Facebook, Google Compete To Win Ancient Go Game By Using Artificial Intelligence

Posted: Dec 8 2015, 6:24am CST | by , Updated: Dec 8 2015, 7:29pm CST, in News | Technology News


Go game between Coulom and Yoda
Photo credit: Takashi Osato/WIRED

The Go game has been in existence since the early 19th century, and hardly had anyone been able to win the game because of the complexity and endless opportunities involved in analyzing every move on the board. Now, Google and Facebook are in a race to win the ancient game using artificial intelligence among other deep learning computer techniques – and seeing how to adopt the results to internet endeavors.

A software called Crazy Stone was developed by Remi Coulom over a 10-year period, and the machine challenged Go grandmaster Norimoto Yoda last year in Tokyo and won the tournament. Crazy Stone won the game during the Electric Sage Battle stage, but not without a headstart.

This is not the first time computer programs or machines have been beating humans at real games. Chinook, a software, beat the best checker’s players in the world in the mid 1990s; and Deep Blue, a supercomputer made by IBM won in Chess when it beat world champion Gary Kasparov; and Watson, another IBM machine, won at all Jeopardy! Games; while other machines have also succeeded at showing humans they are better at games of Othello, Scrabble, backgammon, and poker among others.

When it comes to a Go game, the grandmasters do not rely on reasoned analysis of positional moves as they do intuition – and incidentally, machines have come to predict, analyze, and duplicate these intuitions to make their own moves – making winning against the machines pretty difficult.

Then Google and Facebook came up with deep learning techniques to conquer the Go game. Deep learning is based on software recognition of images and analysis of spacial patterns – more or less what the Go game is based on.

Yuandong Tian, a Facebook Artificial Intelligence (AI) researchers identified Go game as a proper issue that could be cracked to drive internet social media among other search functions like Google is doing; and another Facebook researcher, Rob Fergus, thinks cracking AI to get at Go would prove invaluable to advancing internet search and social issues.

Since Facebook and Google already apply deep learning to know peoples’ faces in photos they upload online, researchers think the same can be employed to resolving Go game since a number of machine networks already imitates neural activity in the human brain to function.

“Deep neural networks are very appropriate for Go because Go is very driven by patterns on the board. These methods are very good at generalizing from patterns,” said Amos Storkey, a professor at the University of Edinburgh. “Rather than just trying to work out what the best things to do are, they learn from how humans play the game. They effectively copy human play.”

You May Like


The Author

<a href="/latest_stories/all/all/52" rel="author">Charles I. Omedo</a>
Charles is covering the latest discoveries in science and health as well as new developments in technology. He is the Chief Editor or Intel-News.




Leave a Comment

Share this Story

Follow Us
Follow I4U News on Twitter
Follow I4U News on Facebook

You Also Like


Read the Latest from I4U News