The US Air Force is using AI-piloted stealth drones as Top Gun instructors

Even as it reassigns Adversaries to face the real adversary

To prepare its fighter pilots for war, the US Air Force sends them into mock battles with mock enemy jets. Traditionally, these “adversary” jets are flown by experienced Air Force pilots or contractors who are, more often than not, experienced former Air Force pilots.
It’s a similar system to that used by the famous US Navy “Top Gun” fighter weapons school and other fleet weapons schools, whose elite instructors and contractors fly as adversaries against Navy and Marine aviators.
Now artificial intelligence (AI) is about to join the ranks of adversary pilots. Last month, the Air Force awarded California drone-maker General Atomics $98 million to modify two MQ-20 jet-powered drones to function as prototypes for a new class of AI adversary.
It’s a big development. As drones and AI both get more sophisticated, air forces are combining them to produce autonomous warplanes. If recent trials are any indication, AI-flown jets have the potential to be much more lethal than human-flown jets.
If the Air Force is going to beat enemy drones in combat, it’s going to have to train – against similar drones. At the same time, AI adversaries could free up Air Force adversary pilots to fly combat missions, instead. 
It was evident four years ago that a new era of AI-driven aerial warfare was dawning. In 2020, the US Defense Advanced Research Projects Agency tapped an Air Force pilot – trained on the nimble Lockheed Martin F-16 – to fly simulated dogfights against a new AI pilot developed by Maryland tech firm Heron.
The mock battles took place in a flight simulator at Johns Hopkins University, also in Maryland. The F-16 pilot, who identified himself to press only by his call-sign “Banger,” got beat, badly. “The standard things we do as fighter pilots aren’t working,” Banger moaned as the simulated AI pilot maneuvered faster and more aggressively to line up gunshots at Banger’s own simulated plane. 
Then last year it was done for real up in the sky, with an AI-piloted F-16 jet taking on a human-piloted one. On that occasion the Air Force wouldn’t reveal who won. But in May, Air Force Secretary Frank Kendall rode along aboard an AI F-16 flying against a human pilot in a traditional turning dogfight. He said that honours were about even, but the human pilot was a highly experienced veteran. Kendall considered that if the AI had gone up against a pilot with less experience – like most US front-line pilots – it would have won.
The Air Force is applying the lessons from these AI air battles as it develops a new class of “loyal wingman” drones that USAF officials hope will accompany manned fighters into battle in the coming years. The wingman drones are designed to be highly autonomous. Basically, they take general orders from human controllers – go here, attack this – but maneuver on their own.
The Americans aren’t alone in developing AI-steered warplanes. The British, Japanese, Chinese, Russian and Australian armed forces, among many others, have also begun developing AI for aerial missions – as well as high-performance drones to host the quick-thinking control algorithms.
AI dogfights are coming, possibly sooner than many people think. If the MQ-20 adversaries work as intended, the Air Force could begin preparing for them. It’s worth noting that the angular MQ-20 has radar-evading stealth qualities, making it a potent enemy in a long-range aerial fight. Most manned adversary jets are older models that lack stealth. They’re poor surrogates for the latest enemy planes.
Nonetheless the MQ-20 is subsonic and was designed entirely for air-to-ground strike missions. It probably won’t be much of an opponent for human pilots in jets designed for air-to-air fighting – provided they can beat its stealth technology. 
But that initial contract is for just two MQ-20s, and they are only the prototypes. Two AI adversaries wouldn’t be nearly enough, of course. If the scale of the Air Force’s current adversary enterprise is any indication, the service might need scores of the mock enemies: and some would need to be high-performance supersonic airframes, probably with other advanced technologies.
A fleet of AI adversaries could render human adversaries redundant. In addition to hiring private contractors flying ex-military planes, the Air Force maintains two adversary squadrons in Nevada – one each flying F-16s and Lockheed Martin F-35 stealth fighters. In real life, if not in the movies, the Navy’s Topgun school is no longer to be found in sunny beachside San Diego: it too is in Nevada to take advantage of the vast aerial tactics ranges above the high desert. It too has many pilots and many manned adversary jets.
Until this spring, the Air Force actually had three adversary units. But the 18th Aggressor Squadron, flying F-16s from Alaska, converted to a traditional fighter squadron in order to devote itself more fully to intercepting the growing number of Russian warplanes probing US defenses around the huge northern state. 
Clearly, the Air Force is desperate to squeeze more combat power out of its existing force structure. Drone adversaries could help – by relieving human pilots of their training duties, and then honing those same pilots’ skills against enemy drones.