A second, slightly broader thought: The prevailing atmosphere of gloom about mathematics &c is based on a perception that neural networks produce discovery without understanding. Turn the crank, a proof comes out. But we have no idea why.
If interpretability actually worked, this might feel