ControlAI's Amadori: Three Filters prove most ASI survival plans fail
Common theories of change ignore three catastrophic pathways from superintelligent AI.
Alex Amadori, a researcher at ControlAI, published 'The Three Filters: Why Almost Every Plan to Survive ASI Fails Miserably' on LessWrong. He systematically examines common theories of change—technical AI safety research, racing to build ASI first, and creating a benevolent superintelligence—and argues they all collapse under the pressure of corner-cutting. The core problem: any entity that builds ASI gains an insurmountable competitive advantage, creating extreme incentives to cut safety corners. This pressure alone makes most plans unrealistic, even if alignment proves technically easy.
Amadori identifies three pathways that can destroy all value: human extinction, all-out nuclear war (the least-bad scenario he considers), and suffering risks (s-risks) that are even worse than extinction. He argues that no single technical fix or race strategy addresses all three simultaneously. The only approach that works, he claims, is global coordination to halt or slow ASI progress, paired with mass public understanding of the stakes. He endorses ControlAI's mission but emphasizes the post is about diagnosis, not solution advocacy.
- Amadori identifies three failure pathways: extinction, nuclear war, and suffering risks (s-risks), each requiring distinct countermeasures.
- Common survival plans (technical alignment, racing, benevolent ASI) fail because they ignore corner-cutting pressure from ASI's competitive advantage.
- The only proposed solution: global coordination to slow ASI + mass societal awareness of risks, as advocated by ControlAI.
Why It Matters
Professionals in AI must understand that technical safety alone can't prevent catastrophe—global governance and public awareness are essential.