GEOINT Grapples with AI Implications, Policy Directives 

Artificial Intelligence is roiling every industry, but its implications for GEOINT are especially striking. After all, AI’s value is predicated in part on its ability to derive useful insights after parsing massive amounts of raw data—a capability with plenty of GEOINT overlap.

Accordingly, GEOINT and government leaders, along with their industry partners, are rushing to figure how best to engage with AI, as well as how to govern its use.

Last fall, the White House released an executive order that adds government oversight to the application of AI—especially its use within the government, which covers a host of GEOINT activities. The order, issued on October 30, constitutes the most substantial U.S. effort at AI regulation to date, outlining a framework both for promoting AI innovation and application and for guarding against risks and misuse. It was an attempt to get in front of technology that carries profound promise as well as catastrophic risk, and that is rapidly transforming both GEOINT and the world at large.

“We’re not talking about a well-established industry that’s gradually moving into a regulatory regime,” says Chris Andrews, COO and head of product at Rendered.AI and a member of the USGIF Machine Learning & Artificial Intelligence Working Group (ML&AI). “This was much more a reactionary application of regulation to rapidly moving technology that has the potential to cause some harm—and the government is trying to kind of figure it out at the same time as industry.”

Those discussions are ongoing and will be addressed at a USGIF-hosted event on March 6: “Mission Focus: Generative AI and High-Performance Computing.”

For now, the executive order stands as the clearest AI guidance; While the order’s scope extends far beyond defense and GEOINT—it includes everything from consumer protections to supports for workers whose industries are facing AI-related labor disruptions—there are several provisions that directly affect government agencies who use GEOINT, as well as their industry partners.

For example, the order calls for tracking the provenance of all data, and for the labeling of synthetic or AI-generated content using digital watermarking or a similar technique. That’s a substantial hurdle for some in the GEOINT community, where use of synthetic data to train algorithms is widespread. The National Geospatial Intelligence Agency (NGA), for example, has trained hundreds of algorithms using bulk-labeled data, according to Don Widener, chief technology officer for intelligence solutions at BAE Systems, Inc., and a member of the USGIF ML&AI working group.

The executive order may require more detailed labeling, forcing some organizations “to throw out years and years of labor, potentially, because all of those old trained algorithms are now kind of on the naughty list,” Widener says.

In contrast, the order serves as validation for the U.S. Marine Corps, which has insisted on using only traceable, explainable AI even that was an uncommonly conservative approach, Widener says.

For now, how the executive order will be implemented isn’t entirely clear and it will likely remain that way for a while. For example, the watermarking component won’t harden into specific, enforceable requirements for some time. According to the order, the Secretary of Commerce has 240 days to research and report on potential watermarking and content-authentication technologies. The Commerce report then triggers an additional, yearlong development and review process before the issuance of formal guidance.

Why so slow? One reason, according to an analysis of the executive order in the MIT Technology Review, is that AI labeling and detection technologies aren’t yet ready for prime time. But there is an argument for not allowing developing policy to slow down technology innovation, instead looking for ways to have policy and technology pace each other.

“The 2023 executive order on AI places strong focus on governance, observability, and monitoring the real-world impacts of production models,” said Jim Rebesco, CEO and cofounder of Striveworks, a machine learning operations company that recently joined USGIF as a Small Business Partner. “We as a community have learned, sharply, that rapid innovation and retraining of models is critical to keeping pace with an ever-changing and often adversarial world. Looking back, one reason the geospatial intelligence community was an early experimenter and adopter of AI/ML was because it was obvious, to anyone who looked, that the sheer volumes of data being collected could not be processed by human eyes alone – we needed to augment and enhance analysts. In direct analogy, if implementation of the executive order is treated solely as a matter of policy, we risk having the pace of innovation slow to a crawl. I think this is a tremendous opportunity for the community to adopt the tooling and automation for governance that both enhances safety but also accelerates innovation.”

Other components of the executive order are more straightforward and will have more immediate effects. For example, every government agency was mandated to designate a Chief AI Officer by the end of 2023 in addition to a broad hiring push to add AI professionals across government.

Creating a dedicated position to lead AI efforts in agencies is a good thing, according to Barry Tilton, technology evangelist at Maxar and a member of the USGIF ML&AI working group. “If the folks [working as Chief AI Officers] have some backing in computer science, cognitive analysis, psychology and all the other pieces that make up an AI room, and they talk to each other, then we can probably navigate some of those problems more effectively than we would otherwise.”

The executive order also landed during a period in which AI products are being developed at a furious pace, and government customers are looking for new applications. One industry concern is that the order will complicate the market.

“We have proposals in flight,” says Rendered.AI’s Andrews. All those potential deals now must “be reexamined in the light of the executive order if they have anything to do with artificial intelligence, synthetic data, computer vision content, or anything like that.”

That’s an inconvenience. It’s also a welcome one for government procurement staff who have been making AI deals without the benefit of knowing the rules of the game long-term.

“If you talk to AI developers or rapid technologists, it’s like, ‘Man, this is going to slow everything down,’” says BAE’s Widener. “But within the full lifecycle and the full chain of acquisition, there’s individuals within that chain that really needed some guidance. For everyone in the Defense Industrial Base, at least now we know how the game is played. Now we can follow the standard.”

To register for the March 6 Mission Focus: Generative AI and High Performance Computing event sponsored by AWS, please visit https://usgif.org/mission-focus-events/.