The Taylor Geospatial Institute’s GeoAI Working Group Launches with a Well-Attended Kickoff with activities scheduled throughout the year.
Geospatial experts from academia, industry, and government gathered in St. Louis to mark the start of the new TGI Working Group, which provides a forum for our collaborative, diverse, interdisciplinary community to explore geospatial and AI topics. The GeoAI Working Group seeks to accelerate the application of AI with multi-modal data and technology to address the collective challenges we all face, including national security, urban planning, agriculture, health, economies, and climate change among others.
The event hosted presentations from Mark Munsell, Director of Data and Digital Innovation and Chief Artificial Intelligence Officer at the National Geospatial-Intelligence Agency (NGA), Josh Delmonico, Executive Director of the Federal Geographic Data Committee (FGDC), and a panel moderated by Abby Stylianou, TGI Fellow and Saint Louis University computer vision researcher that included Samantha Arundel, US Geological Survey research scientist and acting director of USGS’s Center for Geospatial Information Science, and Mike Merritt, Chief Customer Officer at Planet.
Expectations in the room were high for understanding the intersection of artificial intelligence and geospatial science, technology, and data, and the potential benefits and capabilities that it can bring. “I trust that 10 years from now, the work that we’re doing here today in this room will have a profound effect on the world,” Munsell said. He added that the NGA is looking forward to several things on the GeoAI front:
- Improving computer vision models: “Fundamentally, our computer vision models still aren’t very good,” Munsell said. “When we try to detect an object from low-Earth orbit, on either commercial imagery or national technical means imagery, we’re still only getting it right maybe three quarters of the time.”
- Expanding to new modalities: “Today we mostly do electrical optical imagery, radar imagery,” he noted. “We’re not really expanding these models and this work to other types of modalities.”
- Integrating emerging technology seamlessly into analytic workflows: “We have thousands of analysts that look at imagery every day,” Munsell said. “But we’re still not in the phase where those humans are getting a lot of feedback to the models to improve them. That’s something we need to do.”
- Employing large-language models more in analyses of geospatial data: “Language is something that we know is important, we know we have to embrace it,” he said. “I think that our agency is sitting upon a goldmine of the most amount of imagery in the world associated with the most amount of text written by humans with those images.”
- Expanding current object classes: “Today it’s mostly military equipment that we’re looking for,” he suggested. “You can imagine all of the things that would be interesting to national security in terms of going off and detecting – things from energy infrastructure, foundational military intelligence infrastructure…. Super useful sets of object classes that we haven’t even begun to tackle yet.”
- Developing common operations, tools, and ontologies: “That’s really a struggle amongst the bureaucratic agencies – to try to do anything common like that,” Munsell said. “The only way that we’re going to win this war is to develop some of these common platforms, common tooling, common vocabularies.”
The presenters that followed Munsell, in conversation with the meeting’s attendees, identified several key areas where the GeoAI Working Group could focus its energies to help spur on innovation and further TGI’s vision to spark a revolution in geospatial science and technology that will benefit humanity. Some examples included using multi-modality models for AI to better understand and validate geospatial alongside other types of information, the need for humans in the loop when training, checking, and improving AI models, and the value of dual-use capabilities across civilian and military users to continue to spur on innovation.
Josh Delmonico agreed that large-language models will be essential to plugging AI into growing repositories of geospatial data. “How do we integrate geospatial data into large language models?” Delmonico asked. “Some of that large language model work is geospatially informed, not geospatially focused. As a community, we need to think about a roadmap to get us here. We have so much to do that we can’t afford to be duplicating effort across the board. We need to be focused together to achieve the vision that we all see.”
During the panel discussion geospatial experts from academia (Stylianou), industry (Merritt), and government (Arundel) to explore common themes and challenges around GeoAI. Now is the time for AI to become a bigger piece in the analysis of geospatial data, Merritt said. “I think this is the year for AI,” he noted. “Actually, this is the year that we’re really seeing breakthrough and go mainstream.”
Merritt also said that geospatial researchers broadly, and specifically targeted groups like TGI’s GeoAI Working Group could help bring AI and geospatial applications to the next level. “Our single biggest gap in the industry is around solutions,” he said. “It’s one thing to build an AI model. But what’s an AI solution to a real-world problem? That is that next step that we need to get to, because models are not enough. There needs to be a more complete solution wrapped around them that makes business sense and drives real value and adoption.” Samantha Arundel agreed: “Bringing together all of these smart people and talking about these issues is a good way to go about this,” she said.
As the TGI GeoAI Working Group now turns to the business of establishing a charter, electing chairs, and other infrastructural moves that will advance and solidify the effort. The group has already put out a call for chair nominations and held its second meeting on April 4, when space and spatial attorney Kevin Pomfret discussed the regulations, laws, and ethics orbiting AI and geospatial research and development in these exciting times [“The geospatial community did big data before big data was cool,” Pomfret said during his presentation!].
The working group plans to hold its next meeting on June 6th.
Subscribe to receive news and see upcoming events from the Taylor Geospatial Institute.