This is a bit of a reductionist viewpoint. Perhaps you’re disregarding the emergent properties of fitted functions at scale.