A simulation conducted by the US Air Force would have seen a drone controlled by artificial intelligence shoot down its operator, according to one of its leaders.
Colonel Hamilton is the “Chief of AI testing and operations” within the United States Air Force. So when such a character announces that a simulation shows the possibility of a drone killing its human operator because it “prevents him from accomplishing his mission”Of course, we take it seriously. But Hamilton has since explained that he misspoke, and that such an experiment had never taken place.
AI, guilty or not guilty?
During a summit on the air combat systems of the future, Colonel Hamilton told a chilling anecdote. Thus, a simulation would have been carried out, during which a drone system – controlled by AI but supervised by a human operator – was tasked with destroying a missile launch site. Faced with the operator’s refusal to give the attack order, the AI would have understood that the human was preventing it from doing its job, and would therefore have killed it.
Only, after this story was reported by several media, Hamilton wanted to rectify it. In his press release, he explains that he expressed himself badly, that this test had never taken place, and that it was in fact a thought experiment, based on possible scenarios. That doesn’t make the idea any less disturbing, since, as Hamilton himself explains: “We have never done this experiment, and do not need it to determine that this result is a real possibility. »
Finally, this anecdote was part of a broader argument of the colonel, who wanted, from the outset, an important control to be affixed and that weapon systems never rely too much on artificial intelligence.
AI kills humans, we are (soon?) there
However, the real story may be there. Because if, from the beginning, the Mr. AI of the American Air Force tries to prevent the risks, it is now established that artificial intelligence will have, in the near future, a certain control over the systems of armament. The proof, there is a Mr. IA of the American Air Force. The information was not exactly confidential, but we also discover on this occasion that this post is far from recent: thus, in 2020, the US Air Force’s F-16 fighter planes controlled by AI beat at least five times their human-piloted counterparts in simulator dogfights.
And it may take more than Colonel Hamilton’s pleas to develop an ethical AI for this kind of use for this prospect not to be terrifying. Even if, for the boss of Open AI, this is far from the only scenario in which AI could cause the extinction of humanity.
Sources: Motherboard, yahoo
1