22 AI Experts Warn Automated R&D May Compress Progress to Five Weeks
A recent paper co-authored by 22 leading AI researchers, including Turing and Nobel laureates Geoffrey Hinton and Yoshua Bengio, warns that automating artificial intelligence research could trigger a rapid intelligence explosion, potentially compressing a year of technological progress into as few as five weeks. Published by the Cambridge Programme on AI Science and Policy, the report examines the accelerating shift toward AI-driven R&D, noting that cutting-edge models are increasingly capable of generating code, designing experiments, and optimizing research pipelines with minimal human oversight. Internal benchmarks from Anthropic and OpenAI confirm this trajectory, revealing that AI systems now manage substantial portions of research workflows, with developer productivity scaling rapidly through parallelized software agents. The authors describe a self-reinforcing feedback loop: as AI systems assume more R&D responsibilities, they simultaneously accelerate the development of subsequent models. Drawing on historical data, the paper estimates a returns to research effort metric between 1.2 and 1.9, suggesting that increased computational and algorithmic investment currently yields disproportionate technological gains. If automation scales without hitting immediate physical or computational bottlenecks, the report projects that AI iteration speeds could increase tenfold within eighteen months, fundamentally altering the timeline for scientific and engineering breakthroughs. Beyond the technical trajectory, the signatories highlight three critical systemic risks. First, the pace of AI advancement may outstrip societal and regulatory adaptation, leaving defenses against cyber threats, biological misuse, or labor market disruption vulnerable. Second, increased autonomy could degrade human oversight, allowing biases or misaligned objectives to compound across iterative model generations. Third, early adopters or dominant nations could exploit compounding advantages to consolidate unprecedented technological and geopolitical leverage. Rather than calling for an outright halt, the researchers advocate for systematic monitoring and transparency. They urge the establishment of independent auditing frameworks, standardized reporting on AI automation rates within research labs, and government contingency plans for rapid technological shifts. This approach shifts the policy debate from voluntary corporate slowdowns to verifiable metrics, ensuring that oversight mechanisms can keep pace with the accelerating trajectory of artificial intelligence.
