human effort

Generative AI closed three quarters of an education-based performance gap during assisted work, but effort shaped what carried forward

Guillermo Cruces, Diego Fernández Meijide, Sebastian Galiani, Ramiro H. Gálvez, María Lombardi

arXiv working paper

In a preregistered randomized online experiment with 1,174 Argentine adults, GPT-4.1 assistance raised workplace-style problem-solving performance for both education groups and reduced the baseline gap from 0.548 to 0.139 standard deviations. Lower-education participants retained a modest gain after AI was removed, but stronger follow-up performance appeared when intensive assistance was paired with sustained human effort.