoption to productivity growth, labour-force development, and competitiveness, and from this perspective, public distrust becomes a barrier to economic transformation. So, if Canadians are reluctant to use AI, businesses will be slower to adopt it, public institutions will struggle to integrate it, and the economic benefits the strategy promises may fail to materialize. The solution, therefore, is to increase familiarity. And in the government’s view, more exposure will lead to greater comfort and quicker adoption.
This may be a coherent theory of change. But, it is the wrong one for Canada. It applies a traditional way of thinking about trust to a category of technological change where familiarity alone is unlikely to be sufficient.
Federal AI plan commits billions to fund startups, provide training and boost adoption
Why? Well, the AI systems the government hopes to invest in and accelerate adoption of are not simply consumer products that Canadians can choose to use or not. They are predominantly AI systems that will be integrated into workplaces, education systems, public services, and health care. In these contexts, public acceptance depends less on familiarity than on legitimacy.
This reveals the second problem with the strategy’s approach: its emphasis on literacy. Literacy assumes that trust emerges from understanding. But many of the AI systems Canadians will be asked to trust – the ones in those critical sectors – are not ones they will ever be in a position to meaningfully evaluate for themselves. A patient cannot independently assess the technical reliability of an AI-assisted diagnostic tool. A resident cannot audit the algorithmic systems used to allocate public services. An employee cannot evaluate the models informing workplace performance evaluation. The challenge in each of these cases is not a lack of knowledge, and literacy will not solve the skepticism that Canadians have of those technologies.
Under these conditions, trust depends on confidence that governments have done the work to make AI systems safe on their behalf. They need assurance that systems have been tested, that risks have been assessed, that oversight exists, and that meaningful recourse is available when harms occur. Canada has none of these safeguards, and the strategy does not inspire confidence that those governance mechanisms will be developed, let alone put in place, before this major AI roll out.
This is precisely why the strategy’s theory of change breaks down. Large-scale institutional adoption often depends on trust already being present. People are far more likely to accept technological transformation when they believe robust safeguards exist, risks are being managed, and institutions can be held accountable. For Canadians, that trust does not exist, and the strategy’s approach to building it might very well backfire.
So, the challenge for the Canadian government is now to create the conditions under which the trust they apparently want (and certainly need) becomes warranted. The long-term success of its AI ambitions, whether measured by productivity growth, public-sector modernization, or technological sovereignty, will not depend on how effectively Canadians are taught to use AI. It will depend on whether the government can demonstrate that AI systems are subject to meaningful oversight, accountability, and public interest governance. So far, it hasn’t done that.