A Statistical Case Against Empirical Human-AI Alignment