Robust Poverty LLMs that Support Poverty Elimination

​The past decade has been marked by a decline in support and funding for initiatives in the humanitarian and development sector, best exemplified by the dismantling of USAID in early 2025. These trends have left knowledge and funding gaps that further highlight the need for new approaches in addressing global poverty. We deployed a novel, integrated approach for finetuning a foundational large language model on a diverse range of international development data, with the hope that this work could be used as a policymaking tool to better inform more effective poverty and development solutions. 

Robert Krueger and Esther Mao of Worcester Polytechnic Institute will share work connected to inste^d, an AI-driven platform designed to help policymakers and aid organizations better understand multidimensional poverty and identify more targeted, evidence-based interventions. Drawing on WPI’s interdisciplinary strengths in social science, policy, data science, and project-based research, the session will examine how large language models and diverse international development datasets can support more effective poverty diagnosis, policy design, and resource allocation.

Rather than relying only on income-based measures or top-down assumptions, this conversation will focus on how AI can help surface patterns, context, and community priorities in ways that strengthen human agency and improve decision-making. Participants will consider both the promise and responsibility of using AI as a tool for more responsive, beneficiary-centered poverty alleviation.

​This session will be held in English with simultaneous live Spanish translation.