[ganfortran]

> Previous ML methods usually resulted in models, or formulas, that were small enough to "grasp" intuitively.

Which is not true in practice. SVM in reality is an instance based method, with thousands to 10 thousands high-dimensional vectors as its 'parameters'. Random forests, which is famous for its ease of interpretability is no easy meal either when you end up with 100s of trees with outrageous branches. Not mention that real world model are embarrassingly complex ensemble of smaller but still complex models.