Brief
Paper models when AI feedback could lead to self-sustaining acceleration
A new paper coauthored by Parker, Tom, and seven other economists presents simple models of how AI may accelerate AI R&D. It focuses on the strength of feedback effects and whether they could cause self-sustaining acceleration.
The paper does not claim that recursive self-improvement is happening. Instead, it quantifies feedback effects and identifies the most uncertain link: how increased model capabilities would speed up algorithmic progress. The authors say they cannot rule out substantial acceleration, but list bottlenecks—data, compute, experiments—that could cause it to fizzle out. The work is a modeling exercise, not an empirical finding.
Our reading
Our reading is that this is a conceptual modeling paper, not evidence that acceleration is occurring or will occur.
What to do or watch
Watch for empirical work on the paper's most uncertain link: how an increase in model capabilities would increase the rate of algorithmic progress. The unresolved question is whether feedback effects are strong enough for self-sustaining acceleration or fizzle out due to bottlenecks in data, compute, experiments, algorithmic-specific capabilities, and R&D-specific capabilities.
Source details and supporting facts
Each line is stated by the page named above it.
Stated by metr.org
- The paper was coauthored by Parker and Tom with 7 other economists.
- The paper walks through a series of simple models of how AI may accelerate AI R&D.
- METR's priority is to assess risk from frontier AI development.
- Capabilities have been growing rapidly over the past 5 years.
- The most uncertain relationship is how an increase in model capabilities would increase the rate of algorithmic progress.
Sources
- METRText stored 15 September 2026
How this story was checked. Written from the 1 page listed above, stored 15 September 2026; claims checked against that stored text on 15 September 2026.
What that means
- 5 of 5 reported statements were confirmed against the page that carries them; the rest were removed rather than published.
- Figures in the text were required to appear in the stored source text: yes. Identifiers: yes.
- The check reads stored text only: no claim rests on a fresh look that did not happen.
- Where the reporting was silent, the text says so instead of filling the gap.