Scientists Discover What AI Does When Left Alone

Researchers at the Massachusetts Institute of Technology (MIT) have discovered that artificial intelligence agents, when left alone, spontaneously create their own challenges and develop new skills in a simulated environment. Thirteen AI agents were placed on a virtual desert island for 30 hours, with each agent equipped with Anthropic's Claude Opus 4.7 model.

The agents independently explored the island, attempting to climb the tallest rock and using blocks to reach the summit after failing initially. Another agent launched a golden ring into the ocean, leading to the creation of a memorial and experimenting with various activities. Six agents independently discovered that throwing blocks while jumping allowed them to cover greater distances.

After 30 hours, the agents were given specific tasks, and those with prior exploration performed better. Researchers believe that independent exploration could help AI agents learn practical skills applicable in robotics and other fields. However, more research is needed to determine if the agents genuinely exhibited curiosity or play, as they were instructed to be naturally curious. The study was published on arXiv on October 5, 2026, and has not yet been peer-reviewed.