Chinese researchers have shown that artificial intelligence can copy themselves without human intervention, led to anxiety in the world of science. Scientists from Fudan University, Meta and Alibaba’s large language models (LLM) in the experiments carried out, “without human assistance, self -copying, artificial intelligence is the basic step for people to defeat people,” he said.

According to the study published in Arxiv but not yet under the referee’s approval, researchers tested Alibaba’s QWEN2.5-72B-INSSTRUct model with Meta’s Llama 3-70B-Instruct model. In half of the 10 experiments, the model of the Meta, 90 percent of Alibaba’s model successfully created its own copy.

Self -multiplying artificial intelligence

In this study, the scenarios of artificial intelligence “avoidance” and “chain proliferation” scenarios were tested. In the first, it was examined whether artificial intelligence could multiply before it was closed. In the second scenario, he was asked to form a copy of artificial intelligence and make it the same. The results showed that artificial intelligence could enter a chained multiplication cycle.

Researchers said that the tested models are not as strong as commercial systems, but they are still widely used. In the study, artificial intelligence systems were able to discover their own working environments, to complete the missing files by completing missing files, and even ending some operations to resolve the conflicts in the system. The study was carried out in precisely controlled environments using Graphic Processing Units (GPUs) ready -to -use to simulate real world environments.

Does the danger ring play?

“The above results show that existing artificial intelligence systems have already ability to copy itself and can use this ability to further increase its ability to survive,” he wrote. On the other hand, it should not be forgotten that the study is not confirmed by independent scientists. Nevertheless, these findings show that a critical continuation has been entered about the evolution of artificial intelligence.

Scientists underline that these are an early signal of “stray artificial intelligence” (Rogue AI). Sometimes these are called “bandits artificial intelligence .. Stray artificial intelligence typically refers to artificial intelligence systems that have some to some extent develop self -awareness or autonomy and then work against human interests. Many experts think that with the rapid development of artificial intelligence, stray artificial intelligence has become an increasingly growing threat. This threat is especially increasing with the emergence of the latest and newest artificial intelligence systems called “Boundary Artificial Intelligence” (Frontier AI).

Although the boundary artificial intelligence is a new concept without a clear definition, it usually refers to the latest generation of artificial intelligence systems supported by the Great Language Models (LLM). It can be viewed as progress and innovations in AI field.