On January 23, The Rebooting Democracy in the Age of AI lecture series hosted Natalie Buda Smith, the Director of Digital Strategy at the Library of Congress, the world’s largest library, about her team’s work, investigating how Congress can use artificial intelligence to enhance its operations.
Watch the video, read the transcript or view the slideshow of Natalie’s explanation of such projects as:
-
Legislative Data Analysis: The Library is experimenting with generative AI and text analysis on Congress.gov data, specifically focusing on enhancing metadata around geographic entities and organizational names in legislative materials.
-
Bill Text Analysis: They are conducting experiments to develop authoritative and authentic bill summaries using AI, though Natalie noted that current results aren't yet comparable to the quality of summaries produced by Congressional Research Service analysts.
-
Model Testing and Evaluation: The Library tests both commercial models and open-source models (from platforms like Hugging Face) against their domain-specific Congressional data. They've found no single model performs well across all their needs, and accuracy remains a significant challenge when dealing with specialized legislative content that requires high standards of authenticity.
-
Congress.gov API Enhancement: While not strictly an AI project, the Congress.gov API (launched in 2022) processes up to 40 million hits per month and serves as a foundation for making legislative data machine-readable and accessible for potential AI applications.
-
Performance Monitoring: The Library uses AI-powered tools to monitor and analyze the performance of their legislative systems, helping identify inefficiencies and potential issues in their digital infrastructure.
Natalie emphasized that their approach to AI is focused on augmenting rather than replacing human expertise, with a strong emphasis on maintaining authenticity and authority in all their Congressional work. She noted that their experiments are still in relatively early stages, with the most successful applications being those that support staff workflows rather than fully automated solutions.
The talk was followed by a robust discussion with questions such as:
-
Beth Noveck (Northeastern, The GovLab), asking about how the U.S. approach to legislative AI innovation compares to other countries like Chile and Brazil, highlighting the more distributed nature of AI functions across U.S. legislative bodies versus centralized innovation labs elsewhere.
-
Anne Washington (NYU) inquired about research questions that could help support the Library's work with large language models in the public interest.
-
Daniel Schuman (American Governance Institute) asked about initiatives to improve public access to legislative data and ways for the public to collaborate with the Library on tool development.
-
Jim McGlynn (Massachusetts House of Representatives) raised questions about how the Library is balancing immediate action with long-term planning given the rapid pace of AI development.
Moderated by Beth Simone Noveck, the conversation was part of the The Rebooting Democracy in the Age of AI lecture series hosted by the Burnes Center for Social Change at Northeastern University, and its partner project The GovLab, in partnership with the Institute for Experiential AI.
Learn more about AI and the Library of Congress or read the Library of Congress Blog.
For more resources on this topic, check out The Library Of Congress Is A Training Data Playground For AI Companies (Forbes), and Could Congress Leverage AI to Help Restore Faith in US Democracy? by Lorelei Kelly.
Learn more about and connect with Natalie on LinkedIn.
The series continues on February 27, 4 pm E.T. (online and in-person at Northeastern’s EXP Building, Room 610) with The Habermas Machine with MIT and Deepmind's Michael Henry Tessler and Michiel A. Bakker.
March 20, 5 pm E.T. Ed Bice of Meedan will talk about combatting misinformation in elections with AI.
Sign up all of our upcoming events: rebootdemocracy.ai/events
