DOD’s New AI Center Ramps Up

Jill Crisman, Joint Artificial Intelligence Center chief scientist and acting chief technical officer, discusses Defense Department AI strategy at an AFCEA TechNet Cyber event held May 16, 2019, in Baltimore, Md. Twitter photo via the Defense Information Systems Agency.

The Pentagon’s new Joint Artificial Intelligence Center is ramping up its work, adding two new areas of focus—cybersecurity and robotic process automation—and expanding its outreach to academics and foreign allies, its new chief scientist said.

“We want to help the DOD understand [cyber] attacks that are happening on our systems and networks and actually facilitate collaboration across the DOD,” said Jill Crisman, who is also acting chief technology officer for the JAIC. She spoke at the AFCEA TechNet Cyber event in Baltimore, Md., on May 16.

Robotic process automation (RPA) “isn’t strictly AI, but [it presents] a lot of the same problems” in designing and building it, she said. It has many of the same benefits, too. “If we could allow our DOD back-end office support [teams] to automate some of their tasks, you can free up people to work on more complex tasks, that require more human judgement and analysis,” she added.

RPA and cyber “sense-making” are two new national mission initiatives—the lines of effort that JAIC will focus on as it seeks to actually deploy existing artificial intelligence and machine learning technologies across the department. That effort is complementary to the DARPA research and development program looking to develop new technologies.

Crisman said she thinks of the division of labor as JAIC doing “AI now,” while DARPA focuses on what it’s dubbed “AI Next.”

National mission initiatives are complementary to component mission initiatives (CMI), like the under secretary of defense for intelligence’s controversial Project Maven, which uses AI to identify objects in video feeds from drones. JAIC will assist with CMIs but they’ll remain under the purview of their sponsoring components.

The two previously announced national mission initiatives were preventative maintenance—focused on figuring out when engine parts are likely to fail—and crunching real-time data about the spread of forest fires for disaster responders, known in military jargon as Humanitarian and Disaster Relief or HADR operations.

The preventative maintenance effort is being piloted for the H-60 helicopter fleet run by US Special Operations Command, Crisman said, but the AI technology developed there will be scaled out to the Air Force, Navy, and Army over the next several months for use in their own H-60 fleets.

Demand for the technology, she explained to Air Force Magazine after her talk, was driven by a recent Pentagon directive setting a minimum 80 percent readiness level for all platforms. Improving predictions about when parts fail and when they will need to be replaced will make the logistics supply chain more efficient as well.

Subsequently, the same technology can be tweaked for other things, such as gas turbine engines, explained Pentagon spokeswoman Elissa Smith.

The preventative maintenance NMI is also developing “AI data curation tools,” such as the Army’s Composite Learning Algorithm for Records Evaluation (CLARE), Smith added. Tools like CLARE “are designed to make it easier/faster to develop an AI solution” by revealing the relationships between different data points.

“CLARE is currently performing with better than 90 percent accuracy and has been applied to more than 40 million records,” Smith said. The JAIC, working with Naval Air Systems Command, US Army Futures Command, and the US Army Corps of Engineers Engineering Research and Development Center, “is now enhancing CLARE to provide similar capability for Navy and Air Force maintenance records.”

In cyber, JAIC is starting with a market survey of commercially available AI offerings, Smith continued. “Industry has invested heavily in AI for cybersecurity products,” she said, and there was no need to reinvent the wheel. “Unfortunately, not all military needs can be satisfied with commercial products,” she added.

The JAIC, working with other DOD components, was “establishing a campaign of commercial product evaluations that will result in rigorous evaluations” of the “fit and functionality” of existing commercial offerings. These evaluations would help components decide what to buy, but it will also help JAIC identify gaps in available commercial products, which it would seek to fill.

The NMI would initially focus on “using AI-enabled capabilities to improve network event detection, network mapping, and misused account identification …The volume, velocity, and variety of data that can be collected for network and user behaviors make these ideal use cases for AI.”

In the future, Smith said AI could be used for a range of other cybersecurity missions, including “application event detection, cyberspace data engineering, sensor orchestration, automated vulnerability patching, and social media analysis.”

The JAIC stood up last year, and Crisman, who started two months ago, said the center is still in ramp-up mode.

“JAIC is a relatively new organization. We are still standing up many of our processes,” she said.