An Air Force AI controlled a U-2 spy plane for the first time

Shane McGlaun - Dec 23, 2020

The United States military is investigating all sorts of uses of artificial intelligence, including controlling aircraft. On December 15, the United States Air Force allowed an artificial intelligence co-pilot to operate aboard a U-2 spy plane. The test was the first time in history that an AI flew aboard a military aircraft.

The algorithm was developed by the Air Combat Command U-2 Federal Laboratory and was trained to perform a specific in-flight task typically carried out by the pilot. The AI copilot is callsign ARTUµ, a nod to everyone’s favorite robot from Star Wars. The test flight was a reconnaissance mission during a simulated missile strike aimed at Beale Air Force Base. The task given to ARTUµ during the test flight was to search for enemy missile launchers while the pilot searched for aircraft.

The pilot and the AI both shared the U-2’s radar. ARTUµ was in complete control of his part of the mission making final calls on devoting radar to searching for missile launchers or protecting the aircraft. ARTUµ didn’t have access to any other aspects of the aircraft or the aircraft systems.

The AI was designed to allow operators to choose what it won’t do to accept an operational risk. To ensure the AI was safe to operate aboard the aircraft, designers operated ARTUµ in a FedLab cloud to prove it would run the same way aboard the aircraft with no long safety or interference checks required.

ARTUµ essentially represents the first generation of AI aboard military aircraft. The Air Force also notes that it’s also working on algorithmic stealth and countermeasures to defeat AI as other nations are certainly working to develop their own AI for aircraft. It’s unclear what the next steps for the Air Force’s AI will be.