How realistic would AI-engineered chatbots be? — LessWrong