The invention-to-adoption gap keeps shrinking

A new NBER working paper introduces GABRIEL, an open-source library that wraps GPT for extracting, rating, ranking, and categorizing qualitative information from text data. Of note for innovation policy: they apply the framework to study technology diffusion, filtering 18 million Wikipedia articles down to ~25,000 industrial-age technologies and extracting characteristics for each. They find that lags between invention and adoption have shortened from ~50 years in the 1800s to ~5 years today (with interesting heterogeneity, e.g. military and transportation technology diffuse faster than average, while agricultural, medical, and energy technology diffuse slower). The pace of technology adoption is an important input for innovation policy decision-making; CBO recently conducted a preliminary analysis of the budget impacts of R&D investment, within which they modeled a 15-year ramp from outlays to peak productivity effects. If the invention-to-adoption gap is truly compressing as fast as the authors find, R&D may pay off in shorter windows than typically assumed.