The Secretary of the US Air Force, Frank Kendall, recently announced that he would soon be flying aboard an F-16 piloted by artificial intelligence [AI]. “There will be a pilot who will simply observe, like me, the operation of this technology,” he said during a Senate hearing. “I hope that neither he nor I will find it necessary to pilot the plane,” he added.
Mr. Kendall was referring to the X-62A VISTA (Variable In-flight Simulation Test Aircraft), an F-16 from the US Air Force Test Pilot School that incorporates artificial intelligence algorithms from the AACO (Autonomous Air Combat Operations) and ACE (Air Combat Evolution) programs, respectively, run by the Air Force Research Laboratory (AFRL) and DARPA, the Pentagon’s innovation-focused agency.
In detail, it aims to develop AI for beyond-visual-range air combat (BVR). It received significant attention in 2020 during the “AlphaDogfight” competition, where an AI algorithm from Heron Systems defeated an experienced F-16 pilot without his usual reference points after five simulated aerial combats.
“Whether a human or a machine wins the final aerial combat matters little since the AlphaDogfight trials aim to increase confidence in artificial intelligence,” the ACE program manager said. “But if an artificial intelligence earns the respect of an F-16 pilot, we will have taken another step towards achieving an effective man-machine interface for aerial combat, which is our goal,” he estimated.
Regardless, in 2022, the X-62A Vista conducted twelve test flights with the “AACO” and “ACE” algorithms at the controls. At the time, the US Air Force explained that during these flights, the aircraft carried out “individual engagements beyond visual range against a simulated adversary” while the “AI agents” engaged in aerial combat.
The next step was to measure the AI’s capabilities against an F-16 piloted under real conditions. DARPA recently released a statement confirming this.
Thus, between December 2022 and September 2023, in the vicinity of Edwards Air Force Base in California, the X-62A conducted 21 flights during which it faced an F-16 in visual range aerial combat or “dogfight,” which is perhaps the most perilous form of engagement for a combat pilot. It is undoubtedly even more so when the opponent is an AI.
It is worth noting, however, that two pilots were on board the X-62A during these flights to monitor the systems and, if necessary, take control.
At the event, the X-62 VISTA was pitted against another F-16 piloted by a human. The air combat test—close-range aerial combat—initially began with defensive maneuvers, with the primary focus on safety.
Then the complexity of offensive maneuvers increased, culminating in high-speed interceptions, “when the aircraft were within 2,000 feet moving at a speed of 1,900 miles per hour,” explains DARPA.
Colonel James Valpiani, US Air Force Test Pilot School commander, explained that the X-62A began by flying in a defensive posture before transitioning to more offensive maneuvers. “The aircraft were flying aggressively, within 600 meters of each other,” he said.
“The potential for autonomous air-to-air combat has been imaginable for decades, but reality has remained a distant dream. In 2023, the X-62A overcame one of the most significant barriers in combat aviation. This is a transformative moment made possible by the revolutionary achievements of the X-62A ACE team,” said Air Force Secretary Kendall.
Which of the two performed better? DARPA did not wish to disclose this information. “The goal of the test was to demonstrate that we can safely test these AI agents in a critical air combat environment for security,” stated Lieutenant Colonel Ryan Hefron, the ACE program manager.
Although aerial combat was an important and complex test, it is not yet the program’s ultimate goal. ACE’s success paves the way for broader applications of AI in various aerospace sectors, including commercial and defense operations.
“It’s very easy to look at the X-62A ACE program and see that it is under autonomous control. It can fight, but that’s not the main thing. Aerial combat was a problem that needed to be solved so that we could begin testing autonomous artificial intelligence systems in the air. Every lesson we learn is applicable to any task you can assign to an autonomous system,” said Bill Grey, the school’s chief test pilot.