U of Texas will stop using controversial algorithm to evaluate Ph.D. applicants

Abstraction: ML admissions tool encodes historical bias in PhD candidate selection

Key points:

• GRADE (GRaduate ADmissions Evaluator) used by UT Austin CS from 2013–2020, predicted admission probability on a 1–5 scale trained on past admissions committee decisions
• Institutions encoded as "elite," "good," or "other" based on faculty surveys; recommendation words like "best" or "award" boosted scores while "good" or "programming" predicted rejection
• Reduced full reviews by 71% and total review time by 74%; application volume grew from ~250 (2000) to over 1,200 by post-2012
• Critics: race and gender can be encoded in proxy features — women's colleges and HBCUs likely undervalued; gendered recommendation language reflected in scores
• UT officially cited maintenance difficulty rather than equity concerns as the reason for discontinuation
• "The racism of today is being immortalized in the algorithms of tomorrow" — activist Yasmeen Musthafa

Connections: University Of Texas Austin · Algorithmic Bias · Machine Learning · AI Fairness

Source: https://www.insidehighered.com/admissions/article/2020/12/14/u-texas-will-stop-using-controversial-algorithm-evaluate-phd