Import AI 459: AI oversight is difficult; scaling laws for protein folding models; and pricing the extinction risk of AI systems

· Source: Import AI · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Data Science & Analytics, Emerging Technologies & Innovation · Depth: Intermediate, long

Summary

The US AI economy is experiencing unprecedented, yet largely invisible, growth, projected to reach approximately \$250 billion in 2025 with a 2,600 percent annual quality-adjusted real growth rate, according to economists from the University of Virginia, Anthropic, and the Bank of Canada. This rapid expansion, driven by AI inference, is obscured by falling per-unit prices and poses a unique challenge as AI may substitute human labor. Concurrently, researchers from the UK AI Security Institute highlight the inherent difficulty of automated AI alignment, citing issues like "alien mistakes" and non-human-evaluable arguments. In other developments, Stanford University and collaborators released GPIC, a 100M permissively licensed image dataset, while Biohub introduced ESMFold2, a protein prediction model outperforming AlphaFold 3 in some benchmarks for cancer research. An Australian Assistant Minister also urged economists to better price AI extinction risks, emphasizing the need for policy frameworks that consider survivability.

Key takeaway

For policymakers and economists assessing future economic stability, you must integrate quality-adjusted AI output measurements and develop "AI satellite accounts" to accurately reflect the sector's 2,600 percent annual growth and potential labor market shifts. Simultaneously, AI safety researchers should prioritize robust measurement and generalization techniques, including red teaming, for automated alignment systems, recognizing their inherent complexities. This proactive approach is crucial for mitigating unpriced extinction risks and ensuring societal preparedness.

Key insights

AI's rapid, hidden economic growth and complex safety challenges demand new measurement and governance frameworks.

Principles

Method

For AI safety oversight, interventions include recreating research projects, testing agent prediction performance on correlated events, and red teaming automated alignment programs.

In practice

Topics

Best for: Research Scientist, AI Scientist, Policy Maker, AI Ethicist

Related on AIssential

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Import AI.