Medical chatbots can already answer questions and suggest preliminary diagnoses. But do we know how to use them safely and with critical judgement?