The study of deep comprehension and learning traditionally employs behavioral methods and also makes use of instructional, computer-based training programs. By contrast, the neuroscience approach to discourse comprehension mainly uses neuroimaging techniques to reveal the neural machinery of language. The convergence of both approaches is a promising new field. We posit that deep comprehension, as a complex evolutionarily recent skill, recycles several pre-existent neural mechanisms broadly distributed and interconnected in the brain, including sensory-motor, theory of mind, emotional, and self-relevance networks. In addition, deep learning recruits control and executive brain systems to implement self-explanation strategies.