In the real world, things change fast. Stanford researchers invented the “curious replay” training method based on studying mice to help AI agents successfully explore and adapt to changing surroundings.