However, up until now there are no empirical studies that allow for a detailed comparison between high-level human learning of complex routines and the training input and the induced programs of IP systems, to see whether they lead to similar solutions and, when they diverge, to see whether the IP solution is still comprehensible to humans.