Computer Beats Human at Hardest Board Game In The World (it’s not Chess)
The name of the game is Go. The Computer that won is called AlphaGo, made by Google. The Human that lost is called Lee Sedol, the best human in the world at Go.
The challenge was set forth in the following way. AlphaGo would play Lee Sedol in 5 games of Go. Whoever won 3 games gets $1,000,000 USD.
Lee Sedol predicted he would win 5–0 or at least 4–1.
Lee Sedol has lost. 3–0. There are 2 more matches to be played, but already the point has been made. A computer can play Go better than the best human. And Google is donating the money to charity.
What is Go?
Go is a game with origins thousands of years ago in China. It is a game that requires more intuition than chess and is more complex in terms of numbers of possible moves, which makes it a harder game for a computer to win.
How did it happen?
Google built this computer to take as inputs possible positions of pieces on a Go board. Then based on the current position it makes a decision about what the best move is for it to make. It calculates the move that will give it the highest probability of winning. But it does not calculate the move that will give it the highest margin of victory. So it will always play to win, but to it there is no difference between winning by 1 and winning by 100.
The computer was trained by looking at lots and lots of Go games. The more it saw, the better it go. In October 2015, it wouldn’t have won against Lee Sedol. So it loooked at more games and learned more. It would even play games against itself to get better.
Why did Google build AlphaGo?
AlphaGo uses a type of algorithm called Deep Learning, using a structure inspired by human neural networks. This is a major area of research and development in all the world of technology because Deep Learning algorithms are amazingly flexible. If you use Siri and Google’s voice commands, you know how good they are at understanding what you’re saying. If you’ve used both, you know that Google’s is much better. This is because the Deep Learning algorithm used by Google is better than the one at Apple. Google uses similar learning algorithms in Google Maps to identify house numbers based on images. They use Deep Learning for handwriting.
The general pattern of deep learning is much like go. Give the computer a lot of things you’re interested in, tell it what result you want, and it will learn it by looking at all the things you give it. Once you’ve given it enough things, you can give it one more new thing and it will know what it is.
For example, give it 10,000,000 images of cars and it can start to classify those cars into categories. Then with image 10,000,001 it can tell you that it’s a Blue SUV, you’re seeing the driver’s side, and the door is open.
Google is building a computer that can play Go better than humans because it can easily use that game playing computer to figure out which add you’ll like better on a Google Search and make more money.
Google is building a computer that can play Go better than humans because it can easily use that game playing computer to make self-driving cars.
Google is building a computer that can play Go better than humans because it can easily use that game playing computer to read handwriting on documents that have been scanned.
Are you getting it? Google is building a computer that can play Go better than humans because it can easily use that game playing computer to do anything.
I hope you found this article informative. Believe it or not, it took me hours to write this down, so why not pinging me on Twitter and give some ♥ back?