Inductive Biases for Deep Learning of Higher-Level Cognition