The test was set up by DARPA and it didn’t take place in the air but in a computer simulation. Two pilots were competing to see who could win an F-16 dogfight scenario but only one of the pilots, call sign Banger, was human. The other was an artificial intelligence system called Heron which learned to fly combat missions through billions of iterations of trial and error. In the end it wasn’t even close. The AI won the test five hits to zero.
The victor was an artificial intelligence-directed “pilot” developed by Heron Systems. It quickly put the lie to a statement DARPA made just one year ago, “No AI currently exists … that can outduel a human strapped into a fighter jet in a high-speed, high-G dogfight.”
Within the scope of the simulation, the AI pilot exceeded human limitations in the tournament: It was able to consistently execute accurate shots in very short timeframes; consistently push the airframe’s tolerance of the force of gravity to its maximum potential without going beyond that; and remain unaffected by the crushing pressure exerted by violent maneuvers the way a human pilot would.
All the more remarkable, Heron’s AI pilot was self-taught using deep reinforcement learning, a method in which an AI runs a combat simulation over and over again and is “rewarded” for rapidly successful behaviors and “punished” for failure. Initially, the AI agent is simply learning not to fly its aircraft into the ground. But after 4 billion iterations, Heron seems to have mastered the art of executing energy-efficient air combat maneuvers.
The human pilot, Banger, said, “The standard things we’re trained to do as a fighter pilot aren’t working.”
Much of the rest of the piece is given to reassuring readers that we are still a few years away from Skynet taking over everything. The scenario involved in this test was a classic dog-fighting in which pilots are using machine guns within a fairly close visible range. And in that scenario, the AI pilot could fire more quickly and accurately. But contrary to what you might believe from watching Top Gun, a lot of battles with modern jets happen using missiles at a range where pilots can’t even see one another’s planes. In those scenarios being a fraction of a second quicker on the trigger isn’t really decisive.
Still, all of that sounds a bit like happy talk to me. Clearly the weak link in any kind of aerial dogfight is the human being trying to make turns in the air at 400 knots and puling 7-8 g’s while trying not to pass out. In this case, neither pilot experienced those g-forces because it was a simulation, but in the real world those g-forces would have made this much, much harder on Banger and wouldn’t have impacted Heron at all.
And in this scenario both planes are limited by their designs which are intended to keep the humans inside alive. But it’s not hard to imagine a near future in which the human pilot is fighting a specially designed AI fighter that is capable of pulling 10 or 12 g’s. I think the writing is on the wall.
Here’s video of the simulations finals. As you’ll see, it wasn’t close, though Banger did seem to do best on his 5th and final run. I don’t want to diminish him. He’s probably an incredible pilot, but that’s really the point. This AI was pretty impressive.
