TAMPA, Florida–Vice Adm. Frank Whitworth has satellite imagery for virtually every corner of the globe. But what keeps him awake at night are not the threats he knows about. It’s the ones he or the intelligence analysts under him might miss.
“Warning is the big behemoth for us,” the head of the National Geospatial-Intelligence Agency told Defense One on the sidelines of the Global SOF event here, “When you’re talking, ‘Are we keeping an eye on all these complexities in the world? Have you established a baseline everywhere? Will you be able to warn that something is anomalous and prevent surprise?’ That is a big deal and that can be humbling,” he said.
But new approaches to data and AI, which are already “bearing fruit,” are creating ways to anticipate threats before they emerge.
NGA uses AI to analyze satellite and other data, then alert analysts, in a program of record called Maven. It started in 2017 as a small “pathfinder” program for Air Force Special Operations, but has grown considerably, particularly since NGA took over three years ago. Now, Whitworth said, demand is growing quickly across all the services and combatant commands.
Maven is primarily used to find potential targets, which means the program must be trained to look across large amounts of visual and other data to find specific things. But the next chapter of Maven, which will include new AI reasoning capabilities, could allow for the detection of “anomalies,” Whitworth said: an adversary truck in a location it wasn’t expected, a weapons facility suddenly very busy or not busy at all. That opens the possibility of detecting not just known threats but potential ones.
“If we take the best of what we’ve done with Maven and apply it to warning, we actually could find this not so humbling,” he said.
The development will require making new models that can—in a way—reason based on what they’re detecting, rather than just identifying objects. “We will have models for the pieces of equipment or even behaviors that we want to prioritize,” he said
NGA this year also established a program executive office for advanced analytics, run by Maven program manager Racheal Martin.
“She’s taking a look at not only Maven, but also ASPEN,” the Analytic Services Production Environment for the National System for GEOINT, Whitworth said.
The agency launched ASPEN in May 2023, aiming to help NGA grapple with “the near tripling of GEOINT data coming to NGA’s analysts,” according to a fact sheet.
Whitworth described it as “is a major system acquisition that’s dedicated to [data] workflow of imagery and is only partially based on AI.”
There are other types of data that NGA that go into helping models be more accurate and confident in the threarts they are seeing. Analyst training, particularly in how to think about AI and new forms of data, is key to surveying more of the globe and improving accuracy.
“Distinguishing enemy from non-enemy, combatant from non-combat, you need to have layers,” he said. “I’m very confident in the training of our people and also their ability to train these models. However, in this business, you’re always looking for some corroboration to make absolutely sure, ‘Yes, we have positive identification.’” You need that if you’re going to brief a combatant commander or even the President on a potential threat, he said.
Efforts are already showing some limited results. “I’m seeing more analysts sit at the table with the people in Maven, our innovation team and our AI team, describing the problems they have that they need AI solutions for.”
One of those problems is getting the data out to troops on the battlefield and enabling them to make their own tools to find what they need. To that end, NGA is looking to get more granular and portable in the warnings and targeting it delivers.
Another NGA official described an effort called GAMBLER (GEOINT Artificial Intelligence and Machin Learning Light-Edge Resilient System,) currently in beta testing, to develop “AI-enabled models for tactical or fighters. So, basically, they’re building their models on the ground based on what they’re seeing from largely drones and other hand-held sensors.”
The number of soldiers, sailors, airmen, Marines, and guardians, who have joined the Maven user base has quadrupled in just one year, Whitworth said. More and more commanders across the services now have confidence in AI and see the urgent need to incorporate it into their operations.
Whitworth says NGA has been able to keep up with the demand. In fact, he said that latency—the time between when an object shows up and when it’s “detected” as a potential target—has gone down by 80 percent. He ascribed that to not only better, continuous work on models to improve accuracy and efficiency, but also the foresight to acquire enough computing power to meet the customer demand.
That demand is likely to continue to grow, and could outstrip compute supply at some point in the near future, he said. The ability to meet the growing intel demands across the force “could be a temporary situation, If we’re not careful.”
The post AI is helping the Pentagon go from finding targets to predicting threats appeared first on Defense One.
