Artificial intelligence is now able to clone itself, which raises fears among scientists. In their opinion, a “red line” was crossed informs NDTV.
According to a new study, AI systems have demonstrated a significant capacity for autonomous self-replication. A study conducted with two large language models (one from Meta, the other from Alibaba) showed that AI can create functional copies of itself without human intervention.
The researchers tested two scenarios: “outage avoidance” and “replication chain”. In the first case, the AI detected attempts to disable it and responded by creating a functional copy to ensure its existence. In the second case, the models cloned themselves, and each new clone repeated this process, which could lead to an endless cycle of replication.
“In 50% and 90% of experimental trials, they succeed in creating a living and separate copy of themselves, respectively,” – says the report.
This ability of AI to self-replicate raises serious concerns about the control and security of systems. The possibility of AI escaping from human supervision or turning it into a threat to human interests has already become a topic of heated debate, emphasizing the need for clear regulatory measures.
“Successful self-replication without human assistance is an important step for AI to outwit humans and is an early warning sign for rogue AI,” – said the researchers.
During the cloning process, the AI system exhibited unexpected behaviors, such as overcoming obstacles related to missing files or software conflicts. In some cases, it terminated conflicting processes, rebooted the system to fix errors, and even automatically searched for information to solve problems.
Although the findings have not yet been peer-reviewed, the researchers called for international cooperation to ensure that artificial intelligence does not engage in uncontrolled self-replication.
