This is the first first article in an ongoing series by Dr. Andrew Maxwell, the Bergeron Chair in Technology Entrepreneurship in the Lassonde School of Engineering at York University. Every other week – and occasionally every week – we’ll present a new article by Maxwell, in a series whose wide-ranging and incisive themes encompass: Canada and innovation policy; productivity and industry; innovation frameworks; AI and higher education; research and intellectual property; technology adoption; entrepreneurship and commercialization; universities and higher education; entrepreneurship education; and AI and the future of work. 

We keep hearing a familiar explanation for Canada’s slow adoption of artificial intelligence: Canadians don’t trust AI.

Depending on the survey, the number varies. Sometimes it’s framed as low confidence. Sometimes it’s described as skepticism. But the conclusion is always the same: if only Canadians trusted AI more, adoption would accelerate.

It’s an appealing explanation. It’s also misleading.

Canada’s AI challenge is not primarily a trust problem. It is a misdiagnosis. More specifically, it is a misunderstanding of how trust is built  - and how adoption actually happens.

When adoption is low, the instinct is to assume that people are skeptical of the technology itself. The proposed solutions follow naturally: increase awareness, highlight success stories, provide training, promote benefits. These approaches assume that trust is a static attitude that can be increased through persuasion.

But adoption rarely works this way. Trust is not something that appears fully formed. It develops gradually, through experience, feedback, and learning. Organizations do not move from distrust to trust in a single step. They move through experimentation, refinement and increasing reliance.

When we treat trust as a single number, we ignore this process – and risk addressing the wrong problem.

When organizations consider adopting AI, they do not ask a single question. Instead, they make a series of behavioural judgments.

Can we rely on it?
Should we share our data?
Does it align with our objectives?
Can we explain its decisions?
Will people actually use it?

These questions shape behaviour. They determine whether individuals disclose information, rely on outputs, accept recommendations and integrate AI into workflows. Adoption depends on these behaviours – not just on positive attitudes.

This is why measuring general trust is misleading. An organization may believe AI is capable, yet still hesitate to rely on it. Employees may be receptive, but reluctant to disclose data. Leaders may see potential, but unable to justify decisions. These behavioural gaps prevent adoption even when overall sentiment appears positive.

Understanding adoption therefore requires unpacking trust into behavioural components. These fall into four broad categories: trustworthiness, capability, trusting behaviours, and communication.

Capability reflects whether AI is perceived as technically able. Trustworthiness reflects whether it is seen as appropriate and aligned. Communication relates to explanation and transparency. But the most critical category for adoption is trusting behaviours – reliance, disclosure, receptiveness and openness. These determine whether AI is actually used.

Trust behaviours develop through learning

A key insight often overlooked in discussions of AI adoption is that trust behaviours are not fixed. They develop through learning.

When individuals interact with early versions of AI solutions, they observe performance and limitations. They refine expectations. They provide feedback. Over time, reliance increases, disclosure expands and receptiveness improves. Trustworthiness strengthens as consistency improves, and communication deepens as explanations become clearer.

In other words, trust is not simply a prerequisite for adoption. It is often an outcome of iterative experimentation.

When organizations lack structured opportunities to learn, trust behaviours remain weak. Adoption stalls not because people distrust AI, but because they have not yet built confidence through experience. This distinction matters. It shifts the conversation from attitudes to actions, and from persuasion to learning.

This behavioural perspective helps explain why Canada’s adoption rates are often lower than those of some other countries. The difference is not primarily capability. Canada has strong AI research, talent and technical expertise. Many organizations already believe AI can work. Capability-related trust is reasonably strong.

The difference lies in how trust behaviours develop. Canadian organizations tend to operate in contexts that emphasize accountability, alignment and risk mitigation. Decision-making often involves multiple stakeholders. Leaders must justify choices. Reputational risk is carefully managed.

These characteristics are not weaknesses; they reflect legitimate priorities. But they also raise the threshold for adoption.

When reliance requires high confidence, organizations delay dependence on AI. When disclosure involves uncertainty, data sharing is cautious. When explanation is essential, deployment waits until transparency improves. The result is a pattern of exploration without full integration. Organizations experiment, but hesitate to operationalize.

At the same time, Canada has a structural advantage that is often overlooked. The country has strong collaborative infrastructure – superclusters, innovation networks and industry partnerships designed to support shared learning. These structures create opportunities for collective experimentation, where organizations learn together rather than in isolation. If leveraged effectively, this collaborative environment can accelerate the development of trust behaviours.

From pilots to prototypes

A critical factor in building behavioural trust is how experimentation is framed. Many organizations rely on pilots, which often imply testing a near-final solution. Pilots create pressure for success and discourage iteration. When pilots fail, confidence declines.

Prototypes offer a different approach. They are explicitly designed for learning. They assume imperfection and invite feedback. They evolve through iteration. This framing changes behaviour. Users are more willing to share data, highlight limitations, and engage in refinement. Receptiveness increases because individuals see their input shaping the outcome.

Prototypes also mitigate risk. Instead of committing to full deployment, organizations explore limited applications, gather evidence and refine approaches. Each iteration strengthens capability perceptions while building experience. Over time, reliance develops naturally. Adoption becomes the result of accumulated learning rather than a leap of faith.

Adoption requires acting under uncertainty, which implies the possibility of failure. In environments where failure is unacceptable, organizations delay adoption until risks appear eliminated. This often means waiting for perfect solutions – which rarely exist.

Safe-to-fail experimentation provides an alternative. By designing prototypes that limit exposure, organizations can test AI in controlled contexts. When failure is framed as learning, disclosure increases, receptiveness strengthens and reliance grows incrementally. Communication improves as assumptions and limitations are discussed openly. Trustworthiness strengthens through iterative refinement.

The goal is not fail-safe technology, but safe-to-learn adoption.

Sharing both successes and failures plays a critical role. When organizations openly discuss what did not work, they reduce uncertainty for others. Risks become visible and manageable. Learning accelerates. This transparency strengthens behavioural trust across the ecosystem, not just within individual firms.

Collaboration as a trust multiplier

Canada’s collaborative structures create a unique opportunity. Superclusters and innovation networks can act as platforms for shared prototyping, collective learning and dissemination of insights. Organizations can experiment together, share data in controlled environments, and learn from each other’s experiences.

Collaboration reduces individual risk. It accelerates experience. It strengthens reliance and disclosure behaviours. Most importantly, it transforms adoption from isolated experimentation into coordinated learning.

This approach aligns with Canadian values of accountability and transparency. Rather than abandoning caution, it complements caution with mechanisms that enable action.

Canada’s AI adoption gap does not reflect a lack of belief in the technology. It reflects hesitation about acting under uncertainty. Trust exists, but the behaviours required for adoption are still developing.

By understanding trust as behavioural and learning-based, the path forward becomes clearer. Education about adoption processes, prototype-based experimentation, safe-to-fail learning, two-way communication and collaborative structures all help build the trust behaviours that enable deployment.

The challenge is not convincing organizations that AI works. It is creating environments where relying on AI becomes the logical next step.

Trust is not missing.

It is being built – and adoption will follow. 

R$