preprint · 2026-09
What if automating AI R&D triggers an intelligence explosion?
Why it matters here
This working paper examines how AI doing more research could accelerate the development of better AI. Better systems could carry out more effective research, which could produce still better systems and speed up later rounds. Whether that feedback becomes self-sustaining depends on how much useful research AI can perform and whether experiments, data, human decisions or lengthy training runs hold it back. The paper brings together existing studies and economic models; it does not report a new benchmark result.
Observed R&D automationRecursive improvement evidence
Related benchmarks and indicators
Research outcome predictionTests whether a specialized AI system can predict which of two research ideas will perform better before the experiment.Automated weak-to-strong research studyTests whether AI research agents can find better ways to train a stronger model using guidance from a weaker model.PostTrainBenchTests how well AI agents improve small AI models, with ten hours and one H100 GPU per training run.Anthropic R&D Automation IndexModel-rated automation across a fixed basket of R&D work.RE-BenchResearch engineering environments with expert baselines.
What to keep in mind
- The acceleration scenarios are conditional model calculations, not observed progress or a predicted date for RSI. The supplementary model assumes full research automation and simplifies compute and data constraints.
- More researchers can produce diminishing returns or duplicate work. Some research tasks remain difficult to automate, and some experiments must run in sequence.
- The authors identify limited public data on actual research productivity, resource use and the strength of the feedback from better AI to faster AI development.
- The paper’s cited studies are supporting evidence, not additional independent replications. Author affiliations do not imply endorsement by their organizations.