Can generative AI improve hiring? We experimentally embed generative AI (GPT-4) into a teacher-recruitment screening process, comparing three pipelines: (i) human-only, (ii) human with AI assistance, and (iii) fully automated screening. Automation increases downstream hiring success by 11 percentage points (a 73% improvement) over the human-only baseline. In contrast, AI assistance neither improves outcomes nor productivity, as users systematically disregard its recommendations, perceiving it as incapable of distinguishing between signals of teacher quality. Our results provide evidence in favor of fully automated deployment of generative AI for an economically important task, highlighting potential limitations of hybrid approaches involving humans-in-the-loop.
Awuah et al. (Wed,) studied this question.