On the Computational Intelligibility of Boolean Classifiers