Artificial Intelligence

DoD study sees ‘big breakthrough’ with using AI for declassification

The DoD study comes as Congress presses the Biden administration for progress on efforts to streamline classification and declassification.

Justin Doubleday@jdoubledayWFED

June 24, 2024 6:49 pm

3 min read

A Defense Department research project has seen success in using artificial intelligence and machine learning to manage and declassify records. The project leads say the approach could be used to help agencies manage an explosion in digital records.

The research study, “Modernizing Declassification with Digital Transformation” is sponsored by the Office of the Under Secretary of Defense for Intelligence and Security. It’s being carried out by the University of Maryland’s Applied Research Laboratory for Intelligence and Security (ARLIS), one of DoD’s University Affiliated Research Centers.

J.D. Smith, chief of the records and declassification division at DoD’s Washington Headquarters Services, said the research project validated a proof of concept that shows AI and machine learning models can use “contextual understanding” to perform records management and declassification functions.

“The big breakthrough here is the mapping of business rules to contextual understanding models,” Smith said during a June 24 Public Interest Declassification Board meeting.

Previously, machine learning models “weren’t quite there” to understand the context for different types of content, Smith said. He said it’s key for models to understand the distinctions between, for example, a Department of Agriculture document that describes a “kiloton of grain,” versus a DoD document that uses “kiloton” to describe the specific content of nuclear weapons.

“How do you break through that contextual decision making to a computer and train a computer or an algorithm on doing that,” Smith said. “And one of the big break breakthroughs that we discovered is you can actually do that now, with the algorithms that exist with natural language processing, named entity recognition, and other models, you can configure them to train on how to make a contextual decision making.”

Lawmakers want updates on declassification

The DoD project comes as Congress presses the Biden administration for progress on implementing the Sensible Classification Act of 2023. The legislation was signed into law as part of last year’s defense authorization bill.

In a June 18 letter to federal Chief Information Officer Clare Martorana, a bipartisan group of senators requested an update on efforts to develop a technology solution to support both classification and declassification.

“This opportunity to adapt our classification and declassification processes will greatly enhance the government’s ability to maintain accountability of our classified documents and records, streamline critical processes important to our national security, and work to reestablish trust and transparency between the United States government and the American people,” the lawmakers wrote.

Lawmakers are seeking answers to long-standing concerns about what one former official called a “tsunami of digitally created classified records.” The Biden administration has also kicked off a National Security Council-led process to reform the classification system.

‘Playbook’ for information review

Meanwhile, DoD’s declassification study will eventually result in a “playbook,” Smith said, for using technologies to support declassification and record management decisions in government. ARLIS is working on a “system architecture,” Smith said, as well as costs and other considerations.

The playbook will also turn into a request for proposals, he added, to help guide industry’s work with agencies on the supporting technologies.

DoD is looking to partner with agencies, including the Energy Department and the National Geospatial-Intelligence Agency, to further advance the project. Smith also said the DoD is looking to augment a State Department project that has used AI to declassify diplomatic cables.

DoD also plans to convene an interagency meeting this summer to discuss cross-government efforts and standardization.

“The principles that we’re going to explore here and show how we unlock technology to kind of navigate this, it’s applicable to any type of information review and release you’re doing,” Smith said. “Foreign disclosure, FOIA, security review . . . any type of information security review that you’re doing to clear anything, it follows these steps. And how do we map technology to each step to really make things efficient from a reviewer standpoint?”