AI Interpretability, Explainability and the Mechanistic Reality That Compliance Frameworks Miss

This post was either an anonymous submission of an interesting paper or was written by a student; full credit remains with the author (linked).

A critique of how governance frameworks like the NIST AI Risk Management Framework distinguish "explainability" from "interpretability," arguing that this framing — while useful for compliance — glosses over the deeper, harder problem: many deep learning systems are inherently difficult to mechanistically understand, no matter how well their outputs are documented.