What I read from the article was that high-level models are more than just compression - a device for a La-Place-demon to cut on computation time. But I also h

What I read from the article was that high-level models are more than just compression - a device for a La-Place-demon to cut on computation time. But I also have a strong suspicion that this is not what they meant; maybe someone can clarify? In the other case (my interpretation being in line with the intent of the authors): How do they justify this?