DOD AI Priorities Include Data Standards, Workforce Training

As the Defense Department’s new Chief Digital and AI Office ramps up operations, upskilling the workforce and defining frameworks for new artificial intelligence (AI) capabilities are top priorities for data leaders across DOD.

The Air Force, for example, is in the process of building a framework to enable AI and machine learning at the operational level. Historically, the Air Force has implemented emerging technologies on the business side for functions such as personnel and financial management before slowly implementing on the operations side.

Now, the service wants to accelerate AI innovation for operations in theater.

“Our strategy includes following that same progression using the guardrails associated with that framework we’re putting in at the enterprise level,” said Air Force Deputy CIO Winston Beauchamp at an AFCEA DC luncheon Wednesday. “We’ve gotten in our feedback with industry, let’s not put a heavy governance hand on this because you tend to limit your opportunities for innovation if you require every new innovation to go through a board for approval before you move forward.”

Beauchamp said the Air Force’s goal is not fully autonomous weapons systems, but accelerated decision-making, which aligns with the goals of DOD’s Joint All-Domain Command-and-Control (JADC2) plan.

“We are going to look at embedding portions of AI into many of our operational systems to aid our folks in making decisions,” he said. “Man-in-the-loop decision-making aided by the best information available to them, which often includes using AI as a data and analytics source.”

Beauchamp also highlighted cybersecurity as a “very promising area” for AI to make a difference.

Before DOD services and components can truly take advantage of AI, they need to upskill their workforce, said CDAO Data Scientist Lead James Doswell.

“As we are seeing more people come into the platform and try and build these AI/ML models, agencies are sending data to us and the data you receive one day, the next day something is completely off,” he said at the Wednesday event. “It goes through a lot of different transformations. What we’re trying to do is be able to perform monitoring and learning and perform statistical analysis on that data, learning on top of data as we receive it, then be able to detail-document exactly what’s happening. The amount of time it takes to do that, is there any amount of data drift as it comes in that will affect your AI model? So being able to understand the full lifecycle of the data before it even gets to the AI model [is important].”

DOD also needs to clean up its data because much of it isn’t properly labeled or tagged, he added, or is “in various formats that makes it difficult to work with quickly.”

DOD also needs to define common data standards and modernize its business systems so they can handle AI and machine learning (ML) capabilities.

“We need to find a way to use the power of ML and AI to accelerate our ability to plan for and execute operations [for JADC2] very quickly,” Beauchamp said. “That means communicating very quickly between platforms. All of these things are very complex and time sensitive. Having the ability to communicate in that way requires seamless integration across networks and services. This is not something you’re able to do with separate networks, you need high performing networks and seamless interfaces.”