Canada needs a national healthcare AI diffusion strategy
In its new "AI for All" strategy, the government has rightly chosen healthcare as the priority area for AI development.
The initiative will be funded initially with $200 million. Of the health money announced so far, the only line items are $200M for data platforms (the Health Sector Data Space and VITAL). Nothing has been earmarked for getting tools into clinics - into diffusing these technologies to actual clinics in the system.
It is diffusion, not merely innovation, that is the major bottleneck for overhauling the Canadian healthcare system with AI-powered systems.
Consider the domain of healthcare administration & operations (front desk staff).
These are largely structured, repetitive workflows with well-defined steps – fax labelling, fax routing, filling out demographic information forms, and booking appointments – which healthcare-deployed AI agents are capable of doing.
The public healthcare system converts enormous quantities of capital into these administrative workflows which could be more effectively deployed into other critical shortages in the system that are not viable for automation.
In a recent physician survey, 43 percent of those surveyed said they would be willing to see additional patients if they could eliminate paperwork. This is a powerful rationale for pushing AI adoption.
Tools do not appear to be the problem, adoption is.
Clinics are generally small businesses. Where there is no IT staff, there is no procurement expertise, no time to evaluate vendors, and no time to integrate into the electronic medical records system or retrain staff.
Early AI adoption, such as scribe programs, has also been limited by practice differences, privacy issues, compatibility problems, and inadequate change-management support. But as long as physicians are not using AI tools, their risks are medical-legal and privacy-based.
Ottawa itself considers Canada one of the slowest countries to adopt AI at scale. In large part, this diffusion problem is a resource problem.
One government-led AI adoption example: Canada Health Infoway's AI Scribe Program vetted vendors, subsidizing their licenses for one year, and allowed clinicians to choose their AI scribe, now enrolling more than 12,000 primary care clinicians.
The nationwide evaluation of this program found that nearly 70 percent of clinicians who used the technology found they had less administrative burden. One in five clinicians used the saved time to add additional appointments.
So why shouldn't the government use its national-priority approach for intake, referral management, appointment booking, and document processing?
In 2024, Ottawa committed up to $240 million to Cohere under its Sovereign AI Compute Strategy. Its AI for All plan wants government procurement to become a principal market for Canadian technology leaders.
AI agents deployed in service of automating healthcare operations are rapidly converging into a broader form factor: an agentic clinic operating system. Soon systems like this will form the computational backbone of our entire healthcare system. It's in our national interest to ensure that these systems are indigenously developed and diffused into our providers.
Ottawa does not need to choose vendors. Rather, it could subject Canadian vendors to prequalification based on clean privacy, security, and interoperability standards. Grants to clinics would allow them to choose from among them, ensuring competition and keeping prices fair. The vendors would build on new work Canadian or open-source models wherever possible (a federally backed open-source AI initiative led by Mila with Mozilla and Hypertec).
With about 5.9 million Canadian adults without a regular primary care provider or team, each hour clinicians spend on paperwork is an hour they are not spending on a patient. The strategy names physician admin burden as a mission goal, but has no delivery mechanism for it.
It's not too late to change this.
The 24-second version:
