Games Computers Can Play

Computers that play to win

Computers can play games, and some play extremely well. A computer has beaten the best human players in the world at chess and at an ancient game called Go.

How can a machine be so good? It is not magic, and the computer is not really thinking like you. Let's see what is really happening.

Looking ahead

Imagine you are playing chess. Before you move, you think: "If I move here, my opponent might move there, then I could do this." You are looking ahead.

A game computer does the same thing, but much, much faster. It looks ahead at huge numbers of possible moves:

• If I move my knight, what could happen next?
• And after that? And after that?

It builds a giant tree of "what if" moves. A fast computer can check millions of positions in the time it takes you to blink. This is called searching.

Scoring each position

Looking ahead is only useful if the computer knows which positions are good. So it gives every position a score.

A high score means "this is great for me", like winning a piece. A low score means "this is bad for me". The computer then picks the move that leads to the best score it can reach.

This is a bit like following a clear set of rules to make a decision, the same idea as Making Decisions with If in coding.

Learning by practising

The very strongest game AIs do something extra. They learn by practising.

A famous Go program got better by playing millions of games against itself. Each game, it noticed which moves led to wins and which led to losses. Slowly it discovered clever strategies that even human experts had never tried.

This is machine learning: getting better from lots of examples instead of being told every rule. You can read more in How Does AI Learn?.

Why games suit computers

Games like chess and Go are perfect for computers, and there is a reason:

• The rules are clear and never change.
• Each player can see the whole board.
• You can tell exactly who is winning.

When the rules are this neat, a computer can search and score with great power. That is why computers do so well here.

The honest limits

But a game computer is not as clever as it looks. Here are some honest limits:

• A chess computer can only play chess. It cannot drive a car, write a poem, or tie a shoe. It is built for one job.
• It does not enjoy the game or understand why the game is fun. It just calculates.
• In real life, the "rules" are messy and full of surprises. That is much harder for a computer than a tidy board game.

So when a computer wins at chess, it is amazing, but it is winning at one clear task by searching and scoring very fast. It is not generally smart like a person.

To meet more clever machines, visit Meet Artificial Intelligence. And to build a simple game of your own, start with Coding!