The US Air Force has denied conducting an experiment in which an AI-enabled drone attacked its operator after considering them a hindrance in completing the mission.
The clarification has come after a presentation by a US Air Force colonel at the Royal Aeronautical Society’s conference in May went viral.
“The Department of the Air Force has not conducted any such AI-drone simulations and remains committed to the ethical and responsible use of AI technology,” Ann Stefanek, an Air Force spokesperson, said in a statement.
Col Tucker Hamilton, chief of AI test and operations in the US Air Force, described a simulation exercise in which the AI drone was tasked with a suppression of enemy air defence or SEAD mission to identify and destroy surface-to-air missile sites.
The drone was trained to the notion that destruction of the sites was the “preferred option” but final clearance to attack was to be given by a human.
In cases when the drone identified a target but its operator ordered not to strike, it killed the operator as they were stopping it from accomplishing the objective, the colonel said according to a blog.
When the drone was trained to not kill the operator, it started destroying the communication tower used by the operator to relay commands.
