@arstechnica these engines are trained on human interactions, and we rarely criticize an idea on the outset. If people are only getting answers they want to hear from these systems, the people need to be trained to ask better questions, or more questions, to push the AI to provide criticism and point out weaknesses.