AlphaGo ready for next triumph

On the ratings site of the word’s top-ranked Go players, a distinctive name - Google AlphaGo - appeared for the first time on March 14. Holding fourth place, AlphaGo was on its way to comfortably trouncing Korean champion Lee Se-dol, who was listed in fifth place.

The triumph of the self-learning algorithm software developed by Google DeepMind generated both dystopian imaginary and optimism. It triggered enormous public interest on where DeepMind or, more broadly, artificial intelligence research is heading.

Co-founder Demis Hassabis said on Tuesday that the unit will be focused on studying the latest Go match for the next few years and the next challenge is yet to be decided.

Hassabis previously cited the StarCraft video game as a potentially interesting next game, immediately creating online buzz, but he made it clear the ultimate goal of the company is to expand the scope of applications for artificial intelligence and strengthen its ability to solve intelligence.

“It’s to the extent that they’re useful as a testbed, a platform for trying to write our algorithmic ideas and testing how far they scale and how well they do,” Hassabis said. “It’s just a very efficient way of doing that. Ultimately, we want to apply this to big real-world problems.”

Experts also said that the board game battle is more of a platform to demonstrate the performance of DeepMind’s program, though the fields that the company is looking at include complicated issues such as fighting climate change and disease, as well as health care and robotics.

“The real objective is not about playing the game of Go well,” said Cho Sung-bae, a Yonsei University computer science professor whose research focuses on artificial intelligence and pattern recognition. “The Go match showed how deftly the artificial agent inside AlphaGo can process an input and make sense of it.

This ability can translate into a system that takes decisions leading to the desired outcome among a myriad of possibilities and chances.”

Indeed, an algorithm behind AlphaGo, dubbed Deep-Q Network, displayed highly sophisticated decisions that Go commentators couldn’t initially figure out, but they later proved to be crucial for the win.

AlphaGo’s successful performance was built on two types of neural networks that work quite similar to the human brain. One of those programs analyzes the state of the game board, while the other manages the next move. Both systems, like the human brain, can learn from experiences.

Cho said the principle can be applied to many industries.

“In the case of the medicine and health care sectors, there can be a number of ways to treat patients,” Cho said. “The algorithm used in AlphaGo can help doctors and health care professionals learn ideal treatments and prescriptions.”

DeepMind is already venturing into the health care field with a mobile app called Streams. The application is designed to present information to help doctors and nurses detect cases of acute kidney injury.

The Google-owned artificial intelligence company is working with the Imperial College London and the Royal Free London NHS Foundation Trust on the project.

It may be decades before any of the systems are utilized in hospitals around the world, but there are fields that are expected to show more rapid progress.

BY PARK EUN-JEE [park.eunjee@joongang.co.kr]