Investing in Space: How the Pentagon is making use of AI with spy satellite data

Original Article:

Source: Link

CNBC's Investing in Space newsletter offers a view into the business of space exploration and privatization, delivered straight to your inbox. CNBC's Michael Sheetz reports and curates the latest news, investor updates and exclusive interviews on the most important companies reaching new heights. Sign up to receive future editions. Last week I had the opportunity to take part in venture firm Space Capital's NYC summit, which gathered investors and portfolio companies and included a number of panels on key industry topics such as Starship and China. The conversation I moderated was on "Big (Geospatial) Data & AI," with the goal of exploring how the two rapidly evolving worlds of satellite data collection and artificial intelligence interact. I was joined by Nathan Kundtz, formerly of satellite antenna company Kymeta and now leading a synthetic data startup called Rendered, and Rachael Martin, the Maven Office Director at the National Geospatial-Intelligence Agency. It's rare to hear from someone like Martin, who's deep within the intelligence community and has a front row seat on the intersection of classified information and cutting-edge technology. Martin leads the Department of Defense's flagship AI program, Project Maven, from within the NGA, which is effectively a sibling agency to the National Reconnaissance Office. Simply put, Project Maven at NGA is working on how AI can use satellite imagery and data to detect objects and activities around the world. Or, in Martin's words, the NRO will "launch them and we tell them where to go." The introduction of AI into the satellite data realm is one of necessity, Martin emphasized, because "we have billions of geospatially-referenced data points," so "how do we understand them in a way that can provide value?" "We're in a position where you have satellites everywhere but there's so many of them that you still don't know what's going on everywhere, and you couldn't possibly look at all that data and understand it in a helpful way," Martin said. "As the volume of that data grows, it is beyond the capacity of the human mind to be able to derive any kind of useful understanding from that kind of data," Martin said. What's more, "There are many different kinds of geospatial data and you're not necessarily going to use the same kinds of AI techniques to derive value" from each of them, she added. One of the major changes Martin has seen in recent years is that more and more companies in the geospatial realm "want to be part of helping us evolve solutions to some of our challenges." "From a national security perspective, our adversaries are not interested in putting objects of interest where we can find them. And so in many cases, we have to use [artificial intelligence] to help us imagine what they might look like in other scenarios that would be of interest to us," Martin said. And more change is coming: The next step in the evolution of geospatial data and AI, from her view, is applying generative AI "to basically arm non-experts with the ability to expertly use geospatial data." "What we can do with some of the generative AI tools that are coming out is to create the ability for a non-expert to query complex geospatial data and get a response back far, far more quickly than if they would have just outsourced that to a data scientist," Martin said.