This post deserves a more engaged response. I believe deep learning (and statistical/machine learning) presents a good reason for introspection about current linguistic theory, in part because I find the computational model espoused in neural networks to be more biologically plausible than single CPU symbolic computation.

The government and its partners should adopt recommended practices with American Values and the Rule of Law for creating and maintaining trustworthy AI systems that are auditable, traceable, interpretable, and reliable.