The Overstated Cost of AI Fairness in Criminal Justice

2026-08-02

AI Summary

This study challenges the prevailing assumption that algorithmic fairness inherently trades off against predictive accuracy by demonstrating that training data often encode and amplify systemic biases. Through a causal inference framework, the authors conduct an empirical analysis of recidivism risk prediction models on the COMPAS dataset, revealing that existing models not only replicate but exacerbate racial disparities. The findings further indicate that, in criminal justice settings, incorporating appropriate fairness constraints can mitigate estimation bias arising from biased outcome variables—such as rearrest records—thereby enhancing model fairness without compromising, and even improving, predictive accuracy. These results contest the widely held belief that fairness necessarily comes at the cost of performance.

Abstract

A dominant critique of algorithmic fairness holds that increasing fairness reduces predictive accuracy, imposing a cost on society. We challenge that assumption by empirically analyzing the COMPAS dataset. We make two contributions. First, using causal inference methods, we show that racial bias is not only present in the COMPAS dataset but is also amplified by the models trained on it. Widely used models do more than replicate existing bias; they exacerbate it. This undercuts both the assumption that algorithmic decision-making offers a neutral improvement over human judgment and the weaker claim that it merely mirrors preexisting human bias. Second, we reframe the fairness-accuracy tradeoff. Applying fairness constraints does not necessarily cost predictive accuracy in criminal justice. Prediction systems operationalize concepts such as risk through implicit and often flawed normative choices about what to predict and how. The tradeoff claim assumes that the unconstrained model's prediction is an optimal baseline. Fairness constraints can instead correct distortions introduced by biased outcome variables: rearrest data, in this case, captures and magnifies systemic racial disparities. Under some interventions, therefore, fairness carries none of the cost presumed in policy debates. These dynamics extend beyond criminal justice to lending, hiring, and housing, where biased outcome variables reinforce inequality independently of proxy selection. We draw out what this implies for how law and policy should approach fairness adjustments in criminal law.

Problem

• algorithmic fairness
• predictive accuracy
• racial bias
• criminal justice
• fairness-accuracy tradeoff

Innovation

• algorithmic fairness
• causal inference
• bias amplification
• fairness-accuracy tradeoff
• biased outcome variables