Hierarchical in Forcing Learning

There's a field for like a subject under me was learning called hierarchical in forcing learning. And I think the realization I can tell is that maybe those things shouldn't be based into the algorithm or the structure of the agents. Maybe those things should themselves be learned from data, like what skills should it say, what should be the learning algorithms behavior. meta learning has already started becoming having some signs of life in other domains, not your robotics but many computer vision at the time.