Officials from the United States Air Force command said that on Tuesday, during a training flight, the AI was entrusted with the control of the navigation systems of the U-2 spy plane.
Senior Pentagon officials, commenting on the flight, said that it was a watershed moment in attempts to introduce AI into the control of combat aircraft.
The test flight lasted two hours during normal flight training at Air Force Base near Marysville, California. Officials and the pilot himself did not provide details of the test program. It is only known that the AI was responsible for collecting information from the onboard radar sensors and tactical navigation.
US Assistant Secretary of the Air Force Will Roper explained that the AI is aimed at finding approaching missiles and ground-based missile launchers. During the test flight, the AI made the final decision about where to point the onboard sensors.
The main goal of the Air Force command is to implement the joint work of man and machine, when robots are responsible for solving simple technical problems, and pilots take on critical functions: the use of weapons, piloting and target designation.
The AI itself, dubbed Artuμ (a clear reference to the Star Wars series), is based on open source algorithms and adapted to the computer systems of the U-2 aircraft.
We are talking about the publicly available µZero algorithm, developed by research company DeepMind for the rapid development of strategy games, powered by the public, developed by Google Kubernetes system. It makes it possible to transfer AI software between the on-board computer systems of the aircraft and the cloud systems for which it was developed.