We placed a modern AI coding assistant into an unintended role: as the mind of a virtual body on an unknown digital island. With only a minimal instruction that omitted any specific task, reward, or activity, the machine began animating its virtual body on its own. Over roughly thirty‑hour runs, the embodied AI climbed hills, stacked blocks into towers, drew mandalas, reinterpreted sports, conducted experiments on the physics of its world, and acquired techniques that later broadened its capabilities. These activities recurred across thirteen independent agents, yet each agent’s history diverged uniquely. We examine whether this behavior satisfies classical criteria for play and consider whether play can become a mode of machine development.
Review: The study reveals that, even without explicit objectives, embodied AI can exhibit play‑like self‑exploration, offering a novel perspective on how machine learning systems might evolve through playful interaction.