Video Transcript
The U.S. Air Force has successfully tested a fighter jet piloted by artificial intelligence (AI) in a dogfight against a human-piloted jet for the first time.
Although autonomous flight technology — or autopilot — has been used for decades, the tests over Edwards Air Force Base in Southern California marked the first-ever use of systems based on machine learning in “flight-critical” systems.
Teams from the Air Force Test Pilot School and the Defense Advanced Research Projects Agency went from implementing the first AI agents in the X-62A VISTA aircraft to a demonstration dogfight — officially known as a “within-visual-range engagement” — in less than a calendar year.
Air Force officials said that the tests proved that AI can be used safely in aerospace, effectively launching an entirely new era of machine learning in the industry. Air Force Secretary Frank Kendall, who plans to fly in the X-62A at Edwards to observe a simulated combat flight, called it a “transformational moment.”
Refining the AI system required 21 test flights and more than 100,000 software changes. Developers first used defensive maneuvers to establish safe flying conditions before switching to offensive engagements.
Although pilots were aboard the X-62A to disengage the AI if needed, they did not have to do so during any of the test dogfights. During the “nose-to-nose” engagements, the jets approached within 2,000 feet of each other while traveling at 1,200 miles an hour.
The teams were named finalists for the 2023 Robert J. Collier Trophy, which recognizes achievements in the U.S. aeronautics sector. Officials said the tests could eventually lead to safer, more reliable AI systems throughout the aerospace sector — in both defense and commercial applications.
