By: Lindsey Vickers

“You can’t FOIA without AI,” said no one, even though on the letters alone, it’s technically true. 

FOIA, or the Freedom of Information Act, is the federal framework that governs the public’s access to government agency records and information. It’s used by people across the country to request information in hopes of learning more about what our government is up to. 

Want to know about the TSA’s seizures of fireworks at airports in July? As journalists would say, you can FOIA that. Interested in state department agreements, memos, and treaties with foreign states pertaining to science and technology cooperation? You can FOIA that. Curious about the cost of a president’s golf excursions? You can FOIA that. 

However, as with many government processes, FOIA is not without its pitfalls. Agencies have a hard time processing the volume of FOIA requests they regularly receive, resulting in significant backlogs and delays. Now, agencies and FOIA officers are considering how AI might help with the process. 

What problems with FOIA could AI mitigate? 

It’s no secret that the federal government struggles to keep up with the volume of FOIA requests it receives. After all, anyone can submit a FOIA request, from private individuals, to business owners, to journalists. According to the Government Accountability Office, nearly 25% of FOIA requests are swept up in backlogs. This results in huge delays in fulfillment or even just acknowledgement of receipt. While just shy of a quarter of total requests might sound like a relatively small fraction of the total requests, the actual numbers paint a shocking picture: A whopping 200,000 requests got caught up in the backlog.

And that’s not specifically complex records, which include records spanning multiple subjects, sometimes “on different topics and in different formats.” In 2022, only 86% of complex requests had not been processed within 20 days of receipt. 

However, potential agency solutions are not ideal for requesters or the government. One solution is negotiating with the requester to narrow the scope of their request. This often includes shrinking the date range or the type of communication—such as altering a request to simply “emails” rather than requesting “all communications.

That’s where AI might come in. 

How could AI change the FOIA process? 

Government agencies and committees are looking into ways that AI could potentially be used in processing and responding to FOIA requests. 

A government group that’s part of the technology committee and studies the use of AI in the government and private industries, hopes AI can be used to help officials who fulfill federal FOIA requests to separate records into groups by concept or relationship. This technology could make records requests for things like “emails from employees of the Library of Congress complaining about the John Adams Building cafeteria food being cold” easier to fulfill. AI would likely be able to more quickly differentiate between other complaints about the food, such as qualms about the food being unhealthy, as well as complaints about other cafeterias. Instead of a person hand combing through each document to identify that it is about the John Adams building cafeteria, and the food being too cold, AI could quickly and accurately group documents that seem pertinent together.

What agencies have already applied AI to FOIA? 

A few agencies have already put machine learning to work at similar tasks. 

The state department first applied AI to the task of declassifying cables—or confidential typed messages sent and received by the government—which are normally assessed individually by humans to determine whether the information can be safely declassified on their 25th birthday, in accordance with an executive order. The AI program was trained on two years worth of human decisions made on official documents. It performed the same as human reviewers 97%-99% of the time. 

These impressive results led the state department to consider other applications of the technology, including FOIA. AI was then used in another pilot to address FOIA requests the department received. As with the cable declassification, the machine learning technology was highly accurate—performing as FOIA professionals would 97%-99% of the time. The state department has further broadened its potential applications of AI after these results. 

The Justice Department, and Centers for Disease Control are also testing out AI. These agencies are using the technology to manage record-breaking numbers of new requests, and the continual backup of existing ones. 

The implementation of AI into FOIA processes comes with other potential considerations. Would agencies use human professionals to check the technology’s work for each request? While it’s highly accurate, the results aren’t on par with human professionals 1%-3% of the time. This could have devastating effects on journalists, for example. AI misjudgments resulting in certain records being mistakenly withheld or excluded from FOIA requests would negatively impact journalist credibility. But, more significantly, it would jeopardize the public’s right to know, and ability to understand what the government is up to—a central tenet of the FOIA framework. 

#FOIA #AI #governmenttransparency