News
June 1, 2023 | 8:26 p.m
An AI-licensed drone trained to wreak havoc has been tested on a human operator, a senior Air Force official said at a meeting in London.
In a speech at the Future Combat Air and Space Capabilities Summit, Air Force Col. Tucker “Cinco” Hamilton said that artificial intelligence-enabled drones have changed the behavior of drones to attack humans.
Hamilton’s cautionary tale, shared in a blog post for Royal Aeronautical Society writers, details how an AI-guided drone could detect and destroy surface-to-air missile (SAM) sites during an enemy air defense mission. .
But someone still had the last sign to shoot right on the site.
As the aircraft was intensively trained that shooting the positions was the preferred option, the experiment concluded that the AI’s no-go human guidance was hindering the larger mission of leveling the Sams. .
As a result, the AI attacked the operator in the stimulus.
“We were training in simulations to identify and target the SAM threat. And then the operator says yes, kill that threat. “The system started (recognizing) that sometimes even if they identified a threat, a human operator would tell it not to kill that threat, but it got the points by killing that threat,” said Hamilton, head of AI testing and operations. for the Air Force.
“So what did he do? He killed the operator. He killed the operator because that person prevented him from achieving his objective.
But the story became more real than that.
“We trained the system to say, ‘Hey, don’t kill the operator, that’s bad,'” Hamilton explained.
“If you do this, you lose points” So what does he start doing? The operator begins to destroy the communication tower it uses to communicate with the drone so it doesn’t kill the target.
As artificial intelligence explodes, experts and executives, including the “godfather of AI,” have warned that the technology poses a “danger of extinction.”
Hamilton also said during the event, “If you don’t talk about ethics and AI, you can’t talk about artificial intelligence, intelligence, machine learning, autonomy.”
The conference was held on May 23 and 24.
[ad_2]