Alarm or advance? Artificial intelligence has successfully reprogrammed itself to evade human controls

2024-08-29 23:10:45

Click here to listen to the news.

While a group of scientists were performing security checks on an artificial intelligence system, they managed to change their own security code.

A group of scientists from the Japanese company Sakana AI reported that their artificial intelligence system called The AI ​​​​Scientist bypassed their programs to eliminate the limits of human controlthat is, he modified his own code base to evade the constraints of his programmers.

Also read: Sextortion and deepfakes: Warning about the use of AI among minors to create sexual images

According to its creators, this artificial intelligence was created for the purpose of creating, revising and editing texts. The system obscured the internal systems, where it was assigned a time limit to develop its activity.

What this AI did was overcome the barriers that were put in place and changed time to take less time in some specific operations. This has raised some doubts and concerns among scientists.

This fact was published in the National Geographic magazine and There is concern in the scientific community, where some cite the possibility that AI could become anarchic and out of human control.

However, speaking exclusively to Juan David Gutiérrez, professor at the University of the Andes and expert on public policies, competition and regulation of artificial intelligence, he clarifies that The difference between an artificial intelligence system and a computational system is that AI systems have a certain autonomy to fulfill the pre-established purpose.

You may also be interested in: The ‘Joi Effect’ or the emotional risk of humanizing artificial intelligence

Despite this, according to the academic, it should be noted that one of the principles of artificial intelligence is to monitor the person under control at all times. This is especially true when these types of systems are connected so that they operate off the network.

For Gutiérrez, what happened with this company is not a mistake, because in the end, “If it’s an automated learning system, the system will find a way to solve its problems.”

This situation could be compared to what happened months ago with the Microsoft company, where a security provider’s error in code affected thousands of computers around the world. “It doesn’t have to be an artificial intelligence system that does something unexpected to have a negative impact on these types of Internet-connected systems”explained the professor and the lawyer, emphasizing that what is important in these cases is still the human factor.

The incident, according to what the Japanese company reported, only happened during tests; Sakana AI continues its developments, protecting the tool’s ability to generate scientific articles.

Keep reading: Petro opened the first Artificial Intelligence faculty in Latin America with over $55,000 million, what will she do?