Studies Expose AI Chatbots' Flaws in Medical Advice

New studies have revealed significant flaws in the ability of artificial intelligence chatbots to provide reliable medical advice, raising concerns about their growing role in healthcare guidance. Researchers from the University of Oxford and the Icahn School of Medicine at Mount Sinai found that leading AI models often propagate misinformation, perform no better than standard internet searches in helping users make health decisions, and vary dramatically in responses based on subtle changes in input. The findings, published this month in Nature Medicine and The Lancet Digital Health, underscore the risks as surveys indicate that about one in six adults now consults AI tools for health information at least monthly.

The Oxford study involved more than 1,200 participants in Britain who were presented with detailed medical scenarios, including symptoms of conditions ranging from a common cold to a life-threatening brain hemorrhage. Participants interacted with commercial chatbots such as OpenAI's ChatGPT and Meta's Llama to determine next steps, like seeking emergency care or self-treating at home. The results showed that users identified the correct condition only about 34...