For us, but not without us: an Afrocentric framing for models regulating clinical AI

This project examines how racially biased, AI-informed clinical decision-making became normalized in U.S. medicine and advances an Afrocentric, patient-led framework for harm reduction. Grounded in qualitative policy analysis, it traces historical pathways by which racial logics entered model design, guideline translation, payer rules, and deployments. A history of artificial intelligence in medicine (AIM) is paired with analysis of the policy apparatus that turns contested evidence into standards of care. Analytical illustrations show how proxy choices, label selection, and data curation reproduce inequity: population-health cost algorithms that under-refer Black patients; dermatology and imaging models that underperform on dark skin; VBAC and ASCVD calculators that embed race; and kidney indices that devalue Black donors. A capstone case of race-adjusted eGFR demonstrates “Agency Reduction,” whereby technical choices and policy adoption jointly delay recognition, referral, and transplantation. Methodologically, the study centers Afrocentricity, treating patient knowledge as theory-bearing, and interrogates who defines risk, which bodies are coded as “normal,” and how racial proxies are smuggled into ostensibly neutral variables. Findings map the governance levers through which bias becomes durable (clinical guidelines, reimbursement, accreditation) and where change is actionable. The project proposes a two-track intervention: Track A (design reform) replaces race-based labels with equity-aligned alternatives and transparent validation; Track B (governance reform) embeds community authority through patient advisory power, equity impact statements, and oversight tied to payment. Together these strategies move AIM from deficit-based medicine toward accountable systems to operationalize racial justice in clinical AI.