The technical lead for the Pentagon’s new Task Force Lima said this week that his team is currently working across the Defense Department (DoD) on generative AI tech to help the services “unlock access” to troves of knowledge.
DoD established Task Force Lima, which is aimed at assessing and integrating generative AI capabilities across the Pentagon, in August 2023.
“What we’re trying to do is usher in and enable generative AI across the services, which is a really complex task, especially considering there are millions of people across the department,” Glenn Parham, software engineer and data scientist at DoD’s Chief Digital and AI Office (CDAO), said at the Palo Alto Networks Public Sector Ignite event in Tysons, Va., on April 2.
“At a high level, what we can do from the Office of the Secretary Defense perspective, is try to essentially unblock a lot of the barriers on a technical side and on the policy side,” he said.
Parham said that his team was approached to use generative AI to help with cyber defenses across the DoD.
“We’re using large language models to do a lot of summarizations of various notable security incidents and so forth that have arisen over the past several months,” Parham said. “And in terms of the analyst productivity … it usually would take them two hours per notable incident, now they’re able to do it in 10 or 15 minutes, so it’s saving them a considerable amount of time.”
“This is a pilot program,” he said. “There are only like 200 or so users in this network, so it’s not a very large network, which is good to start small. But that’s a program where I’m pretty happy to see us be able to measure the actual amount of minutes saved, because … we’re able to justify that it’s actually enhancing productivity.”
Parham said he has also helped on projects using generative AI to help sift through DoD policy documents.
“In terms of actually using AI to our advantage to navigate existing DoD policy, there are actually several projects that I’ve been a part of to assist with that,” Parham said. “That actually touches on taking existing foundational models, conducting fine tuning operations on DoD policy documents, or you can use things like retrieval augmented generation.”
“If we can use our generative models to unlock access to that trove of knowledge, that’s a huge win for us,” he said.
