We like traditional generalisations because (a) we can understand them; (b) they often enable deductive conclusions; and (c) we can apply them to particulars. But (a) an MLM’s generalisations are not always understandable; (b) they are statistical, probabilistic and primarily inductive; and (c) literally and practically, we usually cannot apply MLM generalisations except by running the machine learning model that resulted from them.
This move away from certainty to a probabilistic understanding of outcomes has an impact on our conceptions of knowledge. For some like Ayad AKhtar uncertainity is actually a good thing, however this is still somewhat corrupted by machine learning and the way it warps our minds.