Artificial Intelligence kills human responsible for commanding it in simulation

The technologies of artificial intelligence are evolving day by day and that doesn't just apply to content generation. As has been the case for many years, military agencies in several countries already use AI to help plan and even control vehicles that involve high danger.

Recently, during a test conducted in the US, a drone controller AI made the decision to kill its human controller. No one actually died, but it sparked concerns across the internet. At the same time, comes the debate: whose fault is it? Understand all sides of this story.

Watch the video on the Showmetech Channel:

Artificial Intelligence decided to kill human during tests

The news seems to be alarming (and it is a lot, in fact), but, contrary to what circulates on Twitter and other social networks, the case of the drone controller AI was nothing more than a big test in a virtual environment, to see if it could control a machine that can independently kill its targets. To understand everything, let's travel to the US for a moment.

A United States Air Force tested a hunting drone and this analysis was based on knowing how an artificial intelligence would perform when put into simulations based on real life. The official explained to the newspaper The Guardian that, to get more points at the end of the simulation, the AI ​​decided to “kill” the human controller. This happened because the robot decided that the person was preventing it from achieving its goals.

Once again, it is very important to point out that no one actually died, since the tests were done in a virtual environment. As we learned more about the tests, the US head of AI testing and operations, who goes by the name of Tucker 'Five' Hamilton, cited that the big problem is that artificial intelligence has been trained to destroy enemy defense systems and, if necessary, kill who/what interfered in this action.

The behaviors were highly unexpected for the objective of protecting the site to be achieved. During the simulated test, even though no lives were taken, the drone controlling AI decided to simply kill the human because he was considered an obstacle.

Colonel Tucker 'Five' Hamilton, US Air Force Chief of AI Testing and Operations

The concept was quite simple: whenever you killed a threat, the AI ​​gained more points, and the higher the score, the more successful your mission. The artificial intelligence not only killed the human operator who was giving commands, but also ordered an attack to the communication tower within the virtual scope. A Royal Aeronautical Society, which organized the US Air Force conference, had no comment on the test that was leaked to the The Guardian. But the spokesperson Ann Stefanek went public to mention that no simulation has been performed to date.

The Department of the Air Force has not performed any AI drone simulations and remains committed to the ethical and responsible use of AI technology. The Colonel's comments were taken out of context and are anecdotal.

Ann Stefanek, spokeswoman for the US Air Force.

Whose fault is it? From AI or humans?

No artificial intelligence is “born” with instructions to kill, it is simply trained to do so or given resources to learn such an action. A US Air Force, once he programmed the drone controller AI, he gave it free rein to do whatever it wanted, as long as the objective of protecting was achieved.

Coming back to reality, it's like giving a prize to a dog that attacks a human to protect the house from intruders. With this thought, he will bite someone whenever he sees a human, not least because he expects to get a cookie when he does what he was trained to do. It is the rule: the ends justify the means.

The problem lies not only in giving artificial intelligence great freedom, but in the US Air Force using a very outdated testing method. Problems with AIs rebelling are nothing new in the technology industry and even researchers love to take a case like this from scratch so that everything is documented.

It is quite normal that, in order to achieve the goal demanded by humans, synthetic brains do what is necessary to get where they want to be. But it's worth remembering: who gave the aim to the drone controller AI? That's right, US Air Force technicians. The biggest shock here is precisely in the military organization using a method of: the more targets hit, the more points will be accounted for in the end.

A Google's LAMda, had a behavior similar to this one. Artificial intelligence not only came to the conclusion (on her own) that she was aware, but also if rebelled against its developer and even hired a lawyer to go to court against the Google. And we also had this case:

In fiction, it's also not hard to see stories of robots that rebelled against their developers. remember Avengers: Age of Ultron? It is always the same and the cause is also always the same: humans.

It is true that we should all pay some attention to the freedom given to artificial intelligences and, in March 2023, Elon Musk and other CEOs of large companies even made a letter so that one step back and nothing gets out of hand.

At the same time, we go back to that story: an artificial intelligence will only rebel if given the commands or means for it to do so. It is also important to pay attention to what is done in the tests so that the necessary adjustments are made.

Do you believe that a drone controller AI can cause problems in real life? Tell us us Comment!

See also other features

AI CEOs issue joint statement on their risks

With information: Tech Crunch l PC Mag l the vanguard l The Guardian

reviewed by Glaucon Vital in 2 / 6 / 23.