DARPA Tests AI Dogfighting (But Won’t Say If the Human Pilot or Computer Won)

The U.S. Air Force and Defense Advanced Research Projects Agency (DARPA) announced successful tests of a new artificial intelligence system using the experimental X-62A VISTA aircraft on Wednesday. The tests—AI dogfights that pitted the X-62A against a human-piloted F-16 aircraft—are being billed as the first “machine-learning-based autonomy in flight-critical systems.”

The Air Force notes that various autonomous systems have been in use for decades, but that machine learning tools have previously been banned due to “high risk and lack of independent control.” And given the past 100 years of warning from science fiction, it’s easy to understand why humans would be leery of AI-powered fighter jets.

The X-62A VISTA is tested with human pilots onboard who can disengage the AI, but the Air Force says its test pilots didn’t have to use their safety switches during any of the recent dogfight tests performed, which were primarily conducted in 2023.

The big unanswered question: Who “won” the simulated dogfight, the human pilot or the AI? The Air Force doesn’t say. But a video released by Edwards Air Force Base gives a peek at what the tests looked like, as you can see below.

“The X-62A is an incredible platform, not just for research and advancing the state of tests, but also for preparing the next generation of test leaders,” Col. James Valpiani, the commandant of the Test Pilot School, said in a press release Wednesday. “When ensuring the capability in front of them is safe, efficient, effective and responsible, industry can look to the results of what the X-62A ACE team has done as a paradigm shift. We’ve fundamentally changed the conversation by showing this can be executed safely and responsibly.”

The U.S. military first started experimenting with artificial intelligence systems in aircraft during the 1980s under a program called the Strategic Computing Initiative (SCI). Dubbed the Pilot’s Associate, the computer-assisted tech was imagined as a kind of invisible R2-D2 that could understand the plain language spoken by the fighter pilot.

DARPA was quite literally trying to build Skynet from 1983-1993, using advanced computers and robotic vehicles, but the tech simply wasn’t advanced enough. That could be changing very rapidly here in the 2020s thanks to tremendous advances in machine learning.

Bill Gray, the head of the U.S. Air Force Test Pilot School, emphasized dogfighting isn’t really the end goal with any of this.

“It’s very easy to look at the X-62A ACE program and see it as under autonomous control, it can dogfight, but that misses the point,” Gray said according to a press release from Edwards Air Force Base. “Dogfighting was the problem to solve so we could start testing autonomous artificial intelligence systems in the air. Every lesson we’re learning applies to every task you could give to an autonomous system.”

The vision is clearly much wider than merely one application, which you can take as either a fascinating bit of technological evolution or a terrifying glimpse at a futuristic robot uprising. Seems like a toss-up at this point.