Augment or Automate? An Early Field Experiment with Generative AI in Hiring
Joint with Kobbina Awuah, and David Yanagizawa-Drott
Revise and resubmit at the Journal of the European Economic Association
Abstract ▾ · SSRN
Can generative AI improve hiring? We experimentally embed GPT-4 into a teacher-recruitment screening process in Ghana, comparing three pipelines: human-only evaluation, human with AI assistance, and fully automated screening. Automation increases offer and hiring rates by 84% and 73%, respectively, over the human-only baseline. Human evaluators apply idiosyncratic standards, while the AI grades more consistently and more generously, advancing more applicants to the in-person assessment. In contrast, AI assistance yields no detectable improvement, as evaluators ignore most of the AI’s recommendations. Taken together, our results suggest that, at least in this context, automation outperforms human-AI collaboration.
Career vs. Social Impact: Gendered Responses in Teacher Recruitment in Ghana
Joint with Kobbina Awuah
Abstract ▾
Organizations recruiting for jobs with a social mission must decide what to emphasize when advertising them. In a field experiment with a Ghanaian NGO, we randomly varied whether emails recruiting university graduates into a teaching fellowship emphasized its career benefits or its social impact. The career message increased male application rates by 71% relative to the social-impact message and left female rates unchanged; we reject that men and women responded equally. Suggestive evidence indicates that candidates recruited through the career message were more likely to receive offers and be hired, and that men, but not women, were less prosocial.