The National Geospatial Intelligence Agency (NGA) is set to issue a significant call to industry this month, seeking over $700 million to train AI-driven computer vision systems, particularly for its Maven program. According to NGA Director Vice Adm. Frank Whitworth, this will mark “the largest data labeling request for proposal in the US government,” highlighting the agency’s dedication to advancing computer vision, machine learning, and artificial intelligence (AI).
NGA collects imagery from various satellites and aircraft, processes it, and then disseminates geospatial intelligence (GEOINT) products, including 3D maps, to users across the federal government. The influx of GEOINT data from an increasing number of satellite and sensor platforms poses challenges for the Intelligence Community (IC). While AI systems can expedite the analysis process, these visual models must first be trained to recognize military targets accurately and identify unusual activities.
A pressing concern is ensuring that AI systems can effectively differentiate between friends and foes, as well as combatants and civilians. This capability is crucial, not only for operational success but also to comply with international laws of war as outlined in the Geneva Conventions. Whitworth stated, “There’s this issue of distinction… and that’s hard. Based on my 35-plus years of experience, especially in targeting, that’s one of the hardest things we do in targeting.”
He further explained that effective warning systems rely on establishing a baseline for behavior or object locations, noting anomalies when something new or troubling arises. Identifying these anomalies involves a detailed process of data labeling, which is essential for enabling the Maven program. Launched by the Defense Department in 2017 to accelerate AI adoption in the military, Maven was transferred to the NGA in 2022 and officially recognized as a program of record last November.
Whitworth elaborated on data labeling, stating, “Data labeling is the process where the human actually identifies the object and informs the model in a way that is understandable.” This task can be reminiscent of online tests where users identify specific patterns or objects, requiring substantial human effort to ensure the accuracy and reliability of each labeled piece of data.
In addition to the data labeling initiative, NGA has launched the Accreditation of GEOINT AI Models (A-GAIM) pilot project aimed at certifying what Whitworth referred to as „large visual models.” This pilot seeks to evaluate the methodologies and testing procedures of AI model development, with the overarching goal of enhancing the responsible use of AI in GEOINT and providing better support to warfighters.
The pilot is designed to be a model for the Department of Defense (DoD) and the broader Intelligence Community (IC). Whitworth noted that the project aims to align with existing DoD standards on ethical AI, and the agency has implemented a training program called “GEOINT Responsible AI Training” (GREAT) for developers. This program aims to address AI life cycle challenges and is expected to be made widely accessible to coders within the national geospatial intelligence community.
“Everyone taking the certification will be asked to sign a final pledge to develop or use AI responsibly,” Whitworth concluded, emphasizing the NGA’s commitment to fostering a responsible AI culture while enhancing the effectiveness of geospatial intelligence efforts.
