AI healthcare keynote speakers: how to choose one
I get asked this more than almost anything else. A conference chair or a chief medical officer sends me a note that says some version of the same thing: we want an AI session, our last one landed badly, who should we book.
The failed session is the part worth paying attention to. It is almost never because the speaker was unqualified. It is because the speaker was qualified for a different room.
Here is how I would think about it, written from the side of someone who has stood in front of those rooms and also sat in them.
Three kinds of AI talk, and only one is right for you
Most healthcare AI keynotes fall into one of three shapes, and the shapes are not interchangeable.
The horizon talk
Where medicine is going in five to ten years. Genomics, digital twins, ambient sensing, the long arc. This works for a board retreat, an innovation summit, or a gala. It fails in front of working clinicians on day two of a clinical conference, because they have patients on Monday and the talk gave them nothing to do on Monday.
The implementation talk
Governance, procurement, validation, workflow integration, what breaks and why. Right for a CIO forum, a quality and safety meeting, or a health system leadership offsite. Wrong for a general audience, where the detail reads as a vendor briefing.
The judgment talk
What stays human when the tools get good. Where a clinician's decision still carries weight, what accountability looks like when a model contributes to a bad outcome, and how you keep skill from draining out of a workforce that has stopped practicing a task. This one travels across audiences, and it is the one most conferences underorder.
Before you shortlist anybody, decide which of the three your audience actually needs. Most booking mistakes happen upstream of the booking.
Four questions that separate a good fit from a bad one
I would ask any speaker these, mine included.
Not a purity test. Some of the best healthcare AI thinkers have not seen a patient in years and their work is stronger for the distance. But the answer changes the talk. Someone who was on shift last month tells a different story about alert fatigue than someone reading about it. Know which you are getting, and which your room wants.
If the answer is vague, it is a horizon talk. That may be exactly what you want. Just be sure.
Anyone who cannot answer this is selling something. Healthcare AI has real failure modes, real equity problems, and real evidence gaps. A speaker with only upside will not survive the Q and A in front of a skeptical clinical audience.
Ask what they need from you to do it. A speaker who wants your incident data, your workflow pain points, and thirty minutes with two of your clinicians will give a better talk than one who wants a logo and a run of show.
What to ignore
Follower counts. A large social audience tells you someone is good at social media.
Book counts, and I say that as someone with more than forty-five of them. Volume tells you a person writes. It does not tell you they can hold a room of eight hundred people after lunch.
Vendor affiliation, in either direction. A speaker employed by an AI company is not disqualified, and an independent one is not automatically neutral. Ask who pays them and decide with that in front of you.
The word futurist, unless the futurist can also tell you what went wrong in a deployment last year.
Ten speakers worth hearing
These are people I would point a conference organizer toward, grouped by the room they serve best. I am on this list because I am one of the people who does this work, and you should read the rest of it with that in mind.
For the board and executive room
Leads Mayo Clinic Platform, with decades across clinical practice, informatics, and policy. Book him when your audience is deciding what to build and what to buy.
At Scripps Research. Wrote the book that put deep learning in medicine on most clinicians' radar. The horizon talk, from the person who largely defined the horizon.
Comes at it as a working health system CIO. If your room is rural or mid-sized and tired of hearing about academic medical centers, he is the closer fit.
For the technical and data room
At Johns Hopkins. Works on machine learning for clinical deterioration and has built systems that run in production hospitals. The right call when your audience wants model validation rather than model enthusiasm.
At Stanford. Sustained work on whether these systems actually help. Book him when your audience has heard the promises and wants the evidence.
For the governance, privacy, and security room
Health data privacy and regulation. Under-booked relative to how much trouble her subject area causes.
At UC San Diego. An emergency physician and a healthcare cybersecurity researcher. That overlap is rare and increasingly necessary.
For the clinical and change-management room
Years inside Microsoft's health AI work, and writes about organizational adoption. Strong for audiences who need the change-management half rather than the technology half.
Physician and health IT background. Works well with mixed clinical and administrative audiences.
Where I fit
Board-certified emergency physician, still working clinically. Five-time TEDx speaker, Chief AI Officer at Phantom Space, author of more than forty-five books on healthcare AI and physician leadership. I served on Singapore's Ministry of Health Regulatory Advisory Panel from 2024 to 2026. Offcall placed me at number 24 on its Top AI Healthcare Voices list for 2026.
The talk I give is the judgment talk. What a clinician still owes a patient when a model is in the room, how to tell a tool that helps from one that only looks like it helps, and what happens to a workforce that stops practicing the thing the machine now does. I speak in English and Spanish, which matters more than people expect for conferences in Latin America and Spain.
If your audience is clinical and you want them arguing with each other in the hallway afterward, that is the talk I would pitch you.
The mistake I see most often
Conferences book the AI talk as a novelty slot. Big name, big room, big screen, and no connection to anything else on the agenda.
The sessions that land are wired into the rest of the program. The keynote raises a specific question in the morning, a breakout works it through at two, and someone from the organization stands up at the close and says what they are actually going to do about it.
That structure does more for your audience than any individual speaker on any list, mine included. Get it right and a good speaker becomes a great session. Get it wrong and the best speaker in the field gives a talk your audience forgets by dinner.
The short version
Decide which of the three talks your room needs. Ask the four questions. Ignore follower counts and book counts. Wire the session into the rest of the agenda. And book someone who can tell you, without flinching, what they got wrong.
Talk to me about your program
If your audience is clinical, executive, or both, and you want a session that people argue about afterward, start here.
Affiliations above were accurate at the time of writing. Healthcare AI is a field where people change roles often, so confirm before you reach out.