AI Alignment at Which Abstraction Level?

LessWrong publishes a new essay by Adam Chlipala that asks at what level we should frame AI alignment. He argues that the usual focus on human values assumes a privileged abstraction without justification, and draws on his experience scaling formal verification to warn that putting humans at the top of the specification may hide hidden bugs in the trusted base. The piece also critiques proposals like coherent extrapolated volition, pointing out potential risks while an AI learns preferences, and suggests looking to higher‑level, less human‑centric specifications. Chlipala notes this is the first of a three‑part series exploring formal‑methods approaches to the alignment problem.

Listen to this story

Hear this and more stories in a personalized audio briefing.

Open The Chonkerton