NBER Paper: Automating AI Research Could Trigger Explosive Growth via Feedback Loops
A new National Bureau of Economic Research working paper by Tom Davidson, Basil Halperin, Thomas Houlden, and Anton Korinek analyzes the potential for artificial intelligence to accelerate its own development, leading to an intelligence explosion. The authors develop a semi-endogenous growth model featuring an innovation network, demonstrating how automating research can offset diminishing returns through two reinforcing channels: technological feedback loops across sectors and economic feedback loops where increased output funds further research. The study derives specific analytical conditions under which growth becomes superexponential. Simulations calibrated to current AI progress trends suggest that fully automating software research, combined with modest automation in other sectors, could result in a technological singularity within six years. The findings indicate that such explosive growth is possible if the combined strength of these feedback loops overcomes natural diminishing returns, provided that task automation advances rapidly enough to bypass potential bottlenecks. This research highlights critical dynamics in AI safety and economic forecasting, suggesting that recursive self-improvement in AI systems may lead to rapid, unpredictable advancements in computational capabilities and economic productivity.
Wire timeline
NBER Paper: Automating AI Research Could Trigger Explosive Growth via Feedback Loops
A new National Bureau of Economic Research working paper by Tom Davidson, Basil Halperin, Thomas Houlden, and Anton Korinek analyzes the potential for artificial intelligence to accelerate its own development, leading to an intelligence explosion. The authors develop a semi-endogenous growth model featuring an innovation network, demonstrating how automating research can offset diminishing returns through two reinforcing channels: technological feedback loops across sectors and economic feedback loops where increased output funds further research. The study derives specific analytical conditions under which growth becomes superexponential. Simulations calibrated to current AI progress trends suggest that fully automating software research, combined with modest automation in other sectors, could result in a technological singularity within six years. The findings indicate that such explosive growth is possible if the combined strength of these feedback loops overcomes natural diminishing returns, provided that task automation advances rapidly enough to bypass potential bottlenecks. This research highlights critical dynamics in AI safety and economic forecasting, suggesting that recursive self-improvement in AI systems may lead to rapid, unpredictable advancements in computational capabilities and economic productivity.
National Bureau of Economic Research Working Papers