Artificial intelligence is transforming medical care across the United States, but the technology often perpetuates existing racial disparities in healthcare rather than eliminating them. Research reveals that AI systems trained on historically biased data reproduce those same inequities when analyzing patient conditions and recommending treatments.
A Duke University study found that an AI model designed to predict breast cancer risk performed significantly less accurately for Black patients compared to white patients. Researchers attributed this disparity to the model being predominantly trained on data from white patients, demonstrating how historical underrepresentation in medical research continues to affect care quality today.
The problem stems from AI learning patterns from existing medical data, which reflects decades of documented bias in healthcare. As Bradley Greger, an Arizona State University professor, explains, these systems are sophisticated but lack insight beyond their training. When fed biased historical data, they inevitably produce biased results.
Dr. Andrew Carroll, a physician in Chandler, believes AI could eventually reduce disparities by incorporating more detailed genetic and ethnic information. He notes that broad racial categories mask important differences—Black patients may have Nigerian, Puerto Rican, or Jamaican ancestry, each with distinct genetic risk factors.
Arizona has responded with House Bill 2175, prohibiting insurance companies from using AI alone to deny claims. Starting July 2026, licensed medical professionals must review all AI-generated decisions, ensuring human oversight remains central to patient care.
See: “AI reshapes healthcare but often adopts bias” (October 7, 2025)
