NGA Eyes Multimodal AI in Next Phase of Geospatial Analysis

The National Geospatial-Intelligence Agency (NGA) sees multimodal artificial intelligence advancements in its future phases of processing geospatial intelligence (GEOINT) data more quickly and accurately at scale, NGA Director Vice Adm. Frank Whitworth said at DoDIIS in Omaha, Nebraska, Tuesday.

“This is a story of scale,” Whitworth said. “It’s appropriate that we’ve inherited this responsibility for GEOINT AI because GEOINT has a scale like none other.”

Whitworth noted AI’s utility in sorting through the huge amount of geospatial data and bring it to human analysts. AI collects GEOINT imagery, labels it and gives it to analysts, he said, helping the human workforce focus on analysis.

“You run as much data through a process that involves also data labels, identifying what that object is and then ultimately creating a model,” said Whitworth “When that data runs through the model, it goes to what we call running inference, which then runs into detections that we, as humans, can use.”

Advanced AI reads more complex images, Whitworth said, and multimodal imaging will transform NGA operations.

“Advancements are now accelerating … to combine text and images,” said Whitworth. “We want to ask questions about geospatial environments and get answers that are multimodal in context. … That will be the beginning. The world will be our oyster.”

Whitworth added that NGA requires accurate and well-organized data to use AI properly and to fulfill agency mission. NGA needs to have usable data for the AI models and the intelligence mission, he noted.

“It’s about the entirety of keeping your stuff organized and in the workflow to include [Geographic Information System] data, to include nav data, and with geospatial accuracy,” said Whitworth. “You can detect all you want. But if you don’t know where on Earth it is with precision and accuracy, you have a problem.”

Whitworth said AI is critical to initiatives like the Defense Department’s Project Maven, which is using AI to identify battlefield targets with satellite and geospatial intelligence. Maven automates analysis of large amounts of imagery using the technology, significantly improving the speed and accuracy of target identification for battlefield use.

“Where it has made its mark here recently is in the combination of a graphical user interface that is very agile,” said Whitworth. “A commander can say, ‘I treasure the confidence of knowing what’s out there in terms of opportunities. But I also want to have a little bit more in the targeting cycle, so that it’s applied mainly to the targeting cycle.’”

Maven has “significantly” increased the number of models it has generated, Whitworth said, since the program came under NGA’s purview to increase the ability of the agency to identify targets with more accuracy.

“The models are way up. I was asked the question when I first inherited, ‘Do you think you’ll just stick with one model?’” said Whitworth. “We have a lot of models because models can be tailored to different situations.”