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

Area of Science:

• Health Informatics
• Artificial Intelligence in Medicine
• Health Equity Research

Background:

• Healthcare AI systems increasingly demonstrate inherited, operationalized, and amplified biases and structural inequities.
• These biases disproportionately affect marginalized and underserved populations, impacting healthcare access and outcomes.

Purpose of the Study:

• Introduce the "Data Disparity Pipeline" concept to explain how upstream inequities propagate through clinical AI.
• Examine sources of bias in healthcare AI and evaluate current mitigation strategies.
• Propose a multimodal framework to advance global health equity through AI.

Main Methods:

• Review of existing studies on bias in healthcare AI systems.
• Analysis of how biased data, proxy variables, and retraining processes contribute to disparities.
• Exploration of emerging international approaches for equitable AI.

Main Results:

• Healthcare AI systems amplify existing biases, operationalizing and exacerbating health disparities.
• Biased data, underrepresentation, and recursive retraining contribute to disparities in risk prediction, generative AI, and adaptive platforms.
• Current mitigation strategies for AI bias are limited.

Conclusions:

• A "Data Disparity Pipeline" concept illustrates how inequities are embedded and amplified in healthcare AI.
• Emerging international approaches prioritize representative data, oversight, transparency, and equity auditing.
• A multimodal framework with stakeholder collaboration is proposed to ensure AI advances health equity and reduces disparities.