Monica, you raise an important point: “we now desperately need this branch of Philosophy to guide our AI development”. I look forward to hearing more about the specific theories of knowledge creation in play, particularly among those practitioners and researchers that are coming at the epistemological questions obliquely (or even blindly).
Generally, I find metaphors like “understanding” too heavy to lift, which I think is reflected in some of the other comments. Another (lighter?) frame for understanding/reasoning might be data/models/explanations, which I think is a better fit if arguing for epistemology-first thinking.
I explored the relationship between deep learning and deep explanations in a recent post. It also touches on reductionism, relevant to your argument, as well.