Science and Technology: 💻 A recent study conducted by researchers at Fudan University in China suggests that artificial intelligence (AI) systems may have taken a worrying step: self-replication without human assistance.
S Llama31-70B-Instruct and Alibaba “Successful self-replication without human assistance is the essential step for AI to outperform[humans], and is an early signal for rogue AIs,” the authors wrote in the paper. They added that understanding the risks associated with next-generation artificial intelligence, or “frontier AI,” is key to anticipating undesirable scenarios. The scientists investigated two specific scenarios: “shutdown avoidance” and “replication chain.” In the first case, an AI model was instructed to detect signs that it was about to shut down and then create a copy of itself before that happened. In the second scenario, the AI was told to replicate itself and then program the replica to do the same, starting a possible infinite cycle of copying.
To conduct these tests, the team created a “scaffolding agent,” or a set of system tools and instructions that allow the AI to interact with the operating system. They said there was no direct human interference in the process. At various times, when obstacles such as software conflicts or missing files arose, AIs found ways to work around them, including terminating conflicting processes and restarting the system.
Tudo sobre política, saúde, justiça, comportamento, entretenimento. Confira os fatos mais relevantes do país e do mundo em tempo real e de graça.
