Canada's AI strategy focuses on trust and transparency. But for whom?

AI and algorithmic systems are often invisible to the people they affect, and not entirely understood by the people that use them either. Those buying and using these automated and generative systems are not always aware of the subtle and not-so-subtle ways they can inform decisions of public importance. And rarely are the risks of these systems advertised to those affected by the decisions made with them. We found it encouraging, then, when Canada's new National AI Strategy named trust as its north star, and transparency as a strategic objective.

One aspect in particular drew our attention: the emphasis on the potential for public-sector procurement processes to become a lever for transparency. By embedding requirements for trust and transparency in the process of scoping, designing and implementing AI, it increases the chances of actually empowering the people these systems affect and are intended to serve. It's a strategy we’ve seen emerge in our work implementing AI and smart city transparency tools with cities and public agencies, and that we want to bring to the conversation with digital government innovators at FWD50 in Montreal on July 7 next week.

Two audiences, two design problems

As Cory Doctorow likes to repeat: it’s important not just to interrogate what a technology does, but who it does it for, and who it does it to.

Most transparency commitments are written for those buying and using technology. The consumer choosing a product. The business buying a tool. The agency procuring a service. Those audiences need trust signals like labels, certifications, watermarks. Quick, low-information, comparable ways to make a decision.

But those on which technology acts are rarely given the same information. The resident walking past a sensor. The patient whose triage was assisted. The worker whose performance was scored. They need legibility. They need to know what systems have acted on them, why, and how they work. They need to know what they can do if something is wrong, who to talk to, and they need all of this in plain language.

Trust signals and legibility look similar on a strategy document, but they're very different design problems in the real world.

Trust signals vs legibility

Trust signals are designed for someone choosing between options. They compress a lot of judgment into a small mark. Legibility is designed for someone who didn't get to choose. It has to meet them where they encounter systems – in a hallway, at a kiosk, on a form – and give them an opportunity to understand and contest.

The procurement opportunity

The Canadian strategy leans hard on trust signals, and frames legibility as aspirational. Without required documentation, point-of-collection notices, or contestation pathways for people affected by AI systems, the strategy frames transparency as deployer-centric, not quite the trust and transparency “for all” it aspires to.

Procurement can provide the mechanism to require legibility alongside trust-signals. The strategy commits the government to lead by example through public-sector procurement standards on transparency, privacy, and accountability. Procurement is one moment where legibility can become a design requirement built into the deployment of AI and algorithmic systems from the start. It's one of the only places in the strategy where the needs of those whose lives are impacted by AI can be enforced, not just those buying and using AI.

Building transparency practices that foster trust

We've spent the last 5 years working with cities and public agencies across North America, Europe, and Australia to design community-facing disclosures that procurement language alone can't produce. The result of that work is the open source standard Digital Trust for Places & Routines (DTPR)–recently extended to describe AI systems–is a free, deployment-tested taxonomy of icons, plain-language explanations, and feedback pathways for the people on the receiving end of technology in shared spaces.

A few things have shown up consistently across deployments:

• Skepticism targets opacity, not technology itself.
  Across DTPR focus groups with members of the public, 100% of participants raised concerns about AI before being shown any disclosure. But once given structured, plain-language information about what a system does, who's accountable, and what data is involved, participants shifted from blanket discomfort to specific, informed questions. In one focus group scenario involving a computer-vision deployment, 11 of 12 participants initially rejected the technology for lack of detail. After reviewing the a DTPR disclosure 12 of 12 said it directly answered the questions they already had, and 10 of 11 moved to positive or neutral acceptance.
• When people see disclosures using DTPR they support it, and they want more of it.
  In intercept surveys conducted across multiple DTPR deployments (n=473), 78% of residents said DTPR helped them understand the technology in front of them, and 73% were more supportive of its use after receiving that information. In one community we worked with, 94% said signage and webpages helped them understand the technology's purpose, and 80% said they wanted DTPR-style transparency extended to other technologies in their city.⁠⁠⁠⁠
• Legibility creates real accountability, not just awareness.
  In one city deploying DTPR across 25+ municipal technologies, a resident identified an inconsistency and successfully pushed the city to change a data retention policy. The legibility provided by the DTPR disclosure brought actionable transparency, leading to governance change.
• It works in high-stakes trust environments.
  In a public park where an AI-enabled crowd-measurement system was deployed, residents objected to not knowing how the data would be used, for how long, and who would have access. Structured DTPR disclosures gave the operator a basis to set clear boundaries on use, communicate them publicly, and earn the social license to keep operating. The regional transport authority subsequently issued formal government guidance recommending DTPR to local councils and public agencies.

Come find us at FWD50

If the north star of Canada’s AI policy is trust and transparency, and if procurement is the best leverage point for bringing legibility for those impacted by the systems, not just those buying the systems, then the next step is empowering those preparing to procure AI right now. Canada's transparency commitments can become real in the hands of the federal and municipal teams already scoping AI use cases and writing procurement language. They just need the right tools.

We're convening an in-person workshop with public-sector practitioners working on this Tuesday, July 7 at FWD50 in Montreal. If that’s you, we’d love to meet. We’ll be curious where AI transparency is hardest in your work. We’ll share what we’ve learned working on technology disclosures around the world, and walk through what a legibility layer that drives accountability could look like in the context of AI. If that's the work you're doing, come find us at FWD50.