From AI Strategy to Execution: Trust, Basic Research, and Canada's Commercialization Dilemma

Introduction

The Canadian federal government released the "National AI Strategy: AI for All," attempting to position AI as the core driver of innovation. But whether the strategic blueprint can be implemented depends on a variable that is often underestimated: trust. Jim Banting, Assistant Vice-President of Innovation, Partnerships, and Entrepreneurship at the University of Toronto, in a conversation with Torys LLP, systematically articulated the critical path from planning to execution.

Why Trust Is the Starting Point for Execution

Banting pointed out that every major technological change in history—from the calculator to the internet—has gone through a period of skepticism. AI is following the same trajectory. The key is not to have AI replace experts, but to make it a lever that amplifies professional capabilities. Only by establishing this shared understanding can the public and businesses develop trust, and trust is the prerequisite for adoption.

Notably, AI's adoption path may differ from that of SaaS. Enterprise customers' concerns about the security of confidential data are slowing deployment at large companies. Banting believes that organizations that can solve the problem of secure, controlled interaction between AI and proprietary data will win in the execution phase. This also means that AI adoption may start with small and medium-sized enterprises, or form an entirely new diffusion path.

Four Shortcomings in Government Support

Regarding how the Canadian government supports AI innovation, Banting proposed four priority areas: first, cultivating application-oriented talent capable of deploying AI, including "pre-deployment engineers" who understand both models and implementation; second, having the government adopt AI itself, demonstrating real value to citizens through example; third, providing "bridge funding" to close the "valley of death" from research to commercialization; and fourth, training and retaining high-performance computing experts to ensure Canada has the ability to build and operate world-class supercomputing infrastructure, including providing testbeds for next-generation hardware.

He emphasized that fundamental research is the upstream of the entire innovation pipeline. Without sustained investment in basic science, the commercialization pipeline will dry up within a decade. Canada's neural network and machine learning research—such as Geoffrey Hinton's early work—is precisely the starting point of today's AI.

The Unique Opportunity of the Canadian Market

Banting believes that Ontario is one of the most noteworthy healthcare markets in the world: its population is large, highly diverse, and access to healthcare services is similar—a combination that is rare in healthcare systems. Under strict privacy protection and clear consent frameworks, unlocking healthcare datasets can yield insights into care and treatment methods that are difficult to replicate. The federal government's CAD 100 million investment in the VITAL platform is an example of leveraging this advantage to accelerate new therapies reaching patients through larger-scale clinical trials.

Startup Ecosystem: Painkillers, Not Vitamins For AI entrepreneurs, Banting advises sticking to the "painkiller" logic—solving problems users are willing to pay for—and prioritizing revenue from day one. He also reminds that the heartbeat of the Canadian economy is small and medium-sized enterprises; profitable companies that create 30 to 200 jobs are extremely valuable outcomes and should not be viewed through the binary lens of "unicorn or failure." AI is a tool that can build more such companies at a faster pace.

Global Trends: AI Sovereignty and Physical AI

On AI sovereignty, the UK and Sweden offer models worth learning from: combining hyperscale cloud infrastructure with data residency, compliance, security controls, policy enforcement, and transparency, enabling governments and regulated industries to run critical workloads within frameworks that comply with national laws and public trust. Banting believes this template of "global-scale infrastructure + national governance controls" suits Canada.

Looking ahead 20 years, software AI will gradually shift toward "physical AI," embedded in everyday environments. Large language models will become a commoditized layer, and differentiated value will be built around proprietary data, forming moats.

Strategic Significance: Long-Term Trends

Canada is in a transition window from an AI research powerhouse to an AI application powerhouse. The real strategic value lies not in building more models, but in establishing an institutional system that efficiently couples research, talent, data, and capital. Trust is not a nice-to-have; it is the gateway to the entire commercialization funnel. Institutions that can connect the four links—data security, government demonstration, bridging capital, and computational talent—will determine Canada's position in global AI competition over the next decade.