Artificial Intelligence

NGA to launch $700M data labeling competition

NGA is also launching a new initiative to set accreditation standards for any AI models that supports the agency's geospatial intelligence analysis.

Justin Doubleday@jdoubledayWFED

September 3, 2024 12:02 pm

1 min read

The National Geospatial-Intelligence Agency plans to release a major request for proposals for data labeling services this month.

NGA Director Vice Adm. Frank Whitworth said his agency’s data labeling RFP would be the largest effort across the U.S. government, with an expected ceiling of more than $700 million. The goal is to help NGA categorize and label geospatial intelligence data, such as satellite imagery, so the agency scales its use of artificial intelligence.

“This represents a significant investment in computer vision, machine learning, and AI,” Whitworth said. “NGA will engage with commercial counterparts to navigate the challenges posed by increasing levels of GEOINT data, together, we will ensure delivery of timely, relevant and AI-enabled GEOINT to our customers, partners and allies.

NGA is also starting a new initiative to set standards for AI models. The pilot program is called Accreditations for GEOINT AI Models, or AGAIM.

“The idea here, just to simplify it, is to ensure that anyone who touches our code, that addresses GEOINT AI, understands that we’re staying principled,” Whitworth said during an Aug. 30 Defense Writers Group breakfast in Washington.

As a combat support agency for the Defense Department, NGA’s analysis often supports military intelligence and targeting. Whitworth said setting accreditation standards will help ensure NGA is following the laws of armed conflict in its use of AI, especially those that require distinction between military and civilian objects.

“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,” Whitworth said.

NGA’s approach will feature evaluating models based on their level of maturity, Whitworth said.

“We’re not addressing autonomy necessarily,” he said. “We’re addressing just the quality of the model.”