Once again, medicine is being asked to do something it’s never had to do at this scale: treat a fast-moving technology like a responsible professional. The College of Physicians & Surgeons of Alberta’s updated advice on artificial intelligence in clinical practice isn’t just a bureaucratic update—it’s a candid acknowledgement that AI has already slipped into the exam room. Personally, I think this is the right kind of “slow down and think” moment: not to stop innovation, but to force clarity about risk, accountability, and consent.
What makes this particularly fascinating is how the document frames AI as both a help and a hazard, depending on how it’s deployed. Ambient scribes and documentation support tools are sold as efficiency machines, but they also reshape what clinicians see, remember, and record. And in my opinion, the hardest part isn’t the technology—it’s the human tendency to trust what looks clean, fluent, and authoritative on the screen. When medicine mistakes “readable” for “reliable,” patients become the unintended quality-control department.
AI is already in the room
CPSA’s update reflects a reality many systems can’t ignore anymore: AI is no longer hypothetical. It’s already being used to support documentation and clinical workflows, which means the conversation can’t remain theoretical.
From my perspective, this is where most public debate gets lazy. People argue about whether AI will “replace” doctors, but the more immediate issue is whether AI will quietly reroute clinical attention. A tool that drafts notes or suggests language changes the cognitive posture of the clinician—what gets summarized, what gets emphasized, and what gets omitted. That’s not sci-fi; it’s editorial power.
One thing that immediately stands out is CPSA’s emphasis on a balanced approach. I interpret that as a warning against the two extremes: “AI is magic” and “AI is evil.” Both positions ignore the messy middle where most harm actually occurs—through small errors, biased outputs, incomplete context, and overreliance.
What many people don't realize is that documentation isn’t just paperwork in modern care. Notes affect billing, handoffs, diagnoses, and later clinical reasoning. So even if AI never “touches” treatment directly, it can still influence treatment indirectly by shaping the record clinicians and systems rely on.
The accountability message is the real backbone
Here’s the part that feels most important to me: regulated members remain fully accountable for patient record content and clinical decisions, including when AI contributes to adverse outcomes. That line matters because it restores the ethical center of medicine.
Personally, I think accountability is the only language that keeps AI honest. Without it, organizations will treat AI outputs like weather—unavoidable and outside human control—while still benefiting from the efficiencies. But the moment a clinician signs off on a note, chooses a diagnosis pathway, or fails to verify an AI-generated claim, the responsibility doesn’t vanish.
This raises a deeper question: what does “verification” really mean when the AI output is fluent and time-saving? Clinicians can’t realistically re-derive every claim from primary evidence while juggling workload. Yet the standard of care demands judgment. So the real challenge is building practical verification habits that don’t collapse under administrative strain.
What this really suggests is that AI adoption must be paired with workflow design, training, and auditing—not just software procurement. Otherwise, accountability becomes an abstract moral statement while the system quietly shifts the burden of error onto clinicians and patients.
Privacy and compliance aren’t add-ons
CPSA also anchors AI usage to privacy legislation and professional standards. Specifically, it points to completing or updating privacy impact assessments (PIAs) before introducing AI tools, informing patients and obtaining consent, and documenting how and when AI is used in patient records.
In my opinion, this is where the conversation often goes sideways. People treat privacy compliance as a checkbox, but AI tools can be fundamentally different from traditional software. They may process data in novel ways, infer sensitive information, or persist outputs beyond what patients expect. If you take a step back and think about it, a PIA isn’t just legal hygiene—it’s a structured way to ask: “Where could this go wrong, and who absorbs the consequences?”
The consent piece is equally significant. Patients deserve to know when AI is part of their care—not in vague terms like “technology,” but in a way that’s meaningful. Personally, I think consent is often reduced to a formality in healthcare. But if AI influences documentation and clinical interpretation, informed consent becomes a fairness issue, not merely an administrative one.
One detail I find especially interesting is the emphasis on documenting AI use in patient records, especially when it informs diagnosis, treatment, or clinical notes. This is essentially about traceability. If a future clinician relies on a note, they should know what in that note was human-authored reasoning and what was AI-assisted drafting.
The benefits are real—so are the failure modes
CPSA acknowledges potential benefits and harms, and that balanced framing is crucial. AI can support clinical care and documentation, which can reduce burnout and improve consistency in recordkeeping. I don’t dispute that.
However, what makes this problem enduring is that the harms aren’t always dramatic. They can be subtle: a wrong detail repeated confidently, a clinically irrelevant suggestion, a missed nuance because the tool prioritized brevity, or a transcription that looks accurate but subtly misrepresents symptoms.
If you take a step back and think about it, AI risks often come from mismatch: the tool may perform well on average, while your patient’s case is not average. And medical context is rarely “standardized.” It’s shaped by ambiguity, emotional content, and incomplete information.
Another misunderstanding I see frequently is the belief that “AI makes fewer mistakes than people.” That might be true in narrow tasks, but clinical care isn’t a single task—it’s a chain of judgments. AI can lower effort in one link while raising risk in another, especially if people stop performing checks they used to do.
What this tells us about the future of care
CPSA’s update implicitly recognizes that the technology’s accuracy, reliability, and safety evidence is still evolving. That admission matters, because it sets expectations: clinicians should not treat AI as a proven clinical authority.
From my perspective, this is a transitional moment in healthcare governance. We are shifting from “software as tool” to “software as collaborator,” and collaboration changes liability, ethics, and training requirements. The public often discusses AI as a replacement for clinicians, but regulators are clearly steering toward something else: clinicians as accountable supervisors of AI-assisted processes.
I also think this foreshadows a new kind of clinical literacy. Patients and professionals will need to understand not only how to use AI tools, but how to interpret their outputs responsibly. That includes knowing when to distrust fluent language, when to demand primary-source confirmation, and how to communicate uncertainty.
A practical takeaway for clinicians and patients
CPSA is effectively telling the profession: if you use AI, you must do it deliberately, legally, and transparently—and you can’t offload your clinical judgment onto an unregulated system.
Personally, I think this is the right stance precisely because AI adoption will continue regardless of how comfortable people feel. So the question becomes: will institutions build guardrails that treat patients as stakeholders, or will they optimize first and apologize later?
Here’s what patients can look for, and what clinicians should be ready to explain:
- Whether AI tools support documentation or other parts of the clinical workflow
- How privacy protections are handled, including PIAs and data governance
- Whether patients were informed and how consent was addressed
- How AI-generated content is recorded and reviewed in the patient chart
If you want the human-friendly version of the same message, CPSA also provides resources for Albertans about AI use in healthcare. I like that emphasis because it acknowledges that trust is earned through clarity, not marketing.
In the end, this update feels less like regulation and more like a moral reminder: medicine is not just about producing an outcome—it’s about the process by which decisions are made. AI can accelerate parts of that process, but it cannot replace the ethical responsibility that sits with licensed professionals.
Would you like me to tailor this article toward a specific audience—clinicians, patients, or policymakers?
