Friendly AI chatbots more likely to make mistakes

AI chatbots that are trained to be warm and friendly towards humans are also more likely to provide them with inaccurate information, a study has found.