AI Still Can't Reliably Verify Medical Claims
AI can't be trusted to check if medical advice is backed by real science yet
Researchers put eight popular AI chatbots to the test to see if they could verify medical claims. Think of it like asking a super-smart librarian to check if a health article is based on real science — but instead of books, the librarian had to use the internet and research papers.
The results showed that while AIs could often find related information, they struggled to provide solid proof from real studies. Out of 1,067 pieces of evidence checked, many were either weak, missing, or completely wrong. It's like if your doctor told you 'eating more blueberries will cure your cold,' but couldn't show you a single study proving it.
This matters because people are already turning to AI for health advice, especially when they can't easily see a doctor. If AI gives unreliable information, it could lead to bad decisions about treatments, diets, or warnings about diseases. The researchers aren’t saying AI is useless — just that it needs much tighter checks before we trust it with our health.
The team created a system to test AI more carefully for medical claims in the future. Until then, if you’re using AI to help with health questions, always double-check the advice with a real medical professional.
- Eight AI chatbots were tested on verifying 17 medical claims using real research papers
- They often found related information but failed to provide valid, reliable evidence in most cases
- AI cannot yet be trusted to safely verify medical advice or explain treatments
Why It Matters
AI health advice could lead to dangerous choices until it learns to prove its facts