Leveraging AI in ‘No-Fail Mission’ for U.S. Intelligence

EXCLUSIVE INTERVIEW – It’s hard to overstate the complexity and importance of the work of the National Geospatial-Intelligence Agency (NGA), which — by its own definition — involves delivering “world-class geospatial intelligence that provides a decisive advantage to policymakers, military service members, intelligence professionals and first responders.” 

As a practical matter, that amounts to intelligence collection on an enormous scale, often involving needle-in-haystack searches with no margin for error.

“We have to be correct,” VADM Whitworth told The Cipher Brief during a recent State Secrets podcast interview. “This is a no-fail mission when it comes to warning and positive identification.” 

VADM Whitworth – who was honored Friday with the Cipher Brief’s Impact through Government Innovation Award at The Cipher Brief HONORS Dinner – spoke with our CEO & Publisher Suzanne Kelly about how the NGA is leveraging the power of artificial intelligence (AI) in its work.

The conversation has been lightly edited for length and clarity. You can watch the full discussion on The Cipher Brief YouTube channel and listen to it on Spotify and Apple Music.

The Cipher Brief: Artificial intelligence is the buzzword these days. It’s much more than that to you, as you’re figuring out new and innovative ways to incorporate the technology at the crazy rate that it’s developing into the mission at NGA. Talk to us about how you’re progressing in terms of bringing AI into that mission.

VADM Whitworth: It’s the natural question. We all want a report card, and there are some metrics. It’s probably important to baseline what we do with AI and ML (machine learning) in this business, in this tradecraft of imagery analysis, and in some cases GEOINT writ large. 

We train models. The idea here is something called computer vision. I think most listeners would be able to relate to computer vision when they go through TSA Global Entry. There’s a camera, it sees your likeness, it compares it to your passport, and then we all marvel as you walk through without even breaking out your passport because it has completely matched you. That’s computer vision. I wish we had an easy [task] like that. 

The task that we have involves an infinitesimal amount of the percentage of a field of view. When you walk through Global Entry, your face is about eighty percent of that field of view of that particular camera. We’re searching for things that are being denied. They’re being hidden. These are secrets that people want to maintain in other countries. And secondly, they’re really, really small compared to the big field of view when you’re taking a picture from space. We approximate this at about 2/100th of a percent of the field of view, for something that we need to delineate its identity. We do that through this process called computer vision. 

For years, humans have been doing this with their own processing unit known as the brain. Now we’re trying to do this so that we can do it faster, with more of the land or the sea available for the review to come up with something. In the case of our warning mission, [it] is [something] anomalous, different from the baseline that we’ve seen over the past. In the targeting mission, it’s something that needs to be distinct — combatants from non-combatants, enemy from non-enemy, adversary from non-adversary. These are very important prerogatives of the Commander-in-Chief, of the Secretary of Defense, and of combatant commanders.

So we take this process very seriously, and it takes humans to train the machine to ensure that it’s right. So when we send all the data that comes from collection through these models, we run something called inference in the computer vision process that yields detections. And the best way to measure what kind of a year you’re having is to measure how quickly you’re running inference.

The intersection of technology, defense, space and intelligence is critical to future U.S. national security. Join The Cipher Brief on June 5^th and 6^th in Austin, Texas for the NatSecEDGE conference. Find out how to get an invitation to this invite-only event at natsecedge.com

At the unclassified level, it’s been a good year. This issue is called latency. Our latency has improved 80% in just one year. We’ve had this be a program of record for just over a year, so I think that’s a pretty good track record. But we’re never completely satisfied.

The Cipher Brief: You’re joining us from Colorado Springs, where you’re speaking at the Space Symposium on the relationship between Space Force and the NGA and how the two of you are collaborating. So it’s not just the Earth anymore, which used to be NGA’s mission, but now it includes space. Talk to us about that.

VADM Whitworth: NGA relies on the warfighting domain of space. And likewise, the warfighting domain of space relies on NGA. So we changed our motto to read, “From Seabed to Space.” This was just about two and a half years ago, and it was the right thing to do to remind people that we look up, not just down. We’re not just looking at the Earth itself, but we are also looking up. This issue of distinction applies up there as well. And so it is appropriate for the United States Space Force Vice Chief and I to be on the stage together.

In the past, we’ve seen allusions in the media to there being some sort of strife or growing pain. I think that while that might’ve been true in the past, right now we’re on a very good run and we wanted to articulate that publicly. 

The future is very good, especially when you use innovation for centers like we are talking about in the Joint Mission Management Center. The last time you and I talked, it was just a glimmer in our eye. Now it’s very real. It’s what we call IOC, Initial Operating Capability, where there are guardians working with NGA and the people who do tasking of these satellites, to ensure that we know the entire denominator, and before any pixel or any analysis is bought, any task is rendered, we make sure that as good stewards of taxpayer money, we’re not paying twice. So it’s good for warfighting, it’s good for speed, and it’s been really good for the relationship between NGA and the United States Space Force.

