Will AI replace your doctor? What a medical futurist actually says

Meskó says the honest answer is no, but that repetitive and database tasks will be automated because there aren't enough people to do them. He describes AI as a tool a physician uses to read tens of thousands of papers rather than the one or two a week a busy doctor can manage. He does say he expects AI to replace him in his own job as a futurist in about ten years.

Meskó's answer is arithmetic. He cites World Health Organization figures of about 6 million missing healthcare workers worldwide today and roughly 11 million by 2030, while the number of patients needing care keeps rising because we diagnose and treat people for far longer than we used to. About a billion people, he says, have zero access to care. In his words it is not a choice, it is a must.

It's Meskó's name for where we are now. He describes three stages: a doctor neighbor era when people discussed a diagnosis with whoever they knew, the Dr. Google era when patients could reach the same studies and websites as their physician, and the Dr. AI era, where what patients can access is clinical reasoning rather than just information. He calls that last shift unprecedented.

Meskó says his team at The Medical Futurist searched the FDA database in 2020 and found 64 AI-based medical technologies. In the episode he puts the current count at roughly 1,400 to 1,500 and expects it to pass 2,000 soon. He notes the FDA at the time didn't even have a separate process for approving these products.

The episode doesn't tell you what to do about your own health, and it isn't a substitute for talking to a clinician. What Meskó does say is that around 1.5 billion people have used a large language model and that, based on data from the model makers, roughly 60 to 70 percent of them have used it for something health related. He also warns that many of these tools are not yet validated or regulated for medical use.

Meskó's position is that patients don't need a physician's permission to look things up, and that trying to stop them was always a losing argument. He says the useful version is bringing the questions a model suggested you ask, alongside your records and any wearable data. He also expects the final decision to stay with the clinician.

Meskó describes AI scribes as tools that listen to the conversation between a patient and a physician and convert it into the exact electronic medical record format that hospital uses. He frames the upside as a doctor who isn't turned toward a keyboard during your visit. He also raises the open question of how a physician should explain that to a patient without denting trust.

Meskó points to radiology first, covering analysis of ultrasound images, CT scans and X-rays, and says the biggest and best examples have come out of that specialty. He then lists smaller specialties with specific use cases, including oncology, cardiology, dermatology and ophthalmology. He gives stroke CT analysis as a concrete example with at least two dozen tools.

Partly, and Meskó says that's the problem. He credits the FDA as the most forward-looking regulator but points out that even the FDA struggles with adaptive AI, the kind that changes with every decision it makes. He asks whether you regulate the product or the company, and says medical associations are producing guidelines but never as fast as they should.

He is a physician and scientist who publishes as The Medical Futurist at medicalfuturist.com, and he tells Nick he has spent about 15 years making forecasts professionally and speaking at medical events. His team published a 2020 study of AI-based technologies in the FDA database, and he co-authored a paper with Professor Eric Topol in Nature's npj Digital Medicine on regulating future versions of large language models. He is 41 and based in Budapest.

Meskó doesn't think physicians can learn every AI tool one by one, and he says trying would turn them into developers. He compares it to the MRI machine: most doctors can't describe how one works in technical detail, but they trust it enough to send a patient in. What he expects instead is prompt engineering and context engineering, with a large language model acting as the interface to everything else.

Meskó went through one of the most comprehensive longevity workups he knows of and says it did not hand him verdicts. He got about 100 diet recommendations and follows roughly 60 of them, and he describes the whole result as a map of routes he chooses how far to walk. His other takeaway is that with enough tests, something will turn up.