Health chatbots sounded just as kind when their advice was dangerous
AI chatbots answered health questions warmly in every language, but the medical advice was much weaker in some. Doctors rated 1,008 answers; dangerous replies were 3.6% in English and 15.5% in Thai and Hebrew.
Why it matters
Millions of people now type their symptoms into a chatbot before calling a doctor. This study suggests a friendly, caring tone tells you nothing about whether the answer is safe. That is worrying for anyone who asks in a language other than English. In one test, none of the 24 answers about stroke said that a stroke needs help within minutes, not hours. If you have a serious symptom, treat a chatbot as a starting point, not a doctor.
Who's behind it: D. Ariel and colleagues, A multi-country team of doctors and researchers.
Summary by the Lemma AI · how we grade