The Efficiency–Equity Tradeoff in AI-Enabled Humanitarian Information Systems: Toward an Intersectional Governance Audit Framework

Abstract

Artificial intelligence (AI) systems are increasingly embedded in humanitarian and disaster response, including flood forecasting, satellite-based damage assessment, displacement prediction, and biometric aid registration. While governance debates emphasize fairness and accountability, most systems are optimized for predictive accuracy, speed, and operational efficiency. We map current humanitarian AI deployments and identify a recurring structural pattern: efficiency-driven design can sideline gendered and intersectional vulnerabilities. Through case analyses of flood forecasting, rapid damage assessment, and biometric registration, we examine how optimization goals, units of analysis, and performance metrics determine what counts as relevant harm, and how the abstraction required for AI modeling compresses complex social contexts in ways that limit both contestability and oversight. We introduce a draft Intersectional Governance Audit Framework organized around optimization logic, administrative embedding, and governance alignment to guide evaluation of humanitarian AI design choices upstream of deployment.