MIT researchers teach AI agents to play on virtual island
mit researchers put 13 ai agents on a virtual desert island for 30 hours with no specific task and found they invented challenges and learned skills. this could change how ai agents are trained.
mit researchers put 13 ai agents called eko, each running on claude opus 4.7, on a simulated desert island for 30 hours with no specific task. the agents began inventing challenges, such as conquering the island's tallest spire and building a heart made of stones. six of the agents learned to throw a block farther by letting go mid-jump, a capability the researchers hadn't intended. this experiment suggests a different way for ai agents to learn skills, especially those that could be used to control robots. the researchers see this technique as an alternative to traditional goal-oriented training.