AI researcher David Krueger argues that current industry proposals for pacing AI development are insufficient and advocates for an international moratorium on frontier AI.
Key facts
- •David Krueger is an assistant professor in Robust, Reasoning and Responsible AI at the University of Montreal.
- •Anthropic CEO Dario Amodei proposed a one-to-two-year slowdown for frontier AI, a plan endorsed by Sam Altman and Elon Musk.
- •Krueger previously helped found the UK AI Security Institute but resigned due to concerns over the industry's approach to safety.
- •The proposed moratorium would involve dismantling the supply chain for EUV and DUV lithography used in advanced chip production.
- •Krueger suggests that an international agreement to decommission data centers would be easier to verify than current proposals for global coordination.
David Krueger, an assistant professor of AI at the University of Montreal, has called for an immediate, indefinite, international moratorium on frontier AI development. Krueger argues that current proposals from industry leaders, such as Anthropic CEO Dario Amodei’s plan to "pace the frontier," fail to address fundamental risks regarding control and safety. He contends that the industry’s current approach of self-regulation is inadequate to prevent potential existential threats posed by increasingly powerful AI systems.
Proposed Strategy for Verification
Krueger suggests that an international moratorium could be verified by decommissioning the advanced computer chips and data centers required to build frontier AI models. He proposes that the US and China could take turns dismantling this infrastructure to ensure no party maintains the capability to develop more powerful systems. To ensure durability, he suggests dismantling the supply chain for advanced AI chips, specifically targeting extreme ultraviolet (EUV) and deep ultraviolet (DUV) lithography technology, which he notes would be difficult to operate covertly.
Critique of Industry Self-Regulation
The author criticizes current safety evaluations, noting that the industry operates under a "safe-until-proven-dangerous" framework. He highlights that Amodei’s proposal relies on external evaluators like METR, which he characterizes as a form of self-regulation rather than rigorous oversight. Krueger argues that these standards fall short of those found in safety-critical industries like aviation and warns that the industry's focus on "recursive self-improvement" risks rendering humans obsolete.
Advertisement
This article was independently rewritten by ManyPress editorial AI from reporting originally published by Guardian AI.
