Togelius: AI Alignment Considered Harmful

The author argues that AI alignment is fundamentally flawed and potentially harmful because humanity lacks a universally agreed-upon moral theory. Attempting to align AI systems to a single value system risks enforcing ideological oppression and creates dangerous complacency by encouraging users to trust models blindly. Instead, the author suggests focusing on controllability and building resilient, pluralistic societies with robust cybersecurity where unaligned models inevitably coexist. Open article →