The National Geospatial-Intelligence Agency (NGA) has completed foundational work for its Maven program and is getting ready to begin soliciting industry for business opportunities related to the program, which largely uses an artificial intelligence capability called computer vision to sift through vast amounts of images to detect and classify targets of interest for warfighters and the intelligence community.
“You’ll start seeing as we as we look to the next few months a lot more activity from the Maven office in terms of contracting,” Rachael Martin, program manager and office director for Maven, said during an industry summit hosted by NGA.
Maven became a program of record on Nov. 7 and just prior to that milestone had its Software Capabilities Document approved by the Joint Staff (Defense Daily, Nov. 2). Those milestones were reached less than a year after the program, previously called Project Maven, transitioned to NGA from the Defense Department’s Intelligence and Security Office.
Already, the Maven office has released a pre-solicitation for data labeling and the request for proposals is expected in the second quarter of 2024, Martin said. The data labeling opportunity will be “core to Maven,” she said, but will also include requirements and needs across NGA.
Data labeling is providing raw data such as images with labels that give them context so that algorithms can learn and observe patterns and make predictions.
Martin said that her office is interested in data labeling as a service, model development as a service, test and evaluation as a service, detections as a service, and AI interface and human centered design. These will be the focus areas for the next few years, she said.
“Our big contracts that we’re looking at in the next year, those are going to be mostly focused on transitioning our core capabilities,” Martin said. “So, things like our data labeling capability, our AI infrastructure, our AI, inference platforms, just our end-to-end AI pipeline, right. So, our straight stick, here’s how we actually can do our AI development.”
NGA will host a Maven industry day on March 7, 2024 to provide more details on its needs and opportunities, she said. Expect more market research and solicitations in FY ’24, she said.
Most of Maven’s focus will be around computer vision but the office will also be looking at other AI capabilities such as generative AI and text image technologies, Martin said. Generative AI is an area of likely “experimentation,” she said.
Generative AI is the creation of new media such as text and images that can be based on a simple query using a computer. Martin said that longer term Maven is interested in querying images to get text results by combining large language and large vision models.
Maven will pursue near-term experimentation efforts through Broad Agency Announcements it has with the Army Research Lab, cooperative research and development agreements, and smaller vehicles within NGA, she said.
Martin’s boss is Mark Munsell, director of the Data and Digital Innovation Directorate. Munsell said that while computer vision gets the “preponderance” of NGA’s investment, the agency will have opportunities for large language models, natural language processing, other AI domains, and image annotations.
Munsell outlined the “trifecta” of NGA’s computer vision needs, which are speed, accuracy, and quality in its daily workflows, including human-machine teaming.
“We need detections that are of a higher quality of object identification,” he said. “We need those detections to be more right more often. We need those detections to be highly accurate in their geolocation accuracy. And we need those detections to be generated faster than they are today.”
Industry is developing multi-modal models that combine AI subsets such as computer vision and large language models that highlight future capabilities that NGA can take advantage of, Munsell said.
“But when you put those two things together, and you chain them in a certain way, and you have the language model doing work for you by parsing information, and taking a detection and doing a correlation, it generates new information, new knowledge, and we see that as a as a great source of future capability,” he said.