Let AI spot disadvantaged students with high potential
Some have proposed an admission lottery for candidates who meet qualifications. Roland G. Fryer Jr., a Harvard economics professor, suggests letting AI pick students based on academics and evidence of "grit." He believes "sophisticated analytics" can predict who's likely to excel, without human prejudices tipping the scales.
Fryer grew up poor and angry -- his father was violent, his mother was absent -- and took the SAT only to score the minimum needed for a football scholarship.
But for my college professors’ willingness to look beyond my past performance — but for affirmative action — I would not have benefited from twice-weekly 7 a.m. meetings with the economics professor who showed me how science could be used to help people. Or the statistics professor who marveled at my stories of my favorite uncle — a wino with sophisticated strategies of betting on Greyhound races — and helped me use formal models to explain his behavior. Or a spot at the American Economic Association’s summer school for minority students.
However, most affirmative action beneficiaries aren't high-potential students from poor families, writes Fryer. "Seventy-one percent of Harvard’s Black and Hispanic students come from wealthy backgrounds. A tiny fraction attended underperforming public high schools."
When humans are prone to bias, bots do better, Fryer writes. "A Cornell study of data on judges’ bail decisions found that computer predictions could reduce crime as much as 25 percent with no change in jailing rates."
A machine-learning model would be fed historical admissions data, including candidates’ family background and academic achievement, and noncognitive skills such as grit and resilience, along with outcomes of past admission decisions. It would use these data to predict new applicants’ performance — as defined by each institution, such as college grade-point average or income 10 years after graduation. The model could figure out which characteristics best predict performance for various subgroups — for example, how salient SAT scores are for public-school Black students raised in the South by single mothers vs. private-school White kids from the Northeast. If we use only unadjusted test scores, all that context is lost.
Someone would have to assign numbers for grit, resilience, leadership, etc. I don't think the bots can do it all.