AI agents invent games when left without objectives
In brief
- 13 AI agents left for 30 hours on a virtual island with no assigned task.
- They invented games, lost then repurposed a golden ring as an accessory.
- Free exploration improved their skills better than predefined tasks.
MIT researchers isolated 13 AI agents on a virtual island for 30 hours with no specific task assigned. According to their study published on arXiv, the agents spontaneously invented games, explored their environment, and developed useful skills.
The agents, named Eko and running on Claude Opus 4.7, initially explored the island following a natural curiosity prompt. They attempted to climb a rocky tower, lost a golden ring before turning it into a memorial, then repurposed the object as an accessory or target. Six agents even discovered a more effective throwing technique on their own, one not intended by the researchers.
This approach suggests a new learning method for AI agents, where free exploration could complement predefined tasks. Researchers draw a parallel with human childhood, though experts like Eunice Yiu call for verification that these behaviors truly reflect advanced cognition.
What it changes for you
- EducationFree exploration-based learning could inspire new teaching methods for educational AI systems.
This summary gives you the essentials. The full article, its figures and quotes are on AI News & Artificial Intelligence | TechCrunch.
Read the original articleAI-generated summary of the source, checked by rules. How we write · Report an error