What is on the horizon for geospatial intelligence (GEOINT)? During day two of USGIF’s inaugural GEOINTegration Summit in September, an all-female panel discussed various ways to advance the geospatial workforce and tradecraft.
AI’s Third Wave
Dr. Ilana Heintz, the lead scientist in analytics and machine intelligence at Raytheon BBN Technologies, outlined three interrelated topics for what she described as “the third wave” of AI technology—explainability, robustness, and democratization.
“When a system gives you a response to your question, it should tell you why it came up with that response,” Heintz said with regard to explainability.
But, like humans, AI systems can make mistakes, which is why explainability and robustness work hand in hand. Robustness protects your AI system by helping it detect inconsistencies or anomalies. One example, according to Heintz, is self-driving cars.
“In one example, if there is a post-it [note] on a stop sign, [the AI might] treat it as a 65 mile-per-hour speed limit. Why would the speed limit be 65 miles-per-hour next to a school?” Heintz said. “So, robustness and explainability come back to explainability because the evidence isn’t there.”
The third topic, democratization, refers to the increase of those with a basic knowledge of AI technologies and capabilities such as facial recognition, face generation, natural language processing, or automated mapping.
“We need be able to predict how [AI] might be used in ways we won’t appreciate and then make our models robust to that and explainable,” Heintz concluded.
An Interdisciplinary Tradecraft
Dr. Cordula Robinson, a senior research scientist with the Kostas Research Institute at Northeastern University, continued the AI discussion, dispelling the myth that automation equals less work.
AI technology is a specialized tool with different components and uses. Therefore, according to Robinson, geography must be taught in schools and more interdisciplinary communication is needed to improve AI and GEOINT capabilities.
“AI means more work,” Robinson said. “We need to understand photogrammetry, geodesic systems, data and coordinate systems, and projections.”
At Northeastern, according to Robinson, there is a campus-wide initiative for interdisciplinary dialogue. “It took us out of our comfort zone. It forced us to become aware of the fruit of our labors in our professions and disciplines and how to explain it and convey it and work with others,” Robinson said. “AI is here. It’s here to stay, and we have to be as intelligent and aware of it as possible. And we also have to put our best foot forward in our discipline.”
Ashley Richter, a technology architect at In-Q-Tel, discussed the interdisciplinary nature of GEOINT from an academic perspective.
“I would love to see more interdisciplinary efforts, especially concerning GEOINT,” Richter said.
GEOINT capabilities are applicable in many other areas of the education ecosystem—geology, for example—that require skill sets such as GIS, photogrammetry, and 3D geospatial imaging.
“But we’re not connecting enough to push forward all these skills sets as a composite that can be [integrated] into the future workforce,” Richter said.
An Educational Challenge
Dr. May Yuan, an Ashbel Smith Professor at the University of Texas at Dallas, focused on some of the challenges of teaching GEOINT-related courses.
GEOINT and GIScience is a combination of geography and computer science, according to Yuan. Therefore, one of the challenges universities face is whether they should teach more technical skills or geographical knowledge.
“A lot of GIS and remote sensing classes emphasize technical skills,” Yuan said. “If students just learn technical skills, they won’t have geographical knowledge to explain patterns, recognition analysis, and spatial patterns from spatial statistical analysis.”
According to Yuan, by elevating geographical knowledge, students will have a better understanding of human and physical processes, pattern generation, and what those patterns mean for the development of society and the physical environment.