[egocodedinsol]

Nguyen claims that the renaissance in neural networks will provide us with concepts to understand the human brain in an analogous way to how the steam engine allowed us to conceive of entropy.

I'm not convinced it goes in that direction yet, though. Neural networks are loosely biologically inspired to begin with, and the idea of the primate visual system as a deep feedforward network predates the recent machine learning advances by many years.

If that's true, it undermines his entire thesis. What's missing: how a concept in machine learning allowed us to conceptualize something new in neuroscience, rather than just describe a process we have a vague intuition for (still obviously useful).

FWIW, and I'm a little biased here, I would argue that it's (high-level,vague) concepts in neuroscience that have been driving machine learning. There are ways we behave and learn that we've been trying to emulate in machines. Someday it will swing back the other way, but not yet.