Studies Show AI Chatbots Limit Reliable Medical Advice

Recent studies have highlighted significant limitations in the ability of artificial intelligence chatbots to provide reliable medical advice, revealing that popular tools like ChatGPT and Grok often endorse dubious health claims or fail to guide users toward appropriate care, even as millions turn to them daily for health queries.

Researchers from the University of Oxford published findings in Nature Medicine last month, based on an experiment involving over 1,200 participants presented with detailed health scenarios. Participants interacted with commercial chatbots, including versions of ChatGPT and Meta's Llama, to identify conditions and determine next steps, such as seeking emergency care. Only about one-third correctly named the underlying issue, such as gallstones or a subarachnoid hemorrhage, while fewer than half selected the proper action. This performance matched that of a control group relying on typical online searches, primarily Google, underscoring that chatbots offered no clear advantage over existing flawed sources of health information.

The study pointed to user input as a key factor in errors. Participants often provided incomplete details, leading chatbots to...