WASHINGTON — To take full advantage of AI, the Army needs a comprehensive simulation of its networks, what’s called a digital twin, where it can train and test both the algorithms and the humans working with them, the two-star chief of the service’s Network Command (NETCOM) said today.
Such a detailed digital model, constantly updated with live data on how the real-world network is performing, would be difficult to build, Maj. Gen. Jacqueline Denise McPhail acknowledged at AFCEA’s annual TechNet Augusta conference. But it would have so much potential, she went on, that she challenged the contractors in the audience to build it anyway.
Chief among its many benefits, McPhail said, the digital twin would provide an ultra-realistic training ground for network operators, cyber defenders, and AI algorithms on how the network works and how to keep it working, even when under attack.
“We get about 1.2 million cyber attacks a day,” the NETCOM commander said — and they’re increasingly sophisticated, she added, far harder to detect than brute-force Distributed Denial of Service (DDOS) attacks.
“It’s no longer DDOS. Now we have to look at behavior: What behavior has changed?” she said. “[It’s] a tremendous amount of volume of data coming in, and how do we secure that data? How do we identify what’s just noise and what is a change in behavior?
“That’s where AI comes in,” McPhail said. “If we can get a digital twin of the DoDIN-A, that now enables us to be able to leverage AI in that model to understand our vulnerabilities, where our gaps are, [and] where we do have resilience.” That acronym stands for Department of Defense Information Network – Army, which means all the military networks for which the Army is responsible.
“It’s a big ask. It’s a big ask! I know it’s a large scale effort,” she said. “I think we start small and build it out.”
RELATED: Army Cyber training AI agents in cyber ‘work roles’ alongside human counterparts
McPhail is using a broader definition of digital twin here than the one enshrined in Pentagon doctrine [PDF], which says it is “a computerized representation (integrated set of models) that serves as the real-time digital counterpart of a physical object or process.” True to that definition, when defense officials and industry executives talk about digital twins, they usually mean a highly detailed simulation of a physical thing, from an aircraft to a supply chain to a virtual wounded soldier used for training medics.
The ambition is to simulate each component and their interactions so accurately that the virtual model — the twin — replicates exactly what its physical counterpart would do in any given situation. That allows users to stress-test the twin in ways that would be unsafe in the physical world, try out alternative designs and upgrades without breaking the real system in the process, and analyze potential problems as they unfold, or even predict and prevent them in advance.
McPhail is effectively arguing she can get the same kinds of advantages from a digital twin of something that’s already digital, namely the Army network. (Obviously the network runs on physical chips, fiber, routers, and so on, but most of the action worth modeling is at the software level).
“We need to figure out what’s broken, but we also need to figure out what’s not broken so we don’t break it later,” she said at one point.