Measuring AI R&D Automation
A measurement proposal for AI R&D automation, tracking how much AI changes research labor, progress rates, oversight, and incident patterns.
Which telemetry indicators provide the earliest reliable warning that AI R&D automation is outpacing oversight capacity?
This is governance that technical teams should actually want: measure the acceleration curve. If AI systems begin automating the work of AI researchers, capability progress, safety progress, compute allocation, and oversight capacity can diverge quickly.
Why it belongs here
ASI strategy needs telemetry. The paper proposes tracking dimensions such as AI R&D spending share, researcher time allocation, and AI subversion incidents. Those metrics are imperfect, but they make a rapidly changing research process visible enough to manage.
Pro-ASI read
The goal is not to slow research by default. The goal is to know when automated R&D is becoming a major input to frontier progress, so institutions can increase evals, security, reproducibility, and deployment confidence at the same tempo.