Essays on AI, healthcare, and the professions that hold them together.

Every experienced clinician has learned to recognize the moment a patient's presentation doesn't fit the pattern — the subtle wrongness that precedes a diagnosis no algorithm would have surfaced. That capacity for productive uncertainty may be the most important thing AI cannot replicate, and the most important thing we must not train out of the next generation of clinicians.

Health professions educators face a genuine dilemma: prepare students for a world in which AI tools are ubiquitous, without producing graduates who cannot function without them. The answer is not to ban AI from the classroom. It is to make the evaluation of AI a core clinical competency.

Training data reflects the world as it was, not as it should be. When that data encodes decades of underinvestment in certain zip codes, certain populations, certain bodies — the model learns to replicate the disparity. Understanding this mechanism is not optional for anyone deploying AI in a clinical setting.

The liability question in clinical AI is not yet settled, and health systems are deploying tools faster than the legal and ethical frameworks can keep pace. This essay maps the accountability gap and argues for a governance model that places responsibility where the judgment actually lives — with the clinician.

There is a version of clinical AI that makes medicine more equitable, more accurate, and more humane — one that functions as a rigorous second opinion rather than a replacement for clinical reasoning. Getting there requires being honest about where current tools fall short.

The conventional wisdom is that AI will allow clinicians to specialize more deeply, offloading the cognitive work of breadth to algorithms. I think this is exactly wrong. The clinicians best equipped to work alongside AI are those with the widest base of knowledge — because they are the ones who can catch what the model misses.

Every time a health system signs a contract with an AI vendor, it makes a policy decision — about what evidence standards are acceptable, what transparency is required, what accountability looks like. Most health systems do not treat it that way. They should.

The overwhelming majority of clinical AI research focuses on physician decision-making. But nurses make the majority of clinical observations. Pharmacists catch the majority of medication errors. Allied health professionals deliver the majority of rehabilitative care. The gap in the research is not a technical oversight — it is a reflection of whose work we have historically valued.