Evaluating Gender Bias in AI Applications Using Household Survey Data

In this study we introduce an approach for evaluating the role of gender in wealth estimates generated using artificial intelligence (AI), geospatial data, and USAID’s Demographic and Health Surveys (DHS) data. In partnership with Ghana Center for Democratic Development, we use the 2014 round of the Ghana DHS to compare the performance of models trained on gender specific household data. Findings indicate that even after controlling for sample size, models trained and validated on male household data outperform models trained and validated on female household data. Alongside the Technical Report detailing the methodology and findings, the associated code and data is available to facilitate future use. The Executive Summary highlights key lessons from this study including the importance of local partners when developing AI applications. This report and associated work was produced as part of USAID's Equitable AI Challenge and aims to elevate equitability discussions surrounding the growing use of AI in development.