NGA deepens push into AI with country’s largest data-labeling effort

Two years ago, the entire world spent an estimated $800 million on data labeling: the painstaking process of annotating images and other information to train machine-learning and AI models. Now, the Pentagon’s mapping agency is prepping a data-labeling effort that will spend nearly that much all by itself.

Within weeks, the National Geospatial-Intelligence Agency plans to release a call to industry for data-annotation services—more than $700 million worth, likely over several years—to help train AI-powered computer vision models to recognize objects and understand satellite images, the agency’s chief said Friday. 

The project will be the “largest data labeling request for proposal in the U.S. government” and “represents a significant investment in computer vision, machine learning, and AI. NGA will engage with commercial counterparts to navigate the challenges posed by increasing levels of GEOINT data,” NGA director Vice Adm. Frank Whitworth told reporters Friday at a Defense Writers Group event. 

Whitworth’s agency is in charge of turning imagery, imagery intelligence, and geospatial information into useful descriptions of places and activities for military commanders.

Data labeling is key to developing machine learning models. Humans tag raw data with labels, which teaches the model how to identify future data. 

“Data labeling is the process where the human actually identifies the object and then—in a way that is understandable by the model—informs the model. So you have to actually label it in a very specific way. And it's part of AI, but it's really the essence of ML for computer vision,” Whitworth said. 

NGA aims to improve AI’s ability to glean intel from its ever-increasing flood of GEOINT data. As part of the push, Whitworth also announced Friday that the agency launched a program to check the reliability and certify large visual models. 

This pilot program, called the Accreditation of GEOINT AI models, or A-GAIM, “will expand the responsible use of GEOINT AI models and posture NGA and the GEOINT enterprise to better support the warfighter and create new intelligence insights. Accreditation will provide a standardized evaluation framework, implements risk management, promotes a responsible AI culture, enhances AI trustworthiness, accelerates AI adoption and interoperability, and recognizes high-quality AI while identifying areas for improvement,” Whitworth said. 

The NGA director called this pilot a “pathfinder” for the Defense Department and the larger intelligence community that will advance the responsible use of GEOINT AI models to ensure that the code and its outcomes stay “principled.”

“We want to make sure that no one fears that there's going to be some sort of singularity where code takes over and makes decisions. This is very much a team between humans and machines, where humans are teaching that entry level analyst, known as a model, on a regular basis, and constantly refining, evaluating itself,” Whitworth said. 

Whitworth’s agency runs the Pentagon’s secretive, flagship AI effort, called Project Maven, which was launched to use AI to filter through mountains of data for things the military needs to see and sense. The effort is now a formal program—no longer a “project”—and is being used by warfighters around the world, Whitworth said, but declined to comment on how it’s been used in ongoing operations.  

“We don't specify that much about the field, because I like to actually just say US EUCOM, US CENTCOM, US INDOPACOM. They're using it,” he said. 

The agency has recently been criticized for moving too slowly to distribute intelligence, surveillance, and reconnaissance data and imagery to combatant commands. And the Space Force and intelligence community have reportedly been in a turf war over how the agencies should split buying ISR information from commercial providers—with the Space Force arguing it needs more power to buy directly from industry. 

Whitworth denied allegations that his agency moves too slowly, calling them “unfair and unfounded,” and said users get information at the same time that NGA does, and that there’s no arbiter in the middle holding up distribution.

“My inbox would be full of emails if we were failing in this regard—and it is not. I do not get negative reports from the COCOMs on our timelines,” he said. 

Some of these tasking questions will be addressed by the “joint mission management center,” or  JMMC, Whitworth said—a hub the NGA set up a year ago to bring together satellite data across the Defense Department, intelligence community and allies to ensure better collaboration between government agencies. 

Whitworth said the center achieved initial operating capability at the end of May. “We actually have people inside of NGA who do this as a matter of their day to day work, but now we have literally seats available for the combatant commands, the services, especially U.S. Space Force, and they are there, and they're collaborating with us.”