By Dr Jacques Conradie — Neurosurgeon
“AI Passes U.S. Medical Licensing Exam.” “Microsoft’s AI Is Better Than Doctors at Diagnosing Disease.” Elon Musk recently told students that medical school is “pointless” because AI will soon outperform doctors.
Read enough headlines like these and it’s easy to conclude that AI has already surpassed the average clinician.
Some stories seem to support that idea. In 2023, a widely reported case described a boy who saw 17 doctors over three years without a diagnosis until ChatGPT suggested tethered cord syndrome. A neurosurgeon later confirmed it.
There is real substance behind the headlines. Reduced to its simplest form, diagnosis is information processing. A clinician takes a history, performs an examination, reviews blood tests and imaging, then weighs probabilities to reach a diagnosis. This is exactly the kind of task modern AI excels at. Given enough high-quality data, it can analyse far more information than any human can hold in working memory. It does not get tired after a long shift or forget a rare disease because it hasn’t seen one for years.
But diagnosis is only one part of medicine.
Patients are not datasets waiting to be analysed. They are frightened parents, teenagers embarrassed to discuss sensitive symptoms, older patients with hearing loss and limited finances, and families struggling to agree on treatment. Good medicine means navigating those realities as much as interpreting blood results or scans.
No AI model has to explain to a family why surgery is no longer an option, or sit quietly with someone after a cancer diagnosis while they process what comes next. Those moments are as much a part of healthcare as making the diagnosis itself.
The phrase that matters is human in the loop. The machine assists. The human decides.
Medicine has always relied on tools. A stethoscope, CT scanner or ultrasound does not replace a doctor. It extends the doctor’s ability. AI should be viewed the same way. Used well, it can improve decision-making, reduce errors and free clinicians from repetitive administrative work. Used poorly, it can produce answers that sound convincing but are completely wrong.
AI will make mistakes. It will hallucinate. Like every investigation in medicine, its output must be interpreted in clinical context. Radiologists have been reminding us of this for years with the familiar phrase: correlate clinically.
That is why I do not believe AI will replace doctors. I believe it will replace doctors who refuse to learn how to use it.
The clinicians who thrive over the next decade will not necessarily be those with the best memory. They will be the ones who know when to trust AI, when to question it and when to ignore it.
The real test of any healthcare technology is simple. Does it improve patient outcomes? Does it reduce errors? Does it save clinicians time or help patients better understand their care? Does it fit into existing workflows instead of creating more work?
Healthcare does not need more hype. It needs tools that solve real problems. Measured against that standard, many AI headlines are still ahead of reality, while others, like AI-powered toilets analysing bowel habits, are little more than novelty.
AI is arriving in medicine whether we are ready or not. The responsibility for the patient, however, still rests with a human being. The machine can assist, but the decision, and the accountability, remain ours.