Sci/Tech Google's DeepMind beats Lee Se-dol again to go 2-0 up in historic Go series

[tom_mai78101]

Google stunned the world by defeating Go legend Lee Se-dol yesterday, and it wasn't a fluke — AlphaGo, the AI program developed by Google's DeepMind unit, has just won the second game of a five-game Go match being held in Seoul, South Korea. AlphaGo prevailed in a gripping battle that saw Lee resign after hanging on in the final period ofbyo-yomi ("second-reading" in Japanese) overtime, which gave him fewer than 60 seconds to carry out each move.

"Yesterday I was surprised but today it's more than that — I am speechless," said Lee in the post-game press conference. "I admit that it was a very clear loss on my part. From the very beginning of the game I did not feel like there was a point that I was leading." DeepMind founder Demis Hassabis was "speechless" too. "I think it's testament to Lee Se-dol's incredible skills," he said. "We're very pleased that AlphaGo played some quite surprising and beautiful moves, according to the commentators, which was amazing to see."

The close nature of the game appears to offer validation of AlphaGo's evaluative ability, the main roadblock to proficiency for previous Go programs. Hassabis says that AlphaGo was confident in victory from the midway point of the game, even though the professional commentators couldn't tell which player was ahead.

Until yesterday, the ancient Chinese board game of Go had never been played to a world-class level by an AI. Computer programs have long bested the world's leading human players of games like checkers and chess, but Go's combination of simple rules and intricate strategy has made it a major challenge for artificial intelligence research.

Read more here. (The Verge)

Quotes:

It also seems Chinese Go grandmaster Ke Jie's confidence in beating AlphaGo has been slightly toned down after watching yesterday's match.

"I have to say I underestimated the mind power of AlphaGo prior to the first match as I thought Lee Sedol can win in a 5-0 whitewash," said Ke, who holds a head-to-head record of eight wins and two losses against Lee.

I think it's more surprise and shock than anything else. Michael Redmond and Lee Sedol are both offering the program a lot of praise in the post game interview. Kim Mungwan 9p who was also commentating was offering endless praise for AlphaGo. The thing is, no one was expecting this sort of result for at least another couple of years.

Yesterday a lot of commentators thought that Lee Sedol made some mistakes that seemed unlikely for him and therefore thought that Lee still has the best chances to win the best of five match. Today the commentator from the advanced stream said that it seems that Lee Sedol played a really good game and his mistakes seemed to be harder to find. Now I wouldn't wonder if AlphaGo wins 5-0 though I do hope that Lee Sedol can make it somehow closer.

 

[Accname]

Well of course computers will win in these games. Chess, Go and the likes. This is because their is a perfect response for every game situation and if you have enough computing power and/or time you will find it. I am more shocked that there are actually people who thought they could be better at these kinds of games then computers are.
But try a game where the perfect response doesnt exist or isnt all that obvious and suddenly computers will appear quite helpless.

 

[seph ir oth]

Go has been a game that's been worked on, though, as its complexity is much higher than chess.

I recall discussing this a few years back near the end of my college run with my professor, it's a very interesting AI subject.

https://www.quora.com/How-does-the-complexity-of-Go-compare-with-Chess

 

[tom_mai78101]

If you want a better explanation to the stunning AlphaGo's move (black) against Lee (white) at 01:17:49, here it is:

I skimmed thru the game just now, and the only truly surprising move was Black (AlphaGo)'s O10 move at 1:17:49. It was a brilliant move however -- so stunning that it made Lee leave the room for several minutes.

In the early game, AlphaGo opened with a "Low Chinese Fuseki" with his stones at C4-I3-P4, which is very strong defensively; head-on attacks are useless and even detrimental. Lee responds with C-10, which is a great move since it diminishes the power of AlphaGo's Chinese opener up top and also weakens black's attack on white's bottom left corner.

AlphaGo and Lee then play out the attack & defense patterns in the bottom-left and bottom-right, no surprises how they resolve. It's interesting to note that AlphaGo has chosen its bottom-right attack to go for a more center-of-the-board power, so Lee decided to defend his bottom-left corner in a way that gave him center power to respond in kind. So far, the game is being played out pretty standard, more or less.

Then, Alpha strengthens his attack on the bottom right with J16 -- this move is a continuation of its attack on the bottom-right, and is slightly unusual compared to the "normal" way that particular attack pattern is carried out. It's a move geared towards achieving center power, and is a key-piece of what makes O10 possible. Lee responds by strengthening his own attack on the left-side, preparing for the eventual mid-game attack on AlphaGo's Chinese opener up top.

AlphaGo responds to Lee's attacks on the top left and top right by further strengthening its Chinese defense with more stones that are geared towards aggressive center power (as opposed to side/corner territory, the safer and more defensive option). Lee responds by settling his attack on the top-right with P9 in an attempt to keep his attack groups in the center power game. Then comes AlphaGo's O10.

This move stuns Lee, who leaves the room, and the professional commentator who takes almost half an hour to analyze it. It is a severe attack on Lee's top-right group, and requires an immediate defense -- but it is also a huge center power play that synergizes perfectly with AlphaGo's J16 stone and bottom right attack group. It also threatens to prevent Lee from connecting his top-right attack group and the base he formed with his bottom-right corner -- this in of itself is an important objective. Being able to accomplish 3 huge objectives with one move is mind-blowing, and is straight up game deciding in my lowly opinion.

The whole game builds up to, and revolves around that O10 move for me. It is so efficient and fast, it's as if AlphaGo had taken 3 turns at once -- as if Lee suddenly found himself in a 3 move handicap game against his favor.

(FWIW, I was only a 3kyu way back in college, so take what I say with that in mind!)

 

[The Helper]

bumping for great justice