The National Geospatial-Intelligence Agency (NGA) is processing imagery 80 percent faster than it did a year ago. However, as satellite constellations and the amount of related data increase, the latency gains could diminish without more computing power, NGA Director Vice Adm. Frank Whitworth said on Monday.
Whitworth described the drop-in latency times in terms of “running inference,” which refers to running the agency’s artificial intelligence models against the imagery data to automatically find targets of interest for warfighters and decisionmakers. The 80 percent decline is a dramatic boost considering a year ago Whitworth boasted that its analysts have sifted through 40 percent more sensor imagery since 2023 because of NGA Maven, the agency’s computer vision-based technology that finds potential targets to attack (Defense Daily, March 21, 2024).
Despite a cascading deluge of data that NGA is processing and expects to process more of in the years ahead, Whitworth feels good the agency’s resource needs will be met.
“I’m confident in our discussions with both on the DoD and on the DNI (Director of National Intelligence) side, that we’ll find those resources and we’ll make sure that we keep up with that demand,” he said at the annual SATELLITE Conference in Washington, D.C.
This year NGA’s theme is accelerating AI, which Whitworth described as “NGAI.” One of the agency’s goals to help accelerate AI is around leadership organization, some of which was done last year and some this year. NGA last fall formalized the role of Chief Artificial Intelligence Officer, a role filled by Mark Munsell, who is also director of the Data and Digital Innovation Directorate (Defense Daily, Oct. 30, 2024).
Munsell is the “director of AI standards,” Whitworth said.
In addition, Trey Treadwell, NGA’s associate director of capabilities, is also the chief acquisition executive, and is in charge of AI programs, Whitworth said, adding that this fits with the agency’s move to a program executive office approach to acquisition management.
More recently, in January, NGA hired Joe O’Callaghan, a retired Army colonel who was the fires coordinator for the 18th Airborne Corps. He was an early adopter of the Maven Smart System.
O’Callaghan is the director of AI Mission and will continue to work from Fort Bragg, N.C. Whitworth said that he is a Defense Intelligence Senior Leader and a senior leader of NGA, “So he is able to make decisions about operations.” He said that O’Callaghan “has hit the ground sprinting.”
This “task organization recognizes the significance of AI” and is the first of Whitworth’s New Year’s Resolutions Whitworth has for accelerating NGA’s approach to AI.
“We will have ops integration teams with all of the people who do normal, structured observations as part of their normal trade craft to ensure that they’re thinking through and processing opportunities for AI,” he said.
Ensuring NGA has the resources, particularly around computing power, which “doesn’t come cheaply,” is his second resolution, Whitworth said.
Next is accreditation and certification. Certification is about responsible AI training, and the former is Accreditation of GEOINT AI Models, which is assessing the quality of AI training data and models.
“The idea here is to allow the standard to be established for accreditation so that you can decentralize the development of these models,” he said. He added, “And if there’s one thing that’s really, really important in this business, it’s having proper PID, positive identification. So, to ensure that those standards are set, when we make a model that helps us with the PID process, we find that the fastest way will be to accredit the model so that then business can be decentralized.”
Whitworth’s fourth resolution is a having his analysts mentor the models, essentially creating a “team approach” to responsible AI.
“Don’t forget, these are some of the best people we have in the world at generating PID, generating warning, generating targeting,” he said. “We need to keep them in a place of constantly training through machine learning, structured observations, data labeling, what’s responsible and a good approach to the model.”
In the past year AI models are already automating imagery collection of things in the past that NGA’s did not hold on to, he said.
“And that’s a very powerful thing,” he said.
The final resolution is GEOINT AI Assurance, which is ensuring that the nation’s adversaries are not able to manipulate the data, he said.