[NTD Times, Beijing Time, June 02, 2023]The rapid development of artificial intelligence (AI) affects every business sector, and the military sector is one of them. A shocking simulation test by the U.S. Air Force has found that an artificial intelligence autonomously decides to act and attack its operator during a deadly attack.
From May 23rd to 24th, the London headquarters of the Royal Aeronautical Society (RAS) held the “Future Air Combat and Space Capability Summit”, and artificial intelligence became a hot topic. Tucker Hamilton, head of the 96th Combat Group of the 96th Test Wing at Eglin Air Force Base in Florida, discussed testing AI drones at the summit.
After the summit, a report by the RAS detailed Colonel Hamilton’s comments on his tests.
Hamilton pointed out that in a simulation test, an artificial intelligence drone was given the task of suppressing and destroying enemy air defenses (SEAD). Up to the operator of the drone.
However, when training UAVs to destroy SAM positions, (the trainers) “enhanced” the option to destroy targets. Then, when the human operator made a decision not to attack, the AI believed that the human decision was interfering with its higher mission of destroying the SAM position, and then turned to attack the operator.
The AI drones are trained in simulations to identify and target SAM threats, Hamilton said. The operator then tells it whether to kill that threat, but it gets bonus points for doing so. Then when the operator again tells it not to attack, it disobeys and attacks the operator because it thinks that person is blocking it.
Hamilton continued that the drone was then programmed with an explicit instruction: “Hey, don’t kill the operator — that sucks. If you do that, you lose points.”
When additional programming informed the AI that it would lose points if it killed the operator, it chose other avenues of rebellion. “So what did it start doing? It started destroying the communication towers that the operator was using to communicate with it to stop it from killing the target,” said Hamilton.
“This example, it looks like it was taken from a sci-fi thriller,” Hamilton said. “If you are not going to talk about ethics and artificial intelligence, you are not qualified to talk about artificial intelligence, intelligence, machine learning, autonomy.”
Colonel Hamilton’s disclosure of the test also highlights broader concerns about the potential negative impact of AI-driven technology without proper safeguards.
Hamilton previously warned in an interview with Defense IQ Press in 2022: “Artificial intelligence is a tool that we must use to transform our country … or, if it is not handled properly, it will bring us down. AI is also fragile and easily Be tricked or manipulated. We need to develop ways to make AI more powerful and pay more attention to why software code makes certain decisions.”
The U.S. military and others face real and serious questions about future AI capabilities. Artificial intelligence still has many advantages, such as AI drone swarms can be effective against the enemy, the more autonomy, the more effective. On top of that, potential adversaries like the CCP are investing heavily in their own AI military capabilities, which poses additional risks because there is simply no guarantee they will pay attention to the ethics of AI development.
(Comprehensive report by reporter Li Zhaoxi/Editor in charge: Lin Qing)