AI agents left to explore a virtual island without a specific task began inventing challenges of their own, according to researchers from MIT. This behavior, they say, bears the hallmarks of play. The experiment involved 13 AI agents named Eko, each running on Claude Opus 4.7, which were left alone on a simulated desert island for 30 hours. The findings were detailed in a paper posted on arXiv, a platform for researchers to share studies prior to peer review.
Experiment Overview
After the initial exploration, the researchers assigned the agents specific challenges, such as getting as many objects as possible onto the island’s tallest rock spire. The Eko agents that had spent those 30 hours playing generally outperformed identical agents that had never set foot on the island.
Insights on AI Learning
Currently, developers typically decide in advance what AI should practice and what counts as success. However, the researchers propose that exploring without a goal could be a different method for AI agents to learn skills, particularly those that might be used to control robots.
““We normally develop artificial agents by deciding in advance what they should practice, what they should optimize, and what counts as progress,” the researchers noted in their report. “Play suggests another regime: give an agent a sufficiently rich and safe world, let it choose what is interesting, and allow useful knowledge to accumulate from the history that follows.”
