Decoding Algorithmic Inequity: Reversing the Generative AI Racial Divide in Healthcare.

Artificial intelligence is rapidly transforming healthcare delivery. Over the past decade, studies have increasingly demonstrated that healthcare AI systems inherit, operationalize, and amplify longstanding bias, health disparities and structural inequities embedded within healthcare systems. In this perspectives paper, we introduce the concept of a Data Disparity Pipeline to describe how inequities originating upstream are propagated through clinical interactions, healthcare datasets and embedded within clinical AI, ultimately impacting healthcare access and outcomes particularly for marginalized and underserved populations. We examine the major sources of bias and prominent evidence across healthcare AI systems, reflecting the impacts to clinical decision support and the limitations of current mitigation strategies. We further explore how biased data, proxy variables, underrepresentation, and recursive retraining processes contribute to disparities across risk prediction systems, generative AI models, and adaptive clinical platforms, while highlighting emerging international approaches prioritizing representative data infrastructure, participatory oversight, transparency, and continuous equity auditing. Finally, we propose a multimodal framework with recommendations for targeted consortium-based efforts involving key stakeholders to drive change across data governance, transparency, clinician behavior, ethics oversight groups, and affected communities to ensure that healthcare AI advances global health equity and reduces existing disparities.