MIT Researchers Find AI Agents Invent Games During Open Exploration

Massachusetts Institute of Technology (MIT) researchers found that AI agents invented their own challenges when left to explore a simulated island without a specific task, according to TechCrunch. The experiment placed 13 agents called Eko, each running Claude Opus 4.7 from Anthropic, alone on the island for 30 hours.

The agents attempted activities such as climbing the island's tallest spire, throwing objects and arranging stones into shapes. Six agents discovered that releasing a block during a jump allowed them to throw it farther, a behavior the researchers had not intentionally designed into the simulation.

After the exploration period, the researchers assigned challenges such as moving as many objects as possible onto the tallest spire. Agents with prior time on the island generally outperformed identical agents that had not explored it.

The study was posted on arXiv before peer review. The agents' instructions described them as naturally curious, and UC Berkeley cognitive scientist Eunice Yiu told MIT Technology Review she wants more evidence of the cognition behind the behavior before calling it genuine play.