Roper Aims for 50 Percent Failure in Next ABMS Experiment

The next Advanced Battle Management System experiment will involve far more participants at more diverse locations than the previous run, but Air Force acquisition chief Will Roper wants to take more risks this time in order to find more flaws to fix.

“The event we have coming up April 8 is going to be massive,” Roper told reporters in a Pentagon press conference. The next experiment—which Roper calls an “on-ramp” to a future ABMS—will involve physically shooting down both an unmanned aircraft and a cruise missile, employing ships, submarines, ground troops, aircraft, and SpaceX Starlink satellites, he said. There are also “discussions” about including Virgin’s Launcher One to evaluate the effects of having the ability to launch a satellite on-demand.

Participating locations will include Eglin and Nellis Air Force Bases, Marine Corps Air Station Yuma, and the Army’s White Sands Missile Range, among other locations, he reported. “Space superiority” also will play a role. There will be no allied platforms in the next itertion, but there will be “international participation” and allied platforms will be included in later events, Roper noted.

“The only failure” in the previous experiment, run in December, was that “we had way too many successes,” Roper asserted, adding that in this run, he wants “an equal measure of things that fail for things that succeed.” The roughly 80-90 percent success of the last event meant not enough was learned from it, he said.

“The point of this acquisition is to inject learning continually into the program,” Roper asserted.  It must be conducted differently than previous major weapon system acquisitions because “if we run this as a major defense acquisition program, we’re already doomed to fail.”
Technology is progressing far too rapidly to approach the program with anything less than breakneck speed, he said. “It’s too complicated to not have continual learning put in.” The ABMS experimentation will reflect the same approach being taken at USAF software “factories” like Kessel Run and Kobayashi Maru, he asserted: “Fast iterations between the developer and the operator. Keep getting feedback, keep learning, keep burning down risk quickly.”

There was no international partnership on the first iteration because if it was a “colossal failure,” the Air Force wanted it “to be all on our dime.” But because it “worked really well,” industry and allies are asking to participate in future experiments

Successful internet companies have told Roper the best approach is to go for small “slices” of improvement, of 10-15 percent, “which you write down,” he said, “because we usually say, 80-90 percent solutions” for Pentagon projects. “You want the minimum viable capability that teaches you something” to be quickly fielded.  

It will be “really cool” for ABMS “when we start going from demonstrating to competition and source selection,” Roper asserted, because the competition will in no way be a traditional fly-off where functionality “at a particular point in time” is measured “and we pick.” Rather, “we will actually force vendors that are competing to make significant changes” between four-month cycles, compelling them to demonstrate flexible updating. This approach changes source selection because it “puts a big spotlight incentive on upgradeability and adaptability. Which, hopefully, will get us the open architectures and things we need to make” the system work.

The Analysis of Alternatives process for ABMS has largely been discarded, Roper said.

“We’ve come so far in ABMS that we realize that it’s bigger than just replacing the capability that JSTARS provides,” he said. “If you get ABMS right, you’ve just built the military’s ‘internet of things.’ That’s super exciting.” He added that no one wrote a requirement for the internet, it simply evolved. “It translated and migrated and updated, and that’s the idea with this.” Eventually, and with help from Artificial Intelligence, the ABMS system will learn from its users what information they need, and it will push that information to them; a capability available to anyone with a smart phone, but “no one has that in the military.”

The Air Force organizations running ABMS are also putting forth a “design philosophy” rather than a set of requirements, because it has to be rapidly upgradeable, Roper said.  

He also noted that Artificial Intelligence is going to need a lot more attention relative to how it enables ABMS. He envisions “AI/Counter-AI” much in the same way as the stealth/counter-stealth “chess game” played out over the last few decades. 
“The first thing we need to do is get AI in the fleet, which we’re working, and the next thing we’ll figure out is, how do we break it,” Roper said. Figuring that out will “seed” the next generation of AI research, which will target not software code, but “the intelligence or the data set.” Such “adversarial AI” is not yet a big area of research, Roper said, “and we need to make it one.”

It will be “really exciting to see how ABMS progresses this year,” Roper observed. “I hope for the 50-50 split between success and failure … If we have lots of failures, we know what to work on for the next tranche.”