Towards Justice-Oriented AI: Reflections on a Research-Community Partnership to Co-Develop Swahili AI Tools

Abstract

AI development is notoriously centralized, concentrating power in the hands of a few. Justice-oriented frameworks emerged to challenge this paradigm, but documented examples of projects that aim to distribute power and ownership across the full development lifecycle remain rare. In this paper, co-authors from a Kenyan activist organization that monitors femicide in the region join with a research team based in the United States to reflect on our co-development of Swahili AI tools for femicide monitoring. We adapted two existing tools previously developed by the research team to the Kenyan context—through data collection, model training, and interface localization—prioritizing situated technical capacity-building by recruiting a local developer to lead the effort. Both tools were successfully deployed and are in use by the community partner. Still, we found ourselves asking a question we suspect many justice-oriented technology efforts also face: "Was the project successful?" Beyond producing the technical artifacts, had we succeeded in sharing power and ownership throughout the development process? To explore this question, we turn to reflexivity as a way to foreground the process of technology production alongside its outcomes. Through collaborative autoethnography, and drawing on Data Feminism and African Data Ethics as analytical frameworks, we examine how structural factors and project design decisions limited or supported justice-oriented principles throughout the development process. We end by distilling broader insights on relationship-building, the roles of access and documentation in distributing control, and how shifting from performance metrics to partner-benefit measures can better ensure project sustainability. Through this work, we propose normalizing reflexive practice as a vital tool for assessing and advancing just AI development, offering both a critical account of one justice-oriented AI project and broader insights that can shape similar efforts.