A multidisciplinary team from the University of Wollongong uses semistructured interviews with 72 stakeholders—clinicians, regulators, developers, and consumer representatives—to assess perceptions of algorithmic bias in healthcare AI. They identify divergent positions on bias existence, responsibility distribution, and handling sociocultural data, and advocate for combined sociolegal and technical interventions, including diverse datasets, open disclosure, and regulatory frameworks, supported by interdisciplinary collaboration to promote equitable AI deployment in clinical settings.

Key points

• Conducted semistructured interviews with 72 multidisciplinary experts to map perspectives on algorithmic bias in healthcare AI.
• Identified three opposing views on bias existence—critical, apologist, denialist—and conflicting stances on mitigation responsibility and sociocultural data inclusion.
• Proposed integrated sociolegal measures (patient engagement, equity sampling, regulatory oversight) and data science strategies (governance, synthetic data, bias assessments) for fair AI deployment.

Why it matters: Addressing algorithmic bias in healthcare AI is essential to prevent perpetuating systemic inequities and ensure equitable patient outcomes across diverse populations.