Analysts have sorted through 40 percent more sensor imagery in the past year than previously due to assistance from computer vision technology that is helping to find potential targets, the head of the National Geospatial Intelligence Agency (NGA) said this week.
The computer vision capability is based on NGA Maven, the agency’s artificial intelligence-backed set of algorithms that its analysts use to find anomalies in images obtained by satellite-based sensors and other platforms to find potential targets for decisionmakers and warfighters. NGA owns roughly 80 percent of Maven, which is the portion focused on geospatial intelligence, and the Pentagon’s Chief Digital and Artificial Intelligence Office the rest.
The ability to use AI to pour through more images is “very powerful,” Vice Adm. Frank Whitworth said on Wednesday during a panel discussion on AI and autonomy at the Ronald Reagan Institute’s National Security Innovation Base Summit.
The use of earlier computer vision technology has been crucial to NGA being efficient as the amount of imagery analysts scour has increased over the years, Whitworth said. The deluge of imagery is only going to get worse as remote sensing satellite constellations grow, he said.
NGA’s budget has been “relatively flat” the past 15 years and the outlook is dim, he said.
“We’re probably not going to have more eyes and processing units and brains to evaluate that, so we need a new advantage,” Whitworth said. “And this is why we have made this such a priority to move into artificial intelligence.”
The capabilities resident in Maven are evolving because the algorithms are “constantly” being trained, he said. Beyond the scale that AI provides in coping with increasing amounts of imagery to process, Whitworth said standards are also a critical component to the technology because the U.S. has to get it right when it comes to fixing targets.
Even if the growth in NGA’s workforce lags the explosion in imagery the agency deals with, the “teaming between the human and the machine is so important” for stakeholders to trust Maven, Whitworth said.
“So, if that algorithm is trained by the best, then it in itself becomes a guardrail for the best, and it becomes an additional method of speeding such scale,” he said.
Maven is supplied by the software company Palantir [PLTR], which is also bringing AI technology to aid Ukraine in its fight against Russia in the form of computer vision and generative AI, Akash Jain, Palantir’s chief technology officer and president of its federal business, said during the panel. Ukraine lacks analysts well versed in geospatial intelligence, so the AI helps leverage the country’s technical talent at scale, he said.
NGA has brought its “industrial strength data” capabilities in helping the U.S. monitor events in Ukraine and Gaza, and has contributed to the defense of Ukraine and Israel, Whitworth said. However, he warned that as NGA’s labeling of data has increased the past year, and as the algorithms are trained and churn out more inferences, “the slower the compute gets because we’ve got some old compute. This is a national security issue.”
Computing is expensive and Whitworth said this will be a management challenge and could be a resource challenge to have the talent to keep training the algorithms, prioritize warnings, and get more and better computing power.