As Air Force adopts AI, it must also defend it, intelligence chief says

With the Air Force increasingly looking to develop algorithms for machine learning and artificial intelligence, it must also get serious about defending them from adversaries, the service’s top intelligence officer said.

While many senior officials across service have touted AI and ML as keys to a more effective, data-driven Air Force, there is currently is no organization within the Air Force that serves to defend the underlying algorithms once operational, Lt. Gen. Mary O’Brien, deputy chief of staff for intelligence, surveillance, reconnaissance and cyber effects operations, said Wednesday at the Air Force Association’s 2021 Air, Space & Cyber conference.

“With our airmen, once we do get the AI, what are we doing to defend the algorithm, to defend the training data, and to remove the uncertainty?” O’Brien said. “Because if our adversary injects uncertainty into any part of that process, we’re kind of dead in the water on what we wanted the AI to do for us.”

O’Brien referenced working with a “maverick” Maj. Ren DeHenre who came up with the idea for a Department of Defense AI red team to help the military services train against adversarial AI and wrote an article on it. DeHenre wanted to join an organization within the Air Force to defend its algorithms, but one didn’t exist. She has since moved on to the Air Force’s School of Advanced Air and Space Studies to continue her work on the subject, O’Brien said.

“There’s an assumption that once we have the AI, we develop the algorithm, we’ve had the training data, you know, it’s giving us whatever it is we want it to do, that there’s no risk, that there’s no threat,” said O’Brien.

That’s not the only assumption senior military officials make when they say they want AI, O’Brien added. Another is that they jump into acquiring an AI tool with the assumption that they have the proper data to power it, and if not, who it can look to provide it. “You’re not going to have AI without that training data to train the algorithm,” she said.

“We’ve got to make sure … that that data is easily available in the format we want it, when we need it, on the platform we need it,” O’Brien said. “That’s an assumption that’s driving, I think, a lot of decisions, that then when we run into roadblocks, it tends to be that we can’t get access to the data and we’re in these various negotiations to have access to it and what format and how we can integrate it.”

However, Deputy Secretary of Defense Kathleen Hicks’ recent Data Decrees should help with freeing up that data for use across the services, O’Brien acknowledged. “The memo has been signed, it’s come out, and I’m waiting for some teeth,” she said.