Are Ideas Getting Harder to Find?
· Published in the April 2020 issue of the American Economic Review; 15 April is used only for sorting.
Summary
Across semiconductors, agriculture, health and firms, the authors find that sustaining progress has required increasing research effort. Their historical accounting estimates declining research productivity: more researchers are needed to produce a given rate of improvement. The paper supplies an empirical baseline for models of idea production.
Why it matters
AI could make research cheaper or more productive, but faster discovery must be assessed against the effort needed to keep advancing the frontier. The historical evidence helps frame what AI would need to change for sustained growth or recursive self-improvement.
What to keep in mind
Research productivity is an accounting residual sensitive to output definitions, input measurement, spillovers and case selection. The findings are neither a universal law of discovery nor a forecast about AI; research automation could change these historical relationships.