If we want to figure out the proper way to control A.I. models, we need to recognize what they really are.