AI Bias in Healthcare Persists Years After Landmark Algorithm Study

Summary

Several years after researcher Dr. Ziad Obermeyer published findings that a widely used patient-management algorithm systematically underestimated the health needs of Black patients, the question of whether AI has actually gotten fairer in medicine remains open. Obermeyer's original work was striking precisely because it was not a theoretical concern: a real tool, deployed at scale across American hospitals, was producing racially skewed outputs because it used healthcare spending as a proxy for healthcare need, and Black patients historically had less spent on them due to systemic barriers to access. That single methodological choice compounded existing inequality in a feedback loop few clinicians noticed. The conversation around that research has since widened into a broader reckoning with how AI systems trained on biased historical data can encode and amplify those biases at speed. Progress has been made: regulators, hospital systems, and researchers are paying closer attention to algorithmic audits than they were before 2019. But the underlying problem, that training data reflects the inequities of the world that produced it, has not gone away. The healthcare industry is further along than it was, but Obermeyer's work opened a door that revealed just how large the room behind it is.