Submission ID 131634
| Issue/Objective | Artificial intelligence is being embraced across global health systems as a transformative solution to workforce shortages, diagnostic gaps, and resource constraints - particularly in underserved settings. Yet the foundational data powering these systems tells a different story. Nearly three quarters of clinical AI training datasets originate from the Americas and Europe - regions representing just 22% of the global population. AI systems perform better in English than in any other language, systematically restricting whose health knowledge is recognized, whose symptoms are accurately interpreted, and whose communities benefit. In low-income countries, which carry the greatest burden of preventable disease, AI research and deployment remain critically underdeveloped. Global health is at risk of adopting a technology that automates and scales the very inequities it claims to dismantle. The problem is not artificial intelligence itself. It is the absence of any standardized equity accountability framework governing how AI tools are evaluated before they are deployed in marginalized settings. |
|---|---|
| Methodology/Approach | A narrative synthesis of peer-reviewed literature, WHO AI governance frameworks, and global health equity measurement tools published between 2020 and 2025 was conducted to map existing gaps in pre-deployment AI equity assessment. Frameworks, including the Health Equity Across the AI Lifecycle (HEAAL) model, the SHIFT principles, and Canada-specific rapid reviews on AI and priority populations, were analyzed for transferability to low-resource and globally diverse health contexts. Findings informed the conceptual development of the AI Equity Audit Checklist (AEAC). |
| Results | The AEAC proposes five pre-deployment audit domains systematically absent from current AI adoption pathways in global health: training data representativeness across geography, language, and ethnicity; algorithmic bias testing in low-resource clinical environments; community and frontline worker involvement in AI design; transparency and explainability standards accessible to non-specialist health workers; and equity impact assessment as a condition of funding and scale-up. Unlike existing technical bias mitigation tools designed for high-income institutional contexts, the AEAC is designed to be actionable by health ministries, NGOs, and global health funders operating in fragmented, under-resourced settings. |
| Discussion/Conclusion | In a fragmented and uncertain world, the appeal of technological solutions to deep structural problems is understandable, but dangerous. AI deployed without equity accountability does not neutralize systemic bias. It encodes it, scales it, and makes it harder to see. The AEAC offers a transferable, low-barrier tool for global health actors to ask the right questions before adoption rather than correct harm after deployment. Funders, multilateral bodies, and national governments must treat equity auditing as a prerequisite for AI investment in global health - not an afterthought. The communities most harmed by algorithmic inequity are the same communities global health has always promised to reach. We cannot automate our way out of a justice problem we have not yet solved. |
| Presenters and Affiliations | Viksit Bali UHN viksit Bali UHN |
