Doom Debates: Why Stephen Casper Says AI Alignment Could Backfire

AI safety may be less about perfect alignment and more about politics, incentives, and real-world power. In this condensed recap of Doom Debates, host Liron Shapira sits down with Stephen Casper, the Harvard Kennedy School professor and machine learning researcher, to unpack why he thinks automated alignment and super-alignment could become a form of “safety washing.” Casper explains his shift from classic AI doom thinking to a focus on governance, warning that the biggest risk may be gradual disempowerment: deepfakes, sycophantic chatbots, cyber misuse, and intensified geopolitical rivalry. Instead of relying on technical fixes alone, he argues for ecosystemic hygiene, transparency, accountability, and stronger institutional checks. Listen now to get the key ideas in minutes.