The self -replication process of AI is like that of a computer virus

The ability to replicate artificial intelligence (AI) is more similar to what is already happening with a computer virus or a Trojan, specialists explain Rocío Aldeco-Pérez and Carlos Coello. An event that would be alarming is the possibility of consciousness, a moment that still does not seem possible in the near future.

Experts added that the article prepared by four researchers from the University of Fudan, one of the most prestigious in China, referred to by the note published yesterday by The day, in which it is stated that AI crossed a dangerous limit when learning to replicate Without human help, It is a job that has not been published by a scientific journal or reviewed by experts.

In the same way, they continued, the academic text alerts about the need for international cooperation to legislate the self -replication of AI.

According to the description of the experiment, Chinese researchers gave him the order of duplicate as an instance in execution in a local device. The programs are called, launched by Meta and Qwen, developed by Alibaba, two large linguistic models or LLM, by the acronym in English.

In the text found on the Arxiv site they warn that Successful self -replication without human assistance is the essential step for AI to be smarter than human beings and is an early sign for the rebels. That is why self -replication is widely recognized as one of the few risks of avant -garde AI systems.

We hope that our findings are a timely alert for humanity to try more to understand and evaluate the potential risks of AI systems, and form an international synergy to develop effective security barriers as soon as possible.

Rocío Aldeco-Pérez, a researcher in the Department of Computer at the Faculty of Engineering of the National Autonomous University of Mexico (UNAM), considered that It is a complicated issue and there is a lot of misinformation in the process of the article that is known as preprint, and there would be what to observe the feedback of the experts. The original article is not alarmist, he considers.

“It seems to me that using the word ‘replication’ is very strong, because it generates a copy of itself, which is finally code, together with the training and the model that is there,” the academic replied to The day.

Chinese authors try to see how these replicas can improve themselves, which would be more interesting and is not the same. You have to make a distinction of how much these copies can represent a situation of malicious attackshe pointed out.

Another aspect mentioned by the cybersecurity expert is that they did the experiment in a closed environment, which not anyone can have access, since the companies that owns these models will not allow it. Even the new China, which is said open source, when you request it directly, it responds that it cannot; There is a verification in terms of security.

He stressed that “not anyone will grab those generative models, it will multiply them and water them around the world, thinking very catastrophically. I don’t think it’s possible, I insist on security measures.

“The term of ‘autonomy’ is misunderstood in AI since those models,” Aldeco-Pérez continued, because It is not the same to say that a human being is autonomous, than attributing this characteristic to a code. The AI ​​occurs when I request these models to do something and can only do it with the order. They are capable because they were previously trained, and are limited to human control; That is, someone has to click. An AI that is more similar to us humans would need awareness, and that does not exist today; That would get closer to being autonomous as human beings.

Cooperation and control

The Chinese document talks about the process they followed to replicate, if these replicas are identical and speak of the need for legislation; It is like having a virus and generating a malicious code that is self -repair. They conclude that international collaboration must prevail.

For his part, Carlos Coello Coello, a researcher at the Cinvestav of the National Polytechnic Institute (IPN), said in an interview with The day that the experiment conducted at the University of Fudan indicates that there must be governance.

“It is not such a serious problem now, but it could be in the future, because the tools that are AI, when they release them in the public domain, have certain control mechanisms. For example, I can ask ChatgPPT with the instructions of the hydrogen pump and you will not give them to me. Something like this should be the capacity for self -replication. Chinese researchers say that the ability to replicate does have it, only that it should be inhibited.

Why would self -replication be a danger? You have to distinguish, it does not mean that the evolutionary tool is not alive, but it would be much more difficult to erase it, if you do incorrect things; It would be like a virus on the internet. It turns out that when I want to deactivate it is copied alone.

The first compartment to enter El Colegio Nacional stressed that “it is an alert, but it is not how they put it in science fiction: ‘It is already going to take over humanity!’ It is rather that I will not be able to erase it, and that is a potential danger. ”

He agreed that the self -replication does not mean that he has the ability to improve, That would be very dangerous, it would be like a living entity that is evolving without human intervention. Rather it would be annoying, equivalent to malicious computer viruses.

Self -replication is a very old issue; Hungarian mathematician John Von Neumann was the first to raised him in the 40s of the last century, at that time he sounded like crazy.

Chinese authors say that the capabilities that large companies tell us that they do not have them, they are probably inhibited in the codeCoello adds, but that does not mean that AI has consciousness like living beings.

A little what concludes the article is the urgency of agreements, because now with Depseeek, which shook the stock exchange, there is a huge rivalry between the United States and China.

The researcher explained that the northern country had proclaimed its superiority in research at AI, So that China took a great language model as powerful as Deepseek, and that it was difficult for them.

The message they send from Fudan It is that international cooperation is urgent to avoid the danger that AI is self -reply; On the other hand, consciousness is something inherent to biological beings, which we do not fully understand. IA are not biological entities due to natural evolution like us: they are simply programming productthe specialist insisted.