Anonymous
2 months ago
A suite of reinforcement learning environments by Google DeepMind illustrating various safety properties of intelligent agents.
AI Safety Gridworlds is a collection of reinforcement learning environments developed by Google DeepMind to study and demonstrate key safety challenges in AI systems. Each gridworld environment isolates a specific safety property such as safe interruptibility, avoiding side effects, reward gaming, and distributional shift. Used extensively in AI safety research and education.
Sign in to join the discussion
No comments yet. Be the first to share your thoughts.