In experiments in which physicians made decisions about treating hypothetical patients, the physicians tended to trust incorrect advice presented as being generated by artificial intelligence (AI), even after given the opportunity to notice that patient recovery data contradicted the recommendations.