"The Overstated Cost of AI Fairness in Criminal Justice", by Professor Ignacio Cofone, University of Oxford, Faculty of Law, and Associate Professor Warut Khern-am-nuai, McGill University - Desautels Faculty of Management is a recently published paper in Indiana Law Journal (forthcoming 2025).
The paper challenges a central assumption in the AI and criminal justice debate: that fairness constraints necessarily trade off with public safety.
Professor Cofone says: "This paper pushes back against the common notion in criminal justice that applying AI fairness in recidivism prediction imposes social costs in terms of releasing dangerous defendants. With an empirical analysis of the COMPAS database and prediction outcomes, it first shows that AI models like COMPAS don’t just perpetuate existing racial biases; they worsen them by about 20%. It then shows that fairness constraints can correct distortions caused by flawed outcome variables, so they don't necessarily degrade, and might improve, recidivism prediction. This has implications for law and policy, since it means one can improve AI fairness without the costs that they are often assumed to have."