Collaborative Research: EAGER: NAIRR Pilot: An AI infrastructure to advance research in reducing poverty and informing social policy

This project builds an artificial intelligence (AI) infrastructure that enhances access to computational power, datasets, and models that support research on social policy and poverty. It will develop a novel trustworthy AI infrastructure that will 1) integrate, curate, and process diverse policy-related data sets; 2) develop and facilitate access to novel AI models that can learn from existing policies, providing insights such as what makes a program successful; and 3) open new areas of research and scholarship, creating multi-disciplinary communities of policymakers, economists, computer scientists, and government and non-government organizations to better address these societal challenges. This project demonstrates how the resources made available through the National Artificial Intelligence Research Resource (NAIRR) Pilot program can be employed to address a major social challenge - poverty. The effort develops a novel, trustworthy AI infrastructure that combines the power of foundation models (FM) with the reasoning capabilities of probabilistic graphical models (PGM) for evidence-based poverty eradication research. This project mitigates several technical challenges related to data, models, and their evaluation, providing foundational and use-inspired advances by: 1) Developing novel methods to build a knowledge database from diverse unstructured poverty eradication literature, such as academic literature and impact evaluations; 2) Combining the linguistic capabilities of LLMs with the reasoning capabilities of a PGM to answer policy-relevant questions; and 3) Developing methods to identify, quantify, and mitigate biases that may be induced in the system, to ensure that these systems are safe and trustworthy when deployed. This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.