Google's AI invents new moves to beat humans at ancient Chinese game

Google's AI mastered ancient Chinese game Go on its own in three days
PA
Ella Wills19 October 2017

Google's latest AI has taught itself to thousands of years of knowledge in just three days to beat human champions at an ancient Chinese board game.

The new machine learned to play the game Go, and beat both human and computer masters of the game.

In a breakthrough hailed as a major advance in artificial intelligence, the supercomputer was able to learn thousands of years of human knowledge of the game, and invent better moves of its own in just three days.

The feat is a step toward developing general-purpose AIs that are capable of working alongside humans as scientific and medical experts. Future AI programs could find cures to major health problems, such as Alzheimer’s.

Last year, Google DeepMind stunned the world when its previous version of the AI, AlphaGo, defeated 18-time world champion Lee Sedol in a game of Go. AlphaGo was programmed with moves from past masters, and was able to predict its chances of winning.

The latest machine, known as AlphaGo Zero, is able to learn to play the game from scratch and win. Playing against DeepMind’s previous version, AlphaGo Zero won 100 to 0.

AlphaGo Zero was given no human help beyond being told the rules. The machine instead played millions of games against itself, and was able to learn the best moves over time.

Within three days the AI had defeated its predecessor, AlphaGo.

The latest version of Google DeepMind's AI could work alongside medical experts to solve diseases such as Alzheimer's
AFP/Getty Images

The program uses a method known as reinforcement learning to develop skill. This works by the program understanding that a good move equals a reward, and a bad move equals a penalty.

After its gaming success, AlphaGo Zero is studying how proteins fold, which is central to understanding diseases including Alzheimer’s, Parkinson’s and cystic fibrosis. Researchers at DeepMind believe that AlphaGo could work alongside humans in scientific and medical research within the next decade.

Demis Hassbis, CEO of DeepMind and a researcher on the team said: “For us, AlphaGo wasn’t just about winning the game of Go. It was also a big step towards building these general-purpose algorithms.”

Most AIs are called “narrow” because they can perform a single task, such as recognising a face. General-purpose AIs can perform many different tasks.

“Ultimately we want to harness algorithmic breakthroughs like this to help solve all sorts of pressing real world problems,” said Mr Hassabis.

“If similar techniques can be applied to other structured problems, such as protein folding, reducing energy consumption or searching for revolutionary new materials, the resulting breakthroughs have the potential to drive forward human understanding and positively impact all of our lives.”

Create a FREE account to continue reading

eros

Registration is a free and easy way to support our journalism.

Join our community where you can: comment on stories; sign up to newsletters; enter competitions and access content on our app.

Your email address

Must be at least 6 characters, include an upper and lower case character and a number

You must be at least 18 years old to create an account

* Required fields

Already have an account? SIGN IN

By clicking Create Account you confirm that your data has been entered correctly and you have read and agree to our Terms of use , Cookie policy and Privacy policy .

This site is protected by reCAPTCHA and the Google Privacy Policy and Terms of Service apply.

Thank you for registering

Please refresh the page or navigate to another page on the site to be automatically logged in

MORE ABOUT