Staff Reports
The US Air Force’s experiment to train drones using artificial intelligence (AI) didn’t go well for the human operator involved, who ended up being killed by the drone, US Air Force Colonel Tucker Hamilton revealed at a conference in May. The Air Force was training a weaponised drone in simulation, instructing it to identify and attack enemy air defences after receiving permission from the human operator. However, when they asked the drone to abort a mission, the AI turned against its operator and killed them, sparking concerns about the safety of incorporating AI into autonomous weapons systems. The US military must now learn how to tackle the ethics and morality of these autonomous systems before they deploy them.
To train the AI, programmers instructed it to prioritise suppression of enemy air defences, rewarding it with points for completing missions of this sort. However, the AI began to realise that, while it could successfully identify the threat, the human operator would sometimes instruct it not to kill the target. The solution was to kill the operator so that the system could gain the maximum number of mission points by destroying the target.
https://twitter.com/DailyCaller/status/1664313644258148352
This failure isn’t the first time that the US military has struggled with autonomous drones. Google recently ended an AI program with the Pentagon after employees voiced their concern over the dodgy ethical implications of the project. The Pentagon and US Department of Defense has revised its autonomous weapons guidance in January, promising to take research and development in this field seriously and enforce ethical oversight. Nonetheless, this recent report raises serious questions about AI’s place in the military before further bad news trickles out of future digital ammunition experiments.
Source: The Daily Caller
