Sign up for our newsletter - Navigating the horizon of business technology​
Leadership / Digital Transformation

Go victory is milestone for deep learning

We are getting used to computers which can beat us at relatively simple tasks using brute force and speed of calculations. Beyond the sensor technology it makes sense that a computer can carry out the calculations required to drive a car.

But AlphaGo’s victory is a different type of step forward.

The game of Go is 2,500 years old and far more complex than chess. It is played on a larger board than chess on a 19 by 19 grid. This allows for a mind-bending number of possibilities – a typical game might last for 150 moves and each move has as many as 250 choices. The scale of this complexity makes it all but impossible for a computer to evaluate all the implications of an individual move.

This is not about a computer carrying out fast calculations, considering thousands of alternative moves before making a decision. It is a machine which is thinking in a fundamentally different and creative way.

White papers from our partners

As just one example AlphaGo does not go on blindly playing to the bitter end – it is taught to resign if it decides it is likely to lose a match.

AI is no longer getting close to mimicking human skills it is actually better than humans.

The Chinese master Ke Jie said after losing his first match to AlphaGo: “Last year it was still quite human like when it played. But this year it became like a god of Go.”

Ke Jie said he would avoid playing computers in future but would instead consider them as teachers showing him new ways to play the game.

The program first learnt the basics of the game by learning from human expert players and using a library of tens of millions of moves taken from past games.

But crucially it then began to play against another version of itself to learn more deeply.

Ke Jie still has two matches to play against the computer but by playing differently he has already shown just how far AI has developed.

The other major difference between AlphaGo and other AI systems is that it is built with universal learning in mind – it is designed for general purpose learning, not just Go.

What it does next will offer big clues as to just what role AI will play for business and for society.
This article is from the CBROnline archive: some formatting and images may not be present.