AI That Plays Games
Do you like playing games? Whether it is chess, checkers, a video game, or a card game, there is something thrilling about trying to outthink your opponent. For a long time, humans believed that playing games well required human intelligence — creativity, intuition, strategy. Then something remarkable happened. AI got so good at games that it started beating the best human players on Earth. Not just winning — completely dominating. And the fascinating part? Watching AI play games taught scientists some of their most important lessons about how machines learn.
The Games AI Has Mastered
Chess was one of the first great challenges for AI. For decades, humans believed that chess — with its deep strategy and countless possible moves — would never be mastered by a machine. In 1997, an IBM computer called Deep Blue beat the world chess champion, Garry Kasparov. That moment shocked the world. But Deep Blue mostly worked by calculating millions of possible moves very quickly — it was powerful, but it was not really learning. Then came AlphaGo. The game of Go is ancient and extraordinarily complex — experts said it would take AI decades to master it. There are more possible Go game positions than there are atoms in the observable universe! In 2016, an AI called AlphaGo beat the world Go champion Lee Sedol — and it did so by discovering moves that no human had ever played in thousands of years of human Go history. Video games are another arena where AI has excelled. An AI called OpenAI Five learned to play Dota 2 — a complex team strategy game — well enough to beat professional human teams. It learned entirely by playing millions of games against itself.
AlphaZero and OpenAI Five both learned by playing millions of games against themselves — a method called self-play. Without any human teaching them strategy, they invented their own — and discovered moves humans had never thought of. This is called reinforcement learning.
The way AI learns games tells us something profound about how intelligence works. When AlphaZero learned chess, Go, and shogi all by itself just by playing millions of games against itself, it did not follow human rules of good play. It discovered its own strategies. Some of its moves looked bizarre to human experts — and then turned out to be brilliant. This teaches us that there may be ways of thinking and solving problems that humans have not discovered yet, because we have always been limited by our own human way of seeing the world. AI exploring different approaches could help unlock those new ways. And game-playing AI is not just about games! The same techniques that taught AI to master Go are now being used to fold proteins, design new medicines, and figure out how to control fusion reactors — some of the hardest problems in all of science.
Games have clear rules and clear goals, which makes them perfect for teaching AI. Once an AI figures out how to win a complicated game, the same thinking can be applied to real-world challenges like medicine, climate science, and engineering.
Flashcards — click each card to reveal the answer
Terms
Definitions
Drag terms onto their definitions, or click a term then click a definition to match.
What is special about how AlphaZero and OpenAI Five learned to play games?
Why are games a useful training ground for AI?
Teach a Human to Play Like an AI
- This activity explores how AI uses trial and error to get better at games.
- Choose a simple game you know — tic-tac-toe, checkers, or a card game.
- Play THREE rounds with a partner. After each round, do the following:
- 1. Think about one move that did not work out the way you expected.
- 2. Decide what you would do differently next time — and write it down in one sentence.
- 3. Try that new strategy in the next round.
- After three rounds, look at your notes. Did you get better? What changed?
- This is almost exactly how reinforcement learning works for AI — try something, see if it worked, adjust, try again. The difference is AI does this millions of times and never gets tired or discouraged. What could YOU learn if you practiced like that?