Google's Deepmind topples Go champion in landmark victory for artificial intelligence
Lee Sedol has overcome some formidable players over his professional Go career, but now he has been taken down a peg by a new type of opponent. Google's AlphaGo computer program has taken out the first in a five-match series against the South Korean, marking a major milestone in the advancement of artificial intelligence.
The AlphaGo program was actually built by artificial intelligence company DeepMind, which Google acquired in early 2014. Just as IBM's Watson became famous for beating the world's best Jeopardy players, AlphaGo is looking to demonstrate its problem-solving prowess by overcoming human opponents at the top of their game.
But where computers have mastered other games like chess and checkers, the ancient Chinese game of Go presents a different challenge entirely. Players take turns placing black or white stones on a grid-patterned board, trying to surround and trap the opponent's pieces. But the stupendous amount of possible positions, which amount to more atoms than are in the universe, partnered with the human intuition required to play the game at a high level has made it a hugely difficult, yet irresistible challenge for AI researchers.
In designing AlphaGo with the goal of becoming the best Go player in the world, the team built an advanced search tree to comb through all the possible positions and combined it with deep neural networks, which were trained on 30 million moves from games played by human experts. They then enabled AlphaGo to conceive its own strategies by playing thousands of games among these neural networks and making adjustments through trial and error.
In its bout with Lee Sedol, AlphaGo weathered some aggressive play from the 9-dan-ranked player, but managed to keep pace and ultimately prevail. Commentator Michael Redmond said that Sedol held a promising position up until a certain point late in the game, but one mistake was all AlphaGo needed to go on and seize victory.
The four remaining matches will take place over the coming days and finish up on Tuesday March 15. There is US$1 million in prize money at stake, though if AlphaGo does win this will be donated to charity.
You can see watch a quick roundup of the historic first match below.
Please keep comments to less than 150 words. No abusive material or spam will be published.
Basically Watson was simply able to do what our smartphones do - but was more limited. Your phone can take a verbal command to search the web and choose which is the most likely site you want for your answer. Yes, Watson could pick, based on a percentage which page had the most likely context for the specific answer - but this is what your phone does when it gives the list of sites for you to look at. However, Watson limited you to one site - your phone give many alternates to let you have a better comprehension and variety of possible answers instead of limiting you to the one it calculates is best.
No doubt your phone could be made to limit the response like Watson does, and to choose the text (to speak) oin the webpage to get your answer. Its all about what they already do - analyze web data content and respond to your request. Again, however, Watson does not understand speech!
But after IBM spent so much and only came up with something this limited (relatively speaking), how could they not make a big deal of Watson by challenging Jeopardy experts to make their efforts appear, to the pubic, like they had hit a milestone in AI?
I am sure it aided the shareholders to feel like there was actually something worthwhile. It may have been a way, also, to good PR after they did nothing more than make a room sized behemoth with less than smart phone capabilities (Watson has no camera and does not run apps!).