While many – if not most – Federal agencies are taking at least preliminary steps towards embracing advanced technologies like artificial intelligence (AI) and machine learning (ML), they are moving with a degree of justifiable caution when it comes to relying on those technologies as part of their cybersecurity defenses, government tech officials said today at an event presented by Fortinet.

The measured pace of AI and ML for security purposes is tied to the crucial nature of the cybersecurity mission, panelists said.

Frank Konieczny, Chief Technology Officer at the U.S. Air Force, said that ML tech in particular requires establishing baselines for access and security to be sure that adversaries are not already impacting relevant data.

And, he said, resulting data outputs from ML applications can still be somewhat opaque, and need to be examined more closely by human operators. “You ask, ‘how did it figure that one out,’” he said, adding that process results in “a lot of time … by a human looking at [the result] again, which we don’t want to do.” He continued, “it’s a question of trust” about AI and ML results and how they were achieved, adding, “it’s the subtle stuff, not the obvious stuff.”

“We are in the very early stages” of AI and ML adoption, said Ryan Cote, CIO at the Department of Transportation (DoT). Speaking of those technologies in general, he said that “marketing hype is at a 10, but delivery is at a 1.”

On the security front, he said, “the best we can do is look at machine learning to improve cyber hygiene” and automate “the simpler tasks” in cybersecurity operations. “I have yet to find a product or service” in the AI and ML realms for security applications “that lets me sleep better at night,” he said.

Rick Pina, Chief Technology Officer-Public Sector at World Wide Technology and a 25-year U.S. Army veteran, said some of the disparity between the potential of AI/ML technologies and the current pace of deployment has to do with policies and architectures in place on Federal networks.

“Things you can do in a laboratory are great … but when you have to deploy in the real world, all of this AI cool stuff that you saw in Silicon Valley is not going to happen on your network” because of the different network architectures used by tech developers and government agencies, he said.

The overall problem of cybersecurity is “much larger than any of these small pinpoint efforts” that are being developed with AI and ML tech, said Konieczny.

Nevertheless, “we are going down the path” with emerging technologies, “but it takes time,” he said. Konieczny said some of the “easy stuff” has already been accomplished with robotic process automation (RPA) technology, “but when you get to cybersecurity, it’s tough.”

“We are trying to take that first step to put our toe in the pool” and launch an initial application of AI or ML tech for cybersecurity applications, Cote said. On the non-security front, the CIO said DoT has had success in using the technologies to scan the agency’s written policies with the goal of “reducing complexity in regulation.”

Asked about the “primary drivers” for their agencies’ next steps in AI/ML implementation, Cote said “the primary driver for me is privacy.” Konieczny replied, “Ours is going to be mission, and how can you drive AI into that … with extremely low latency.”

Read More About
More Topics
John Curran
John Curran
John Curran is MeriTalk's Managing Editor covering the intersection of government and technology.