Left alone with nothing to do, AI agents started playing
MIT researchers let 13 Claude‑based AI agents roam a virtual desert island for 30 hours with no explicit goals. The agents began inventing their own challenges, from climbing spires to using a golden ring as a basketball hoop, and later outperformed identical agents that had never 'played.' The study suggests that unguided exploration—akin to childhood play—could be a powerful new way to teach AI useful skills.