Integrating levels of explanation: a Grand Challenge for the next decade of cognitive science