Weekend Read: National Archives Sunshine Week Panel Focuses on Artificial Intelligence and Government Access

Federal agencies must prepare to integrate artificial intelligence and machine learning capabilities while continuing to ensure government transparency, according to the March 14, 2024 panel hosted by the National Archives and Records Administration (NARA), Artificial Intelligence: The Intersection of Public Access and Open Government. Expert panelists from NARA, The State Department, Department of Justice, and Library of Congress gathered at the William G. McGowan Theater at the National Archives during Sunshine Week—the annual celebration of open government—to discuss how artificial intelligence intersects with the Freedom of Information Act (FOIA) and access to public information.

Here are some big takeaways from the panel:

Eric Stein, Deputy Assistant Secretary for the Office of Global Information Services (A/GIS) and Senior Agency Official for Privacy at the State Department, outlined the agency’s goal to improve the experience of FOIA requesters and State Department employees processing those requests through its FOIA AI Pilot Project. According to Stein, the purpose of the project is to: 1) identify similar FOIA requests—both existing requests and as new requests are received, 2) utilize automation instead of a manual process to search State Department Central Records and Archives, and 3) improve the searching capabilities on the public website and overall customer experience.

Currently, the agency spends $65 Million on annual FOIA processing costs, employs 267 full-time FOIA staff to process requests, and has 20,000 pending FOIA cases with around 15,000 new requests received annually. “Manual processes are inefficient,” said Stein. “In certain cases, the Department was duplicating efforts at various stages of the FOIA process, sometimes re-processing documents from other cases.” With the help of the pilot-produced 360 FOIA Matching Tool, Stein stated that FOIA staff will be able to more effectively search FOIAXpress and State Department Reading Room results, ultimately cutting down on processing inefficiencies.

For additional resources on the State Department’s FOIA AI Pilot Project and Artificial Intelligence Strategy, see:  

• June 23, 2023 National Security Archive Unredacted Post, Questions for the State Dept. After News it Launched a Pilot Program for Declassification Work, Anniversary of Rosenberg Execution for Espionage, and More: FRINFORMSUM 6/23/2023
• September 7, 2023 FOIA Advisory Committee Meeting State Department Presentation, Piloting Machine Learning for Freedom of Information Act (FOIA) Requests
• November 9, 2023 State Department Fact Sheet, The Department of State Unveils its First-Ever Enterprise Artificial Intelligence Strategy

Similarly, Bobak (Bobby) Talebian, Director of the Office of Information Policy (OIP) of Department of Justice, highlighted the DOJ’s commitment to “improving [FOIA] access and the requester experience using artificial intelligence.” Talebian reviewed the DOJ’s new “advanced search tool” on FOIA.gov released in October 2023, which utilizes a “combination of logic and machine learning functionality.” 

Perhaps the most charged moment of the panel came during the question-and-answer session, of which National Security Archive Fellow and Washington Post FOIA Director Nate Jones was an audience member. Jones asked Stein about how the State Department’s AI integration into its FOIA processes might present an opportunity for agency proactive disclosure, specifically regarding provision (a)(2)(D)(ii)(II) of the FOIA Statute which requires agencies to release records to the public that have been requested three or more times. Stein responded that while the State Department “doesn’t actually keep track if a record gets requested three times” because the agency follows the “release to one release to all” policy, AI will help FOIA staff make associations they weren’t previously able to make, ultimately improving record tracking and release. Talebian expanded on Stein’s response, adding that “identifying proactive disclosures is very challenging for agencies to do…and there is a potential for AI to help augment what agencies are already doing at a human level that can help bridge the gaps.”

Another audience member pressed Talebian about the potential for DOJ proactive guidance in federal agency use of generative AI, remarking that “if it’s not DOJ [leading], who does the buck pass to?” Talebian skirted these remarks, stating that OIP exclusively manages FOIA processes, and could not comment on a more global use of AI within federal agencies.

Other general takeaways from the panel:

According to panelist Abigail Potter, founding member of the Library of Congress Labs team, data management—much more than data processing—will be one of the biggest challenges that agencies face when implementing AI capabilities. Additionally, issues of streamlining processes across all federal agencies—ranging in size, resources, and structure—as well as transparency to requesters are still yet to be solved. Gulam Shakir, NARA’s Chief Technology Officer, highlighted the imperative for agencies to think about the authenticity of their documents amidst generative AI so “the user can verify if it’s real or pulled out of thin air.”

Outside of those agencies represented on the panel, it remains unclear how other agencies will integrate AI and machine learning technologies into their FOIA processes—and how these changes will impact requesters.