Air Force acquisition boss Will Roper on Sept. 10 made the case for closer collaboration between military services on artificial intelligence, saying the Pentagon must get the groundwork right in order to successfully expand so-called “smart” information technology across the department.
“To really make AI real, you have to do a lot of work that is not the glitzy algorithms doing cool things at the edge,” Roper said during a Pentagon-run AI conference. “You have to lay in the infrastructure.”
First, the military needs to treat information technology systems from email to wireless networks as a “warfighting system,” he said. DOD has long seen IT as an afterthought, but is now ramping up its investments to get the same connectivity the private sector and everyday people have enjoyed at home for years. That includes pursuing cloud storage so personnel can access information from anywhere around the world, not only on siloed networks.
“I’m willing to predict that in future warfare, before a combatant commander counts the number of munitions or platforms that they would take into combat, they’re first going to look at their digital infrastructure and data, knowing that if they are outclassed in those two facets, they will lose against an adversary that’s faster,” he said.
Roper wants to pursue “AI as a service,” another iteration of the Air Force’s push in recent years to contract out business services and certain enabling technologies like 5G wireless networks to private industry, rather than providing those in-house.
But that still requires dollars to work. According to one market analysis, the Air Force plans to spend $500 million on AI-related projects in 2020. That’s about the same as in 2019, though DOD is broadly looking to boost spending on AI.
“In the world of IT, remember what we’re struggling with: getting it funded,” Roper said. “You can’t take a picture of IT. You can take a picture of airplanes, and ships, and tanks. We’ve got an uphill battle to make this seem as important, as flashy, as sexy, as the leading edge of warfighting. … It’s the invisible power that will allow those physical things to dominate.”
As AI becomes more ingrained in combat systems, the military needs to make sure the software’s abilities and effects are easily understood by those who use it.
The Air Force hasn’t yet succeeded at routinely pushing software updates directly to its fighters, bombers, and other assets conducting missions around the world. The service is pursuing software development environments it can use to test out code and make sure it will run smoothly when it gets to the Airmen who will use it.
“More and more things at the commercial edge are going to have safety-critical functions: flying drones, self-driving or partially self-driving vehicles,” Roper said. Humans will trust those systems with their lives, he added, and the Air Force can learn from the work of private companies to keep people safe. “Knowing the code is going to run there without crashing, that could not be a more important problem for commercial industry to solve.”
AI is not yet advanced enough to get a machine to “think” on its own to wage war, but it can come in handy for pointing out possible courses of action and crunching equations that would take a human longer.
First- and second-wave AI are “mature enough currently to benefit pretty much any activities that we’re doing in the Air Force,” said Jean-Claude Lede, an autonomy adviser to the Air Force Research Laboratory commander. “That is really imperative that we use it to accelerate and improve our decision-making broadly.”
Researchers are trying to compress the typically multiyear, multiphase test and evaluation process into a shorter, more fluid one that lets the service continually deliver updates as technology improves.
The service is rolling out resources that let Airmen with ideas for software applications like AI algorithms start with a ready-made foundation to build on. From there, they can tailor the code to their particular purposes without starting from scratch each time.
The Air Force is cleaning and opening up its legions of data to be more accessible across the Air Force, so people can plug that information in for purposes like schedule and mission planning, targeting, and trend analysis.
“We’re building [the Air Force Cognitive Engine] to be an intuitive platform that will eventually enable anyone to leverage their data to complete their tasks faster and more accurately than ever before,” Lede added of one project underway within AFRL. The lab says that venture is a “one-stop, open-source software application with various interconnected tools that enables users to develop complex AI solutions in real time.”
DOD’s Joint Artificial Intelligence Center and others across the department are experimenting with the Air Force’s PlatformOne software development environment, and the Defense Advanced Research Projects Agency is increasingly interested in tools like PlatformOne as well.
The armed forces need to stop being territorial about their tech projects and collaborate more often, a main reason why the JAIC was founded in 2018. There’s no such thing as too many AI partnerships inside the Pentagon, Roper said.
“If you’re an AI expert and you have a grand vision or a grand design that is simply not getting out of the starting blocks, then come see me,” he said.