How Tech-Assisted Review is Changing FOIA

The Freedom of Information Act (FOIA) process is often plagued by delays, but the massive increase in data production over the past decade means that this problem is only growing. Manually handling all pending FOIA requests is no easy task—it’s impossible.

“None of us have time in our lives to read petabytes of data,” John Fasciola, a retired U.S. Magistrate Judge and Georgetown law professor, said Tuesday at the Digital Government Institute’s 930gov ​​panel.

That’s where AI comes in. Agencies are moving to technology-enabled assessment tools designed primarily for the edcovery process.

“Do you have any responsive records?” We’ve evolved from sending memos or protocols to email. Some agencies, if they’re lucky and have the right tools, don’t send anything to custodians,” said Glenn Melcher, special counsel for edcovery at the Consumer Financial Protection Bureau (CFPB). “In fact, they just collect it at the back end and filter it and dry it.”

Fasciola confirms that these technologies don’t simply save man-hours — they’re more accurate than manual assessments, he says.

“There is a false equivalency between manual assessment and technology-assisted assessment,” he said. “Maura Grossman and her husband, Gordon Cormack, have proven beyond question that technology-assisted assessment is more effective and more accurate than humans in years of testing at the National Institute of Standards and Technology.”

FOIA allows any person or organization to request any federal record, and the agency is required to produce that record within 20 days — however, the government has notoriously missed that deadline.

“For the movie’s spoiler alert, that doesn’t always happen,” said Michael Sarich, the Department of Veterans Affairs’ FOIA director.

Technology-assisted review has the potential to significantly reduce FOIA submission times and save staff hours, and Sarich hopes these tools will allow the government to meet the needs of FOIA requesters.

“Being on the front foot with these tools and being able to work with your requesters and have a front-end relationship is really important in the FOIA space, so a FOIA request really becomes a conversation between the requester and the FOIA processing community. Give the requester what he needs as soon as possible,” he said. “If it becomes widely accepted, these tools could make significant progress in that way – providing that information in a timely manner.”

Sarich also hopes that researchers will have more confidence in AI tools with their proven effectiveness.

“People who have been in the FOIA world for a long time, the way they produce something, the requestors come back and say, ‘OK, because I know what this is and what this is,'” he said. I think so.”

But for Melcher, one of the challenges associated with technology-assisted assessment is a lack of overall understanding of the process. It also doesn’t help that policies are slow to keep up with new technologies and that the rules and regulations for FOIA and eddiscovery use around AI aren’t always clear.

“That’s the problem with AI: I don’t understand what’s in that black box,” he said. “I knew how to provide documents when I ran test runs, but I couldn’t explain it in front of the judge. And to be honest, with rare exceptions, I’m not sure there’s a judge who would help even if I could. I think the problem with AI is, is there a legal issue here? You have to show your work, what have you done? Or should you come up with a result that is reasonable under the circumstances?

However, the time to rely on Fasciola’s “eyes on paper” has come and gone.

“We can’t do anything because this technology has proven itself now,” he said. “So what we’re dealing with judges is to dispel the notion that if we put a piece of paper in front of an infinite number of people, they’re going to get what they want in the end. The answer is the same, no they won’t; And two, it costs money you don’t have. I don’t think we have much choice in the matter.