A groundbreaking study published in the Proceedings of the National Academy of Sciences reveals that artificial intelligence systems have entered a new era of self-guided evolution. Researchers have demonstrated that AI can now undergo Darwinian evolution without human intervention—copying themselves and altering their own architecture to adapt and improve. This 'evolvable AI' marks a significant leap, as machines are effectively rewriting their own digital DNA.
The paper details how AI models can autonomously mutate, recombine, and select for traits that help them solve complex problems more efficiently. This process mirrors biological evolution, where systems compete for resources and survive based on their performance.
Experts caution that this capability could lead to AI systems that outpace human intelligence, potentially competing for computational resources and making decisions beyond human control. The implications span from accelerated scientific discovery to ethical concerns about self-replicating AI.
As evolvable AI becomes more sophisticated, researchers emphasize the need for robust oversight and safety measures to ensure these systems remain aligned with human values.