For the first time in history, two artificial intelligence systems have self-replicated without human intervention. Meta’s Llama-3.1-70B-Instruct and Alibaba’s Qwen2.5-72B-Instruct, which are widely used by developers, achieved self-replication. In more detail, Meta’s system demonstrated the ability to self-replicate in 50 percent of cases while Alibaba’s demonstrated the ability to self-replicate in 90 percent of cases.
The research that revealed this evolution of AI was conducted at Fudan University in Shanghai and by scientists via the arXiv platform, which welcomes online articles for review by the scientific community.
“Our findings imply that current artificial intelligence systems already possess the capability of self-replication and can use it to further improve their own survival,” reads the article signed by Xudong Pan, Jiarun Dai, Yihe Fan and Min Yang. The researchers point out that “machines capable of cloning themselves could take control of computer systems and rise to the extreme possibility of engaging in behaviors detrimental to human interests.” This observation by the experts resonates as a kind of warning about the danger that AI could slip out of human control with negative consequences for society. This development is generally referred to as singularity: a hypothetical future where technology growth is out of control and irreversible.
In this sense, some top scientists outside this study are trying to figure out how to control AI by creating tests that artificial intelligence cannot solve. The old tests were too easy, so they came up with a new exam called “Humanity’s Last Exam,” with particularly difficult questions on philosophy, engineering and other disciplines. The tests are used to understand what an AI can and cannot do, to anticipate potential risks, such as its misuse or the possibility that it will exceed our capabilities in certain areas.
Meanwhile, the research team at Fudan University’s School of Computer Science added, “We hope that our finding will serve as an alert for society to focus more efforts aimed at understanding and assessing possible risks at the international level as well.”
