The authors analysed trial registries to measure the growth and focus of artificial intelligence in economic research. They extracted data from the American Economic Association and the Registry for International Development Impact Evaluations from 2019 into a partial 2026. By filtering for terms like machine learning and specific language models, they compared the volume, location, and subject matter of AI trials against general research. They also measured the average planned duration of these trials and tracked how often researchers partnered with government institutions.
Artificial intelligence interventions are on track to make up one in five registered trials, up from one in thirty in under four years. The authors found that the average planned AI trial runs for 11.3 months, which spans more than three generations of model updates. Furthermore, AI trials in the AEA registry are government-related 10.7 per cent of the time, compared to 14.1 per cent for trials overall. The authors argue that lacking government involvement limits adoption, while lengthy trials risk producing outdated findings. They point instead to an eight-week trial of an AI guided learning intervention in Sierra Leone, which tested over 1,700 students and recorded a 0.26 standard deviation gain.
Building an evidence base for artificial intelligence without involving the public sector leaves ministries with results they cannot easily use. Because machine learning models evolve rapidly, lengthy evaluations often test systems that no longer exist by the time the results are published. The opportunity lies in designing rapid evaluations that measure real capabilities while the tested models remain available to the public.
Today’s links: Assorted links for 7 September 2026.