Public Interest Computing: Pluralistic Design of Ethically Responsible AI and Policy

Abstract

AI-enabled systems profoundly shape our daily lives, influencing our interactions with the world. Despite their vast impact, advanced AI technologies, like large language models (LLMs), often elude billions grappling with poverty and global climate challenges. This inequity underscores the urgent need for policy-oriented AI research focused on driving social change and sustainable development. This talk centers on real-world field research dedicated to designing AI-driven decision-making systems in resource-scarce environments. Emphasizing distributive justice, I illuminate sociotechnical challenges faced by mainstream AI designs rooted in evolving norms, algorithmic biases, shifting data distributions, and enduring colonial legacies. The talk delves into innovative Responsible AI designs to augment climate change mitigation policies and addressing market failures that have led to over 300,000 farmer suicides. I introduce Public Interest Computing as a framework proposing the integration of social and ethical values into Human-AI decision systems from their inception, ensuring foundational guidance for equitable AI design, implementation, and governance. By prioritizing historically marginalized communities and local partners, this framework harnesses AI-enabled insights to scientifically inform high-stakes policy decisions and improve the well-being of invisible millions. Throughout the talk, I emphasize the crucial need for conscientious progress in public sector AI, highlighting its significance not just as a focal point but as a pathway to equitably distribute the benefits of digital technologies across society.