Module 13 of 16 · Investigate: Patterns & Recognition

AI Game Masters: Reinforcement Learning

AI that learns by playing! Rewards, strategies, and game-winning bots.

20 chapters

What this module covers

AI Game Masters is about AI that learns by trying things. Students meet the parts of every reinforcement learning problem: an agent, an environment, actions and rewards, and why the agent cares about future rewards as well as the next one.

The module then covers the main methods: Q-learning, deep Q-networks, policy gradient methods and actor-critic methods, with game AI and robotics as the worked examples. The last chapter connects it all to RLHF, the reinforcement learning from human feedback used to train chat assistants.

Dive deeper into how AI sees, hears, and understands. Explore computer vision and language AI through fun activities.

Chapter topics

  1. RL Fundamentals
  2. Agents and Environments
  3. Rewards and Returns
  4. Q-Learning
  5. Deep Q-Networks (DQN)
  6. Policy Gradient Methods
  7. Actor-Critic Methods
  8. Game AI with RL
  9. Robotics and RL
  10. RLHF (How ChatGPT Learns)

Plus 10 more chapters in the app.

Activities

  • Train a Game Bot
  • Reward Design Challenge
  • RL Playground

After this module, students can

  • Describe a problem as agent, environment, actions and rewards
  • Explain how Q-learning updates its estimates
  • Explain how RLHF shapes a chat assistant's answers

Go deeper

Start learning in the app

The first 3 chapters are free, no card needed.

Get the appSee all 16 modules