Helen Mirren Says She Uses AI Instead of Going to the Doctor. That’s Not the Part That Interests Me.

Helen Mirren said something recently that made me think about a part of medicine we rarely discuss. The 81-year-old actor was talking with her A Talent for Murder co-stars at the Toronto International Film Festival when the conversation turned to health. Olivia Cooke and Alden Ehrenreich described themselves as hypochondriacs. Mirren put herself at the opposite extreme: someone who tends to ignore health concerns and avoid doctors. When Ehrenreich joked that he would like to have a doctor living in his house, Mirren pointed to something many of us now have instead: AI. She described it as an extraordinary place to take a question when something hurts or feels different.

The obvious question is whether that’s safe. But another question interests me more. I have spent 36 years taking care of people with cancer. By the time someone reaches my office, an enormous number of decisions have already happened. Someone noticed something. They decided whether it mattered. Perhaps they waited. Perhaps they searched online. Eventually, they decided to tell another human being. Medicine usually enters the story after that decision. AI is beginning to enter before it.

What happens before you arrive

When you wake up with an unfamiliar pain, the choices available to you have changed significantly. For most of human history they were limited: ignore it, ask someone you trusted, or seek medical care. Then the internet gave us another option, and millions began typing symptoms into search engines. Now AI alters that interaction in a way that matters. A search engine gives you pages to sort through. An AI system gives you a conversation. You can describe where something hurts, how long it has been happening, what makes it worse, what medications you take, what you are worried it might mean. You can ask another question, clarify an answer, ask again. It can feel remarkably close to consultation.

I don’t think the answer is to tell people to stop asking AI about their health. People have always sought medical information outside doctors’ offices, and AI can be genuinely useful for some things. It can explain an unfamiliar laboratory result in ordinary language. It can help someone understand the difference between two treatments. It can generate questions to take to an appointment or help make sense of something a physician said, something the patient only half-absorbed while frightened or tired. Medicine asks people to absorb complicated information precisely when they may be overwhelmed, and a tool that helps close that gap has real value.

But there is a threshold where explanation quietly becomes reassurance, and that’s where the problem begins. A symptom does not have to disappear for someone to delay seeking care. It only has to become sufficiently explainable. A persistent cough becomes allergies. Fatigue becomes stress. A change in bowel habits becomes diet. A new lump becomes something probably benign. And most of the time, the reassuring explanation may actually be right. Most headaches are not brain tumors. Most back pain is not metastatic cancer. Most strange sensations that appear during an ordinary week will never become dangerous. That is precisely why triage is difficult. The challenge is not simply identifying the most likely explanation. It is recognizing the uncommon situation in which the most likely explanation is wrong.

When I meet a patient, I am not simply processing the words they say. I am trying to understand the story around those words. How quickly did this change happen? Does the symptom fit the rest of their medical history? Is there something about their age, medications, previous illness, or family history that changes what an otherwise ordinary symptom means? Sometimes the most important information emerges only because I ask a question the patient did not know was relevant. But here is the vulnerability built into any conversation with an AI system: much of what it knows about you is what you decide to tell it. You may not know what matters. And a confident, coherent answer can feel more complete than the information behind it actually is.

For years, physicians worried about patients arriving with information they found on Google. I suspect we are entering a much more consequential era. Patients may increasingly arrive after having already had a long conversation about their symptoms — with possible explanations, questions they want to ask, interpretations of laboratory results, and some sense of how urgently they should seek care. Sometimes that will make the medical encounter better. The patient may be better informed, the questions sharper, medical language less intimidating. A person who might otherwise have ignored a warning sign could even be persuaded to make an appointment. But the opposite can happen too. The conversation may reassure someone enough that the medical encounter never happens. Once AI becomes the first opinion, it can influence whether there is ever a second.

I write about the signals we often miss in our health and our lives. If you’re curious about what your own body may be telling you, my free 3-minute Capacity Checkup can help you see where strain may already be showing up in sleep, mood, focus, and body.

But after 36 years in medicine, I think one of the most valuable things a good clinician does is not simply answer the question a patient brings into the room. Sometimes it is recognizing that we should have been asking a different question all along.