NGA’s new artificial intelligence chief previews near-term priorities

In his early months as the National Geospatial-Intelligence Agency’s first-ever chief AI officer, Mark Munsell is determined to refresh that spy hub’s existing AI strategy to cover emerging and powerful frontier models, and inventory all of the programs and initiatives that should be under his team’s purview.

During an interview with DefenseScoop on Tuesday, Munsell reflected on his motivation for taking on this nascent role, and discussed those and other key priorities on his near-term agenda as NGA’s newly named CAIO.

“I think you’ll see a lot more energy around implementing generative AI in the analytic workflow,” he said.

Historically considered America’s super-secret mapping agency, NGA is the Defense Department’s functional manager for geospatial intelligence, or GEOINT. That discipline involves the capture of imagery and data from satellites, radar, drones and other means — paired with expert analyses to visually display and monitor physical objects and geographically referenced activities on Earth.

As Munsell suggested, the maturing realm of generative AI (which broadly incorporates disruptive models that can pull patterns within existing data to generate original content across many forms) presents heaps of opportunities and risks for NGA analysts.

The emerging technology and associated, advanced frontier models were already top of mind for Munsell as NGA’s director of data and digital innovation. In addition to serving as CAIO, he’ll continue to lead that directorate — and in doing so, he’s set on “putting AI into the hands” of the agency’s users.

“And what I mean by that is a lot of these … big models being developed by the big companies like Anthropic and Google and OpenAI are still not accessible by the folks at NGA. They are still not usable by the folks in NGA. So I would consider myself successful when our analysts and the officers at NGA have access to those — the world’s best models — on a top secret network,” Munsell told DefenseScoop. 

One primary task that’s already in-play for his team related to this intent, involves revamping NGA’s guidance for how its personnel adopt artificial intelligence to include additions that better confront generative capabilities.

“A new AI strategy will be published by NGA probably after the [2024] calendar year” ends, Munsell said.

Among other responsibilities as CAIO, he’s also working closely with the agency’s first-ever Responsible AI Officer Anna Rubinstein on pursuits to establish more technology governance assets and facilitate initiatives geared at strengthening the trustworthiness and reliability of GEOINT models.

“We know that the work that we do with AI is serious work — positively identifying targets, accurately geo-locating them — it’s serious work that could be used in warfare. And we knew how important it was for us to establish that program of responsible AI implementation and to have a senior officer in charge of it,” Munsell said. 

As CAIO, he now oversees “a couple-hundred-person organization” with offices, divisions and branches. Some of that staff will also be supporting Munsell in an ongoing but complex effort to produce a full inventory of all NGA initiatives and programs on artificial intelligence. 

“We’ve been doing the inventory for a couple of years now, but it’s one of the harder things to do,” he said. 

“There are designated programs that are building and making and implementing AI, but there are hundreds of programs that could sort of suddenly be using AI. And so at some point you do want to know every program and every initiative that’s using AI, so you can have some level of audit on it, some level of security on it, and some level of safety evaluation on it. But sometimes the AI is embedded in software that you’re buying commercially off-the-shelf, and then you’re bringing [it] into the agency and you might not even know it,” Munsell explained.

For that and other reasons, NGA’s inventory will be, as he put it, “less literal,” as well as “a little more logical” and “a little bit more about investments in programs that are deliberately either building or buying or implementing AI.”

In response to questions from DefenseScoop regarding why NGA tapped him as its first CAIO at this particular point in time, Munsell noted that the move is now obligatory for federal organizations under multiple recent directions from the White House and Congress. 

But in his view, the spy agency has also been ahead of the game, especially over the last four years. He called out defense “visionaries” who impacted NGA’s path to realizing AI and enabling the Pentagon’s high-stakes Maven computer vision effort — including former and current NGA and DOD officials such as Phillip Chudoba, Sandra Auchter and Drew Cukor.

“For me, it’s an honor to be the first [CAIO]. But in a way, I’m standing on the shoulders of giants — and I think you should say that, because it’s so true,” Munsell told DefenseScoop.