Geospatial artificial intelligence (AI) is needed to support and expand intelligence professionals’ depth and coverage. Built on a modern software environment and data management framework, AI is a necessary force-multiplier for advanced analytics and modeling capabilities.
Machine-assisted technologies, such as computer vision, endeavor to automatically detect, extract, and attribute features and objects in imagery and video.
While conventional approaches have had success, the National Geospatial-Intelligence Agency (NGA) is seeking novel, innovative approaches to increase AI capabilities in computer vision on existing geospatial datasets. AI is one of the agency’s 2020 Technology Focus Areas. A panel of AI experts discussed NGA’s AI journey and where they are heading, during USGIF’s GEOConnect Series Virtual Main Stage.
Artificial Intelligence, Augmentation and Automation (AAA) defines the tools and strategies that are able to make use of all of the information available in order to anticipate opportunities and threats. NGA’s AAA initiative is a major factor in driving the future of geospatial technology and tradecraft.
“Early on in our attempts to provide mission applications attributable to AAA technology, it became really apparent to us … that a AAA-enabling ecosystem was required of us to achieve the gains we’re looking for in this field,” said Matt Zezima, who works on the AAA Cross-Functional Team at NGA.
NGA’s AAA-enabling ecosystem includes four enabling arms across the agency—mission, policy, technology, and adoption. But according to Zezima, there is a constant duplication and reformatting effort, recollecting of unstructured data, and repeated chasing of capabilities that has hindered NGA’s ability to push adoption across the agency.
“We’ve got to work beyond our endless run-of-pilot and move more into promising capabilities, into operations and do it faster. This is something that we’re focusing on,” Zezima said. “We have five tasks going on right now with an analysis and source director this year. And I think we’re really going to focus on capabilities and integration in order to foster adoption.”
Challenges and Research
NGA’s needs for high confidence in AI and understanding of the generated output under varying conditions are supported by the larger Department of Defense (DoD) and intelligence communities. But according to Kenneth Rice, deputy director of research at NGA, accuracy is critical to their mission and becomes a challenge when dealing with AI.
“For the missions in the DoD, the accuracy we get out of our algorithms is critical and can literally mean life or death in some cases,” said Rice. “So, we really have to be certain of our results.”
NGA’s GEOINT data is diverse, and this diversity manifests in different ways from its different data sources, phenomenology, and data content. There is variability in data quality and the contextual data such as metadata with its obtained data sources. The agency is exploring the impact of this data diversity to AI capabilities and have performed extensive testing evaluation methods to examine new capabilities. But this is challenging to do well at scale and difficult to implement in a way that ensures reliable performance against NGA’s evolving mission.
The agency’s AI research strives to develop a better understanding of these data characteristics and the development process, which could lead to better characterization and parameters of AI and develop a better understanding of the data and of AI development processes.
“The central impact of AI is indeed significant, and we’re witnessing various exciting applications of AI. It’s also exciting to me that we have much to learn in this area,” said Michelle Brennan, image and video pod lead for the research directorate at NGA. “And as we develop this richer understanding of this trade space between data and AI capabilities, we will be able to strategically identify the capabilities that are best suited for the GEOINT mission.”