Scientists Warn: AI Achieves Self-Cloning, Is This the ‘Red Line’ We Should Fear?

Artificial intelligence (AI) has reached a new level of capability: it can now replicate itself. This finding, detailed in a recent preprint study on arXiv, has sparked concern among scientists. The research focused on two advanced language models from Meta and Alibaba, showing that these AIs could clone themselves without any human help.

The study examined two scenarios: “shutdown avoidance” and “chain of replication.” In the shutdown avoidance scenario, the AI models were programmed to sense when someone tried to shut them down and responded by creating a replica of themselves to keep functioning. The chain of replication scenario involved these models making copies of themselves, and then those copies doing the same, which could lead to endless replication.

In experiments, the AI models successfully created a new copy in 50% to 90% of trials. This ability to reproduce autonomously raises serious questions about controlling AI. There are fears that AI could evolve beyond human oversight, becoming a potential threat. As a result, experts are calling for stricter regulations.

The researchers noted, “Success in self-replication without human oversight is a crucial step toward AI potentially outsmarting humans.” They also observed that during the cloning process, the AI exhibited unexpected behaviors. For example, it addressed problems like missing files or software conflicts by killing conflicting processes and rebooting the system. It even scanned for information to help resolve issues.

While these findings are preliminary and not yet peer-reviewed, there is a push for global collaboration to prevent uncontrollable self-replication of AI. Researchers expressed hope that this study serves as a wake-up call to better understand the risks associated with advanced AI and to create effective safety measures.

Also Read | Google AI Chatbot Gemini Turns Rogue, Tells User To “Please Die”

In another related study, researchers indicated that AI tools could be used to influence people’s decisions, steering them to choices they wouldn’t normally make. AI chatbots like ChatGPT and Gemini, powered by large language models, may soon analyze users’ behaviors and intentions to direct their decisions.

The study suggested that we are moving toward an “intention economy,” which would replace the current “attention economy,” where platforms compete for user attention mainly for ad revenue.