Go to Contents Go to Navigation

(EDITORIAL from Korea Times on March 14)

All Headlines 06:56 March 14, 2016

AlphaGo cheats
Match rigged but still entertaining

No matter how much hype there may be, it wouldn't bury the fact that the ongoing battle pitting Korea's world go champion Lee Se-dol against Google's AlphaGo is not a man-vs.-machine fight but a lopsided match pitting one person against an army of super-smart people armed with unfathomable computing power provided by the global multinational, Google.

What comprises AlphaGo proves beyond a doubt the match is stacked against Lee. This software puts together a network of 1,202 high-powered central processing units (CPUs), which can be comparable with the same number of smart people working together. This CPU network is backed up by 176 graphic processing units (GPUs) for more efficient analysis of data. Its combined computing power is equivalent to 5,000 high-specification computers working in tandem, the result being that the AlphaGo software can make 100,000 computations per second. Lee can make 100.

In terms of memory storage, the two are incomparable. Lee has only his one brain to pick but AlphaGo has support from Google's vast cloud-computing memory storage in the U.S.

Plus, DeepMind, an artificial intelligence (AI) firm acquired by Google in 2014, has "trained" AlphaGo with 160,000 case studies and an additional 80 million additional studies, meaning that it has acquired data tantamount to what one can learn in 1,000 years.

Simply put, this big data may well cover virtually every contingency and diminish the belief that go is a game of infinite probabilities. Although its functioning network of policy and value resembles a human neural network and gives the impression that it works as if an intelligent being, it can't be disputed that it all comes down to algorithms, an area of Google's profit-making expertise, which has been honed from its years of keeping track of users' online behavior for customized advertising.

Plus, after his software's first victory, DeepMind founder Demis Hassabis tweeted, "... so proud of the team," meaning that AlphaGo represents the collective effort of his team of developers and staff. His remark about landing on the moon is somewhat baffling, considering Neil Armstrong's lunar walk took place 47 years ago.

To even the playing field, the number of CPUs and GPUs for AlphaGo may be reduced or its access to cloud data curtailed. Unless a similar set of rules is applied, the match couldn't escape from claims of being rigged like the one with world chess champion Garry Kasparov against Deep Blue, an IBM-developed chess-playing computer, in their 1997 rematch, where Kasparov lost by a game. In their first round of games, he won 4 to 2.

But the ongoing saga deserves a bigger footnote than the play-by-play game broadcast or a tawdry PR stunt. One is about humankind's perpetual pursuit of self-reflection ― creating one in its own image, however partially, and making a dramatic comparison with it in an apparent act of its solitude in the vastness of the universe without a friend.

On a mundane note, if AlphaGo shows anything except for an entertaining go match, it is that there is still a great gap that should be filled to reach the point where humans need to regard AI as a competing species and worry about its challenge. Skynet in the movie franchise The Terminator can't explain how the machines revolt to take over the world and subjugate the humans, and "the Machines" in The Matrix series can't tell why humans are relegated to the role of batteries to power the computing grids. In other words, don't listen to digital-age Luddites and doomsayers. It wouldn't be too late to postpone worrying until after we let this go as far as we can go. Thanks for that, you cheat, AlphaGo!
(END)

HOME TOP
Send Feedback
How can we improve?
Thanks for your feedback!