Judith Miller got her imaging results last year. She didn’t just file them away. She asked Claude.
The AI, built by Anthropic, gave her a breakdown. Miller walked into her follow-up armed with questions. She felt ready. The chatbot “enabled me to better understand,” she said. It wasn’t about replacing her doctor. It was about shared decision making.
This is the new normal.
Polls say a third of adults have asked an LLM to explain labs or diagnose symptoms. Robert Wachter, a doctor at UCSF, thinks it doubles every year. “I suspect they’ll double again next year.”
Claude is not designed or marketed for clinical diagnosis.
Anthropic agrees with that stance. The tool prepares you. It doesn’t treat you.
Why are we doing this? Because of data.
The 21st Century Cutes Act dumped our records online. Test results. Clinical notes. A glut of jargon. Dave deBronhart knows the feeling. You read the report. The big question hangs there. What does this mean?
Previously, you had to wait for a doctor to translate it. Now? Chatbots do it in seconds. Fear drops. Anxiety eases.
But they’re wrong sometimes.
LLMs hallucinate. They confirm biases. They present guesses as facts. Cait DesRoches warns about the lack of guardrails. There isn’t much research on what happens when we treat an algorithm like a clinician.
“I don’t think we have any real idea how well this works for the average patient.”
The risks are real. One Seattle man died recently. Leukemia. Treatable. He refused care because AI told him he had a rare complication he didn’t actually have. That is a bad ending.
A Nature Medicine study found users got diagnoses right only a third of the time with help from models.
Should we stop using them? Most experts say no.
Use them with your eyes open. Adam Rodman calls LLMs the best tool for empowerment ever invented. With a huge caveat: you must use them appropriately.
Here is how people try to fix the broken bits.
- Pretend the bot is a doctor. It collects data differently.
- Ask it to rethink its answer.
- Get a second opinion from a different model.
- Never type in your social security number. Protect your privacy.
The goal? Better questions for the real human doctor.
Wachter calls this healthy. But it eats time. He spends ten minutes debunking Dr. Chatbot before the appointment is even halfway through. “I’ve got fifteen minutes. ”
For the uninsured or those stuck on waiting lists? The AI might be all there is.
Laura Adams says access is at a crisis level. Don’t compare the bot to a perfect physician. Compare it to nothing at all. It beats having no care.
The horse is out of the barn. We need literacy now. Not ignorance. Education. Teach people how to do it right.
New models are coming. Maybe board certified ones? Wachter suggests it could happen.
For now? Be skeptical. Miller knows the truth. The chatbot stitches words together from statistical patterns. It is impressive. It is also just predicting the next probable word.
“It’s not absolute truth,” Miller said.
