Doom Debates: Why Stephen Casper Says AI Alignment Could Backfire

AI safety requires a shift from technical alignment toward robust governance and institutional accountability. While Stephen Casper, a Harvard Kennedy School professor, maintains a 5-10% probability of existential catastrophe, he rejects dramatic "superintelligence takeover" scenarios in favor of a messier, gradual disempowerment driven by geopolitical competition and power concentration. Current technical safety efforts often function as "safety washing," enabling companies to deploy powerful systems while bypassing critical questions of liability. Instead of pursuing automated alignment or scalable oversight, the focus must move toward "ecosystemic hygiene" to address tangible harms like sycophancy, deepfake pornography, and algorithmic bias. Building transparency standards and institutional resilience is essential to prevent AI from quietly eroding democratic power structures and intensifying global conflicts.