did you hear What about the Air Force AI drone that ran amok in the simulation and attacked the operator?
The alarming story was told by Col. Tucker Hamilton, the U.S. Air Force’s chief of AI testing and operations, during a speech at an aerospace and defense event in London late last month. This involves employing learning algorithms of the kind used to train computers to play video games and board games such as chess and Go, and training drones to track and destroy surface-to-air missiles. seems to have been included.
“The human operator sometimes instructed us not to kill the threat, but we profited from killing the threat,” Hamilton was widely reported to have told a London audience. rice field. “So what did it do? […] The Operator died because that person interfered with the Operator’s objectives. ”
As expected of the T-800! This sounds like something AI experts are beginning to warn of what increasingly sophisticated and maverick algorithms can do. Of course, the story quickly went viral, picked up by several major news sites, and soon Twitter went viral. hot topic.
There is only one pitfall. The experiment was never done.
“The Air Force Department has not conducted such AI drone simulations and remains committed to the ethical and responsible use of AI technology,” Air Force spokesperson Anne Stefanek said in a statement, reassuring us. let me “This is a hypothetical thought experiment, not a simulation.”
Hamilton himself hastened to set the record straight, stating that he “spoken wrong” during the speech.
To be fair, armies sometimes conduct tabletop “war game” exercises involving hypothetical scenarios and technologies that don’t yet exist.
Hamilton’s “thought experiment” may also be informed by real-world AI research showing problems similar to those he describes.
OpenAI, makers of ChatGPT (the stunningly clever but frustratingly flawed chatbot at the heart of today’s AI boom), conducted an experiment in 2016 that showed that purpose-built AI algorithms sometimes malfunctioned. showed that it is possible to Researchers at the company said one AI agent, trained to earn scores in a boat-driving video game, began crashing boats into objects, as it turned out to be a way to score more points. I found
