Algorithms Can Be Used In Sentencing

The Wisconsin Supreme Court unanimously upheld a man’s prison sentence which was based at least partially on an algorithm that determined he was at high risk of re-offending. The algorithm in question, COMPAS, covers criminal and parole history, age, employment status, social life, education level, community ties, drug use, and beliefs in a 137-question test meant to determine which offenders are at high risk of recidivism. The appellant, Eric Loomis, appealed his six-year sentence for attempting to flee the police, arguing that COMPAS violated his due-process rights because Northpointe, the company behind COMPAS, will not reveal details of its algorithm. But, “if used properly, observing the limitations and cautions set forth herein, a circuit court’s consideration of a COMPAS risk assessment at sentencing does not violate a defendant’s right to due process,” Justice Ann Walsh Bradley wrote for the Wisconsin high court. The limitations and cautions include orders for state officials to present COMPAS scores to judges with disclaimers about the proprietary nature of the test and its lack of validation testing in Wisconsin, and studies that question whether COMPAS disproportionately classifies black offenders as higher risk than whites. The risk score also cannot justify making a sentence longer or shorter, only to determine eligibility for diversion programs and imposing terms for probation. “Using a risk assessment tool to determine the length and severity of a sentence is a poor fit,” Bradley wrote.