AI in Healthcare: The Year in Review and The Road Ahead

To: Canadians concerned about prosperity
From: Will Falk
Date: May 7, 2025
Re: AI in Healthcare: The Year in Review and The Road Ahead

The mood at an AI in healthcare conference last month was resolutely hands on because artificial general intelligence at human levels is being reached by AI systems this year in many realms.

Recent research shows this in healthcare: AI is performing at specialty licencing exam levels. The promise that we see in the research literature is shifting into practice. Implementation challenges are being solved by clinicians, health systems, and vendors across the country.

Let’s review progress this past year and then look ahead.

Ambient AI scribes that take notes of patient visits shifted the AI conversation in 2024/25 from “interesting” to “indispensable.” The administrative burden on doctors calls for an AI point solution to scribing. Ontario already tops 10-percent adoption in primary care AI scribe usage, and a province wide bulk buy this month should bring this to 50 percent by next spring.

Infoway is procuring 10,000 additional licences to seed the rest of the country. Most clinicians will use ambient scribing in their practice by the end of the decade.

OntarioMD lit the fuse. Its evaluation – analytics by Women’s College’s WiHV with the eHealth Centre of Excellence and the Centre for Effective Practice – showed 70-90 percent documentation time savings for more than 150 clinicians, with 83 percent wanting to keep the software. 

With physician organizations – and privacy officers – all endorsing the approach, scribes became a regulatory wedge that lets AI into routine ambulatory care. It is “just a point solution” but also a pointy solution.

Vendor response to scribes has been immediate as major Canadian and US EMR vendors moved to develop scribe partnerships and AI strategies. Big players such as WELL Health and Telus Health are investing in partnerships or developing their own platforms. Interoperability will be critical.

Clinical decision support is following the same curve. Pathway.MD, OpenEvidence, and specialty specific tools are replacing yesterday’s reference mainstays – UpToDate, Google, and even Wikipedia. 

Patients are using AI every day: Automated insulin dosing for diabetes, symptom trackers for Parkinson’s, and more. These normative clinical systems rely on best evidence and clinical pathways to provide real-time recommendations.

Whether “best evidence” is local, national or global will be a fascinating discussion. Often the best answer may be patient specific. For example, a New York Medicaid patient has a different formulary and a different “best practice” than an elderly Ontario Drug Benefit patient.

Canadian hospitals and governments have matched the private sector energy. Many have active multi-year research efforts that have been underway and already had scale. Examples include:

• GEMINI (Toronto’s Unity Health) streams data from 35 hospitals – 12 billion clinical datapoints – with Quebec and Alberta joining.
• UHN has hardened data governance and green lit dozens of internal AI pilots.
• HALO (Hamilton Health Sciences + BC’s health authority) is wiring Fast Healthcare Interoperability Resources plumbing between EMRs.
• Fraser Health built an operational digital twin and is deploying agents.
• Humber River lets an AI agent triage its emergency department front door.

The flood of AI agents will now come. Based on solid foundational population data models, agents will assist with progressively more complex tasks. Starting with administration, answering patient calls and emails, care coordination, and screening, they will become clinical and more independent in the next year.

Here are several predictions for the coming year:

• AI joins the care team. First as a “medical student” or “medical office assistant” (MOA). Med student level AI will take histories, consolidate information, and may suggest differential diagnoses.
• Scope of practice enhancement. AI will upend all the skill-mix discussions as scopes of practice expand. Patients and their informal caregivers will have access to much better clinical information. Nurses, pharmacists, and other mid-level providers will increase their scopes comfortably.
• Scribes become copilots. Scribes are table stakes. By year end, documentation tools may layer in real-time coding, drug interaction checks, and differential diagnosis prompts. Pre-visit data collection and results consolidation will be in maturing products.
• Population health data. Unstructured and semi-structured data becomes more accessible as digital twins and NLP translation tools are put in place to allow for consolidated views across health systems. PDFs are “deblobbed” and become a meaningful part of the care record. Non-hospital care records become more valuable and usable.
• Agents deployed. Agentic AI – all those doctor office appointment calls not least – is in its infancy in healthcare. It is growing exponentially. Dozens in 2024, hundreds in 2025, thousands in 2026 or 2027.

AI is fundamentally changing the way that we work and deliver care. Each of us needs to commit to personal learning to understand these systems, getting our hands dirty and learning when AI works and when it doesn’t. Technology helps but human compassion and caring need to be central to everything we do in healthcare.

Will Falk is a retired management consulting partner who spends his time with start-ups, as a policy fellow at four institutions and a volunteer. This last includes AMS Healthcare where he is Board Chair.

To send a comment or leave feedback, email us at blog@cdhowe.org.

The views expressed here are those of the author. The C.D. Howe Institute does not take corporate positions on policy matters.

A version of this Memo first appeared in Canadian Healthcare Technology.

Want more insights like this? Subscribe to our newsletter for the latest research and expert commentary.