Algorithmic Criminal Justice

In May 2016, two reporters at ProPublica named Julia Angwin and Jeff Larson published an investigation of a risk assessment tool called COMPAS, which was being used in Broward County, Florida to help judges make bail and sentencing decisions. COMPAS was built by a company called Northpointe and had been adopted by courts across the United States; it produced a score from one to ten predicting a defendant’s likelihood of reoffending. The ProPublica analysis found that Black defendants were nearly twice as likely as white defendants to be falsely flagged as high risk, i.e., to be scored as likely to reoffend when they did not, while white defendants were more likely to be falsely flagged as low risk and then go on to commit new offenses. COMPAS does not use race as an input. Its inputs include things like criminal history, age, employment status, and residential stability. These variables are legally permissible and individually plausible as predictors of behavior. They are also deeply…