According to LessWrong, a researcher is proposing a methodological framework for fundamentally questioning the Agent Foundations research agenda—a decades-long effort to ground AI alignment in formal mathematical foundations. Rather than continue incremental work toward potentially unreachable goals, the author argues that alignment researchers should systematically enumerate all possible formulations of core concepts and demonstrate whether the core objectives are even achievable in principle. The approach borrows from Kant's historical refutation of theistic proofs. Applied to Agent Foundations, this might either refute the agenda or reveal unexplored research directions worth pursuing. The author notes the post is incomplete and explicitly invites substantive criticism on the core framing.

Listen to this story

Hear this and more stories in a personalized audio briefing.

Open The Chonkerton