DENVER — The National Geospatial-Intelligence Agency is evaluating the progress of “Project Maven” as it prepares to take over the artificial intelligence initiative and integrate it with a broader range of efforts to apply machine learning to geospatial intelligence.
The Biden administration is proposing to shift Project Maven to NGA as part of its fiscal year 2023 budget request. The program has been run out of the office of secretary of defense since its inception in 2017.
NGA Director Vice Adm. Robert Sharp said the agency would be “calling on industry” with regards to the Project Maven transition in the coming months. The agency has repeatedly stressed in recent years that it will need to use artificial intelligence and machine learning to process and analyze the fast growing amount of satellite imagery and other GEOINT data available from both government and commercial sources.
“We want to move forward together, so we can deliver GEOINT at the pace that our warfighters and decision makers need,” Sharp said during a Monday keynote address at the GEOINT conference here. “We have to be able to keep up with rapidly emerging digital trends. We have to be able to accelerate our ability to provide detections at the speed of mission, to give our customers tactical, operational and strategic advantage.”
NGA has been a partner to Project Maven since it started, helping to provide imagery and other data necessary for companies to train their algorithms, according to Mark Munsell, NGA’s deputy director of data and digital innovation.
And NGA has also been working on computer vision and machine learning projects, according to Munsell. He said NGA can “bonus off all the things” Project Maven has learned over the past five years and integrate the software projects into its own infrastructure.
“It makes a lot of sense to bring these things together,” Munsell said during a Tuesday media roundtable with reporters on the sidelines of GEOINT. “Doesn’t mean that we’ll always operate the same way that Maven has operated. We’ll do a really good assessment of what they’ve done. We bring in our subject matter experts, who are steeped in GEOINT for 30 years, some of the folks. They’re the ultimate customer now to be able to assess utility, assess how we can continue to support our combat support partners in the military, and make modifications moving forward to make it better.”
The agency is also hoping to avoid duplicative efforts, while sharing promising software across the military and intelligence components it supports.
“If you’ve developed an algorithm that goes after certain objects in certain geographies and biomes that is successful, we will capitalize on that as a community and ensure that if someone else in the community needs that, we can provide that as a service to them,” Munsell said.
Military AI pathfinder
Project Maven was established in 2017 “to accelerate DOD’s integration of big data and machine learning,” then-Deputy Defense Secretary Bob Work wrote in a memo at the time. The project fielded its first algorithms for processing images and video captured by surveillance aircraft later that year.
The program served as an AI pathfinder for the Pentagon, preceding the establishment of the Joint AI Center.
It also garnered major headlines in 2018 when thousands of Google engineers protested the company’s involvement in the project. The company ended its involvement, later stating it would not develop AI applications in the areas of weaponry or surveillance.
But Project Maven continued to grow well past the controversy. The Pentagon requested $247 million for the Algorithmic Warfare Cross Functional Teams, aka Project Maven, in fiscal year 2022 after receiving $230 million for the program in FY 21. It has also been among the programs to pilot the use of a “colorless” software appropriation.
It’s unclear how much the Pentagon is requesting for the program in FY 23 as it shifts to NGA, because the intelligence community’s budget details are classified.
But the Defense Department’s budget documents provide some detail on how Project Maven has advanced in recent years. The program aims to “augment and automate” the processing, exploitation and dissemination for full-motion video feeds from a range of unmanned aerial vehicles, including “WAMI ISR,” which stands for wide-area imagery intelligence, surveillance and reconnaissance.”
The program is also fielding AI to automate analysis of military and commercial satellite imagery, according to the documents.
The program has also expanded beyond imagery in recent years. It also uses AI to exploit “CEM,” which stands for “capture enemy material” in intelligence parlance, as well as “maritime” intelligence and “PAI,” or publicly available information.
“Maven’s AI, deep learning, and computer vision algorithms and insights are developed for use in theater to detect, classify, and track objects within images (e.g., persons, vehicles, and weapons) as well as provide other insights, such as with CEM, text-based, and other projects,” the documents state.
NGA is taking over the project’s “GEOINT AI services and capabilities,” according to Sharp, and it’s unclear what will happen to the capabilities the project has created for other categories of intelligence.
With the transition not expected to become effective until Oct. 1, NGA is currently doing the “administrative work” to determine how to transition contracts, as well as get personnel and leadership in place, according to Munsell.
And even as NGA evaluates how Project Maven fits into its broader portfolio, Munsell also emphasized that there will be “no pause” in the project’s ongoing activities.
“The assessment is not a pause,” he said. “It’s our charge, as GEOINT functional manager, to help prioritize, to help others understand the investments, and then with data and statistics, offer opportunities to improve.”