The Cipher Brief: How are you thinking about the private sector as a partner now, as your mission continues to evolve?

VADM Whitworth: They’re so essential. We learn from each other. I think in the issue of speed when it comes to acquisition, all of us are still learning and trying new things. We’ve been using CSOs (commercial solution offering). It’s like a pre-vetted relationship that allows you to move very quickly when something appeals to your mission. [For example] the CSO known as Project Egger has been employed for maritime domain awareness. That has afforded us better speed than we would have had in the past. And that was in direct relationship to a requirement uttered by one of our primary combatant commands. In this case, it would be INDOPACOM. 

The Cipher Brief: We heard earlier this year at the worldwide threats hearing that the new administration is putting a focus on stopping drug trafficking along the border as one of their priorities. And it seems like NGA is ideally positioned to help understand what’s happening there. What can you tell us about that?

VADM Whitworth: We had two decades of concerted activity-based intelligence during the counterterrorism years. Some of that actually does apply to the problem set that you’re describing — how to ensure that we can detect and ultimately help our law enforcement partners with this scourge. It’s happening as we speak. Many of the briefings that I take as we listen to the Commander-in-Chief’s priorities, and we react and do activity-based intelligence and behavior-based intelligence, it transports me back to the days when I was in Afghanistan. It transports me back to the days in Iraq. We were hunting terrorists and trying to localize that activity and ensure that it had a proper end. It does feel as though we’re utilizing our best practices.

The Cipher Brief: I know that NGA has also been really focused on the Arctic and mapping that region. What can you tell us about that? 

VADM Whitworth: We have a responsibility to know the Arctic and to understand the navigability of the Arctic, just because we have a responsibility for safety of navigation. So it’s not necessarily about climate or climate change; it’s because we need to ensure that we monitor all the ice. We need to ensure that we understand the poles; most of the change regarding the Earth occurs at the poles, which has an effect on the spin rate of the Earth and topography that ultimately yields accuracy. So we have a responsibility to understand the Arctic. 

From a strategic competitive perspective, yes, we must absolutely invest in time and energy dedicated to knowing when navigability is changing, and what that will portend either for the Russians or the Chinese and their behaviors up there. So we do have a responsibility there. 

The Cipher Brief: What do you see coming in the next six to 12 months? What should we be thinking about?

VADM Whitworth: When it comes to targeting, our marquee item obviously is NGA Maven. And we’re making great progress. We mentioned latency, and that’s been demonstrated by NGA Maven. Warning is where we’re trying to take the best practices of AI, ML and targeting and applying it to the big behemoth problem that is warning, and to sight when something is anomalous on this earth of ours. And we’re also looking for ways to improve our confidence. I’ve been a targeting professional now for almost 36 years, and I’m always looking for the most positive identification to ensure that it’s not just based on one source. I’m trying to ensure that we have multiple sources overlapping that point to a high-confidence, positive identification. GEOINT is very good on its own, and I trust GEOINT and our people, but it’s always nice when you can add layers. Adding layers is where we’re making progress, as we speak, in AI and ML. 

I think one year from now, we will talk about multimodal and its impact on our readiness. We have at least three multimodal tests going on right now, and they are actually showing that this is going to make a difference in our speed and the completeness of positive identification, which will in turn help us with analysis. It will help us with exploitation targeting. It will help us indirectly with our collection and our tasking, because you have to understand where to look in order to have good collection and good exploitation. It’s kind of the left-hand, right-hand relationship that we have as the functional manager for GEOINT. So I’m excited about multimodal right now, and the application of AI for the future.

The Cipher Brief: If you were explaining multimodal to someone who may not have a crisp understanding of what it is, how would you describe it?

VADM Whitworth: It’s taking several layers and ensuring that a model can actually translate a computer vision solution to what might be something in text. So generative AI, GenAI, that’s a lot of text. It’s very agile, because text is light. It’s not as dense as an image. But a lot of times those models don’t necessarily understand the data inherent to what we do with location, or what computer vision does with regard to a physical image. We’re now starting to find ways that multimodal allows those things to talk to each other. Some of these are showing great promise and it will allow us simply to have a conversation with these models and applications. We’re using NGA Maven as one of the best test cases of that. 

But the same is true on warning. We have to be correct. This is a no-fail mission when it comes to a warning and positive identification. So it makes total sense that multimodal, combining layers, is being done at NGA.

Read more expert-driven national security insights, perspective and analysis in The Cipher Brief