I really suspect this is how the plateau of productivity will look for machine learning. It will be all about building synergy with conventional algorithms, which can provide the rigour, transparency and reproducibility trained models can't. Computational efficiency sometimes, too, although maybe not in this case.