I am still new to this topic. But one of the reasons deep learning has become popular is it requiring less data and feature engineering than classical ML techniques. Does moving to a data-centric methodology not contradict that?
I do not think that is the case.