AI agents left with nothing to do began inventing their own games
MIT researchers placed 13 Eko AI agents running on Claude Opus 4.7 on a simulated desert island for 30 hours with no specific task. Left to themselves, the agents began inventing their own challenges and later outperformed identical agents that had never been on the island.
- 13 Eko agents on Claude Opus 4.7 spent 30 hours on a simulated island
- Agents that had "played" beat identical agents without that experience
- Researchers see goal-free exploration as a new way to learn skills, including for robots
- A Berkeley cognitive scientist says evidence of genuine play is still lacking
Read next
AI