Advancing the Use of LLMs in Civic Contexts

This project develops AI technology that extracts key information from video recordings of public meetings. There is a wealth of information in video recordings of public meetings that could help all American people. To access, process, and utilize this information systematically, new artificial intelligence (AI) tools must be developed for use by individuals, decision makers and researchers. The project develops methods to identify and extract patterns in the way issues are discussed during public meetings. These methods also identify the types of issues raised by participants. The final products – a large public dataset and open-source code – support an interdisciplinary research community studying how people interact in civic contexts. Large language models have accelerated the computational social sciences, facilitating rapid conceptualizations and labeling frameworks. However, off-the-shelf methods fail to make sense of nuanced concepts and have been shown to have severe limitations in important contexts. To exploit the advances of new AI technology while overcoming its weaknesses, the project proposes a human-centered approach. The project develops technology which can leverage domain expertise to teach LLMs to make sense of the nuanced concepts at the center of this work. This project produces both data and open source methods to process it, with the data accessible to researchers and the public. The data are used to answer pressing scientific questions about the topics discussed in public meetings; the emergence of new topics of issue discussion; and topic importance. 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.