Should AI systems be programmed with the ability to disobey humans when they determine the human's request could cause harm to society?