Robin Hanson Compares AI Value Drift With Human Value Change

Summary

In this opinion essay, Robin Hanson examines why concerns about long-term value drift are focused so heavily on artificial intelligence. The central doomer argument is that AI systems could eventually become more capable and powerful than humans, after which the distribution of values among powerful AIs might shift toward radically anti-human goals, potentially including killing humanity. Hanson notes that today’s large language models appear more pro-social, law-abiding, and respectful of humans than many people, but argues that their current behavior does not prove that future AI values will remain stable. He explains the concern through the idea that values are distributed across many parts of a mind rather than stored in an isolated, unchanging “lockbox”; changes to those systems could therefore produce value drift. Hanson accepts that humans can monitor and adjust AI while they remain in control, but says the risk debate assumes that increasingly capable, fast, and opaque systems may eventually become difficult to manage. He then compares this with human descendants. Hanson argues that admired human values are shaped largely by culture rather than DNA, and that faster social and economic change has already accelerated human value change. He describes his own concern about this human drift as a form of “human doomerism.” The essay also challenges the belief that recent moral change mainly reflects progress toward moral truth, arguing that no specific moral arguments clearly explain broad recent shifts. Finally, Hanson rejects the idea that carbon-based biology is uniquely suited to consciousness or truth-tracking, suggesting that these properties may not depend strongly on carbon versus silicon.