How May U.S. Courts Scrutinize Their Recidivism Risk Assessment Tools? Contextualizing AI Fairness Criteria on a Judicial Scrutiny-based Framework

Technical fairness concepts have constitutional foundations but face implementation gaps: The paper finds that major AI fairness criteria (procedural, group, and individual fairness) can be linked to U.S. constitutional protections like Due Process and Equal Protection, but current policies and regulations often fail to operationalize all of them—notably leaving out individual fairness considerations entirely.

Recidivism risk assessment tools lack consistent legal standards: Existing policies adopt procedural and group fairness approaches inconsistently, with some policies contradicting each other on whether certain demographic features should be subject to fairness protections, creating uncertainty about how courts should evaluate these high-stakes criminal justice tools.

Current judicial scrutiny frameworks may be insufficient: While U.S. courts use established "scrutiny" categories to evaluate potentially discriminatory government practices, the paper suggests these legal frameworks may not fully capture how courts should specifically scrutinize the use of demographic data in AI systems like recidivism risk assessments.

AI and legal communities operate in silos: The AI/HCI and legal communities have developed separate understandings of fairness that don't align well, hindering the creation of sustainable policies and designs for high-stakes applications—suggesting better collaboration between technical experts and legal scholars is needed.

International approaches vary significantly: The framework's application across the EU, China, and India reveals different jurisdictional approaches to AI fairness, indicating that responsible AI governance requires context-specific solutions rather than one-size-fits-all standards.