Algorithmic Fairness: Debiasing Artificial Intelligence Admissions Essay Scores Reduces Black-White Group Differences By Valuing Diverse Experiences

Natural language processing is increasingly used to score open-ended text but can often exacerbate adverse impact. I examine the effects of debiasing machine learning admissions essay scores via multi-objective optimization. While reliability and cognitive saturation decrease slightly from optimizing diversity, the main effect is changing what experiences receive high scores.