The State Department is looking at artificial intelligence and automation tools to process Freedom of Information Act (FOIA) requests more quickly and improve its level of service to requesters.
Eric Stein, the department’s deputy assistant secretary for the Office of Global Information Services and co-chairman of the Chief FOIA Officers Council’s Technology Committee, said he’s also looking at ways to use these emerging tools to improve FOIA processing governmentwide.
“I think people are afraid of AI, and maybe they should be. Maybe they shouldn’t be, but my take is, we’d like to get people comfortable with the concepts of AI and machine learning,” Stein said.
The Technology Committee a few months ago went through each the most recently published Chief FOIA Officer reports governmentwide, to better understand what tools agencies are using, and what the committee can do to address their problems.
Agencies generally are seeing an increase in FOIA requests, and are looking for ways to stay on top of this workload. The federal government in fiscal 2022 received a record high of more than 900,000 new FOIA requests.
“What we found is that there are tools in place being pushed to their limits and a lot of assumptions about how records are captured, stored and searched. It’s still very manual,” Stein said.
Federal employees can process most FOIA requests involving unclassified records while working remotely, but Stein said FOIA professionals still need to work in the office to handle classified records.
“Because of that, there’s a balance in how do we recruit and retain people that maybe don’t want to go into the office,” Stein said.
State Department pilots AI for FOIA workloads
The State Department received nearly 14,000 new FOIA requests in fiscal 2022. Nearly another 21,000 FOIA requests are pending, according to the department’s annual FOIA report for FY 2022.
To stay ahead of this workload, the State Department developed its e-Records Archive, which holds more than 3 billion department records.
Stein said the archive reflects the work of chief FOIA officers over the past decade identifying standards for metadata and capturing records, so that records are optimized for search.
“We developed it in such a way that the data standards are in place, so that we could use that data down the road,” Stein said. “I think a lot of agencies probably have Outlook and different emails tools … but they may not have a central archive they can search across, the way we can here at State.”
The State Department is also experimenting with automation to improve the process of filing a FOIA request.
Stein said the department, in a pilot to improve the process of declassifying records, trained a machine learning model on years of humans reviewing and declassifying records.
The model is now as accurate as human FOIA professionals about 97-99% of the time. Stein said the pilot so far saved the department about half a year’s worth of work.
State Department, he added, is also looking at ways AI could help it use information already released under FOIA to support additional incoming requests.
“Right now, it’s more like you do a search and it looks for the key term or this or that. But making more sophisticated connections among information, it might give you what you need, or at least help you to scope your request, in a way, because we do want to get information to people as quickly as possible, [and] a lot of requests are very broad,” he said.
The department is also considering ways to use AI to help it refine records searches.
“If you look for a very specific term, you get a result. But if that term has the same letters in it as other words, you can end up getting like a million potentially responsive records, when really, you wanted a very narrow, specific thing,” Stein said. “Maybe people colloquially use a different term for something. The machine learning tool could help narrow those results to get more timely responses.”
While automation is showing promise in some FOIA pilots, Stein said the federal government still has a lot more work left to do to understand how best to use this emerging technology.
“I think some people are going to fail at some of this work, and that’s OK, in my opinion, as long as we learn from it, and we can save time by sharing here’s what works, what didn’t work,” Stein said. “Which is also why we’re so proud of the machine-learning pilot that did work, because we’re sharing this work with other agencies.”
The State Department also recently soft-launched its new online request platform, which allows users to track the status of their FOIA requests. Stein said the new platform is meant to reduce the number of calls the department’s FOIA office receives, asking for this information.
“We’re rethinking our online experience to help people find information they want,” he said.
The new platform also supports the department’s “release to one, release to all” policy of posting documents released under FOIA to its virtual reading room. In fiscal 202, the department posted over 6,200 records online after it released them to requests.
Stein said FOIA professionals under this policy are trying to post as many documents online as possible each month, but are looking at ways to streamline this workload through automation.
“A lot of it just comes down to that’s still a manual process. How do we automate it? How do we do a better job of getting information out? It seems to be working pretty well, but there’s still a long way to go.”
Demystifying AI for agencies
One of the barriers to adoption is getting more of the FOIA community to understand the opportunities and limits of what AI can do.
Stein said the Technology Committee held an “AI 101 course” a few years ago, in order for more FOIA professionals to develop a common understanding of this technology.
“People just assume everyone knows what artificial intelligence is. It is, on one hand, complicated, and on the other kind of easy to understand if you break it down,” Stein said. “If you’re coming from a place where you think we just do a Google-like search across all records at every federal agency, that’s just not where we’re at. And that changed the whole discussion.”
The Chief FOIA Officers Council also held a “NextGen FOIA Tech Showcase” in February 2022 to identify technologies that could help agencies process their FOIA requests more easily.
The showcase gave agencies an opportunity to learn more about AI, machine learning and tools to improve the customer experience of filing FOIA requests.
Stein said the showcase helped break down some barriers and understand to tools the private sector can offer. But some agencies have unique considerations when it comes to FOIA processing, and AI tools may not be the best way to address those challenges.
“If you’re an agency that gets thousands of requests annually for a specific form, you may not even need AI and machine learning. You many just need a simple tool or an application that could redact certain boxes,” Stein said.
The Technology Committee is looking at collaborative interagency platforms, in an effort to help agencies trying to optimize FOIA processing at agencies with limited IT budgets.
“Maybe a couple of agencies could some together, thinking in a different way, in new ways. [It’s] not just one agency working on it, but several agencies working together, especially those that have high volumes of requests. We could really see some efficiencies through a shared platform and ways to manage information well and securely too,” Stein said.