From
Send to

[Weekender] AlphaGo’s win would mark AI watershed

March 4, 2016 - 17:55 By Kim Young-won
Expectations are running high in the technology sector with the upcoming battle between a human Go champion and a computer program.

Go is a board game in which two players place black and white stones on empty intersections or points of a board that is lined with a 19-by-19 grid.

Google’s artificial-intelligence program AlphaGo garnered media attention after defeating a professional Go player in October for the first time in an even match.

(123rf)

It has been about six months since that first victory, and AlphaGo is gearing up to pit its wits against Lee Se-dol, one of the world’s most prominent Go players, with a 71.9 winning percentage.

If the computer program wins, it will be a watershed moment in AI research.

Even if it loses, the software, which can learn from experience, will be able to further evolve and it might be able to defeat the best human player in the ancient Asian board game in the coming years or even months. The AI program can absorb a massive amount of data and learn through trial and error just like a human.

Behind the development of the Google AI software is DeepMind, a London-based company which was founded in 2010 by a group of computer engineers including its CEO Demis Hassabis. The firm focuses on developing neural networks, which work like a human brain, exchanging messages among installed network nodes.

Initial models of AlphaGo were built to master retro video games and it did master many, such as Space Invaders. The company eventually turned to the mastery of Go, which was thought to be next to impossible due to the immense number of possible moves, at more than 10 to the power of 700.

The company was acquired by Google in 2014.

DeepMind’s CEO said in a statement that the “complexity is what makes Go hard for computers to play, and therefore an irresistible challenge to artificial intelligence researchers, who use games as a testing ground to invent smart, flexible algorithms that can tackle problems, sometimes in ways similar to humans.”

DeepMind trained its AI program with over 1,000 Go matches. The software can predict human moves 57 percent of the time, according to the company in January. Its prediction capability may have improved since then.

After AlphaGo completes its goal of becoming the champion of the Go game, its next steps will be to tackle complex issues such as climate change and diseases such as cancer.

Last month, DeepMind announced the launch of a new project DeepMind Health, which will address health-care issues with AI. “It was clear to us that we should focus on health care because it’s an area where we believe we can make a real difference to people’s lives across the world,” said the firm.

By Kim Young-won (wone0102@heraldcorp.com)