Outdated data infrastructure, inadequate metadata standards, limited application programming interface (API) access, and policy challenges are among the barriers the Department of Defense (DOD) must overcome to operationalize artificial intelligence (AI) across the department, according to Cameron Stanley, DOD’s chief digital and AI officer.
Speaking Sept. 8 at the 17th Annual Billington Cybersecurity Summit in Washington, D.C., Stanley said organizations that have successfully adopted AI continuously refine how they use the technology, identify gaps in their data, and work to address them.
“This is something that you’ll see continuously throughout the department. It’s not just the systems themselves, the technology themselves, or even the data quality itself,” Stanley said.
Stanley said DOD needs to make data more accessible, develop digitally savvy operational personnel who can push vendors to demonstrate their capabilities, and integrate AI into decision-making workflows.
Getting there, he said, will require the department to change how it approaches technology.
“The ‘way we’ve always done [things],’ is the most risky course of action,” Stanley said, adding that the department needs to “think differently” if it wants to get AI tools into the hands of warfighters faster.
Modernizing data for AI
Much of DOD’s data remains locked in aging systems and repositories that were not designed for modern data processing, which creates hurdles for accessing and using that information in AI applications, Stanley said.
“You can’t just access that data using an API, where it’s not structured the right way for modern data processing systems to actually have access to it,” Stanley said.
Metadata standards present another challenge. Stanley said the department needs clear, high-quality metadata and tags for data objects so that classification requirements can be considered in advance and machines can process the information more effectively and quickly.
Stanley said that since 2018, DOD has pursued an approach to improving access to data sources by establishing standards for how department components hold and make data accessible.
“We can actually mandate that department elements have to adhere to certain data standards when it comes to their holding data and their accessibility of that data. So, we’re looking at a number of different ways to do that,” he said.
Helping allies advance AI
Stanley also discussed DOD’s efforts to help U.S. allies and partners adopt AI. “What we’re seeing from our allies and partners is they are behind,” Stanley said. “They don’t have the resources that we do, they don’t have the experience that we do. They don’t have the scale that we do.”
The United States is working with a number of partners to better understand their AI needs, Stanley said. Part of that effort involves helping allies avoid mistakes the United States made during its own AI development.
“There’s so many instances where you can point to and say we made a mistake and we should not have gone down that route,” Stanley said. “We wasted a lot of time and money learning that part of lesson. Do not learn that lesson. Go this way instead.”
Stanley said the approach is showing results with Five Eyes partners, several NATO partners and allies, and several Asian allies.