The Pentagon Is Building an AI Product Factory

The Pentagon’s research chief is deep in discussions about the newly announced Joint Artificial Intelligence Center, or JAIC, a subject of intense speculation and intrigue since Defense Undersecretary for Research and Engineering Michael Griffin announced it last week. Griffin has been sparse in his public comments on what the center will do. But its main mission will be to listen to service requests, gather the necessary talent, and deliver AI-infused solutions, according to two observers with direct knowledge of the discussions. Little else about the center has been decided, they say.

“We are looking right now, as we speak, at things like how we structure it, who should lead it, where it should be, how we should structure our other research. These are ongoing questions we are addressing this week,” Griffin said at a House Armed Services Committee hearing.

To prove its worth to service leaders, the center’s first projects will follow the model of the Air Force’s AI-powered Project Maven. Services will bring problems that might be eased by AI — say, reducing human workload in classifying objects discovered in intelligence, surveillance, and reconnaissance data — and the center will marshal computing resources, contractors, and academics toward a solution.

“Everyone is very happy with Project Maven,” in terms of speed of delivery, quality of product, and organizational structure, according to one of the observers. “There’s an element [of the Defense Department] that’s asking, ‘How do we make a Project Maven factory?’”

The JAIC will also get something valuable out of these visits and assignments: access to endpoint users as well as access to data.

Currently, the goal is to stand center with a staff of about 200 people in about two years, said the two individuals. But both acknowledged that it usually takes longer than expected to staff a new government center. Somewhere in the future — one observer called it a “far-off” ambition — the center could develop into a major national lab on par with the Sandia National Lab for nuclear research, which boasts a staff of 10,000.

There may also be some secondary effort to explore how to use AI safely and ethically, issues at the forefront of the Silicon Valley discussion of AI. Another secondary effort may explore the foundations of much more advanced — read: less practically usable in the near term — AI.

Following the Maven example, the military will rely mostly on contractors and third parties for its AI, and the center could help. “Project Maven has a staff of about 12 (or so) and a budget of $70 million dollars. So are they reliant on outside contractors,” noted one of the individuals.

The center will include a physical building — the two individuals anticipate a lot of arguing about its location —  But there will also be a heavy university component. Little pieces of the JAIC, or rather JAIC activity, will exist or take place in universities across the country that routinely partner with the Defense Department on projects.

This is less of a surprise. The center rises out of a recommendation from the Defense Innovation Board: “This center should coordinate research in these areas across the Department, and liaise with other labs in the private sector and universities, and should also conduct educational efforts to inform the Department about the implications of these advances for the Defense enterprise.” The board is chaired by former Google chief executive Eric Schmidt, who reiterated that point in his testimony alongside Griffin on Tuesday.

“The impact of AI and ML [machine learning] will be felt in every corner of the Department’s operations, from critical tactical operations such as Intelligence, Surveillance, and Reconnaissance (ISR), targeting, cyber defense and autonomous land, air and sea vehicles; support operations such as personnel billeting, training, logistics, and threat analysis and war-gaming,” the recommendation continues. “The Board likens this situation to that which existed in the first (nuclear weapons) and second (precision munitions and stealth) offsets. Indeed, both AI and MLare key components of the Department’s Third Offset thinking.”