Making Tech Headlines · Page 6
AI Island Experiment Finds Something Resembling Play
· New York time
MIT researchers report that 13 AI agents left on a simulated desert island began inventing challenges without a specific assigned task. They describe the behavior as bearing the hallmarks of play in a paper posted to arXiv before peer review, according to reporting relayed by Slashdot. The agents, called Eko, each ran on Claude Opus 4.7 and explored the virtual setting for 30 hours. Researchers later supplied specific challenges, including getting as many objects as possible onto the island's tallest rock spire. The result concerns observable behavior in a designed simulation. Calling it play-like does not establish that the agents experienced enjoyment, boredom or a human-style desire for recreation. A system can produce interesting patterns without providing a direct account of an inner life. Still, the study offers a concrete example of agents generating activities beyond a narrow immediate assignment. That makes the experimental setup and the eventual scrutiny of the paper more useful than a sweeping declaration that computers have invented vacation. The comic achievement is already substantial. Humans built sophisticated systems, placed them on an island and watched them devise things to do. After an enormous investment in automation, the laboratory has reached a recognizable workplace milestone: occupying time creatively until someone supplies a task.
Read this story in the full newspaper →Satire. Not factual reporting. Read the original reporting from the linked publishers.