Lee Sedol vs Alpha go

[trout]

1st game;

AlphaGo defeated Lee Sedol by resign.

AlphaGo vs Lee Sedol 1:0

[Spaceman-Spiff]

Damn, Lee was playing so well at the beginning.

I want AlphaGo to win, but still pretty shocked.

[Mef]

1st game;

AlphaGo defeated Lee Sedol by resign.

AlphaGo vs Lee Sedol 1:0

I must say this trout -- It is quite admirable the way you dutifully report so many pro results here, in such a succinct and objective fashion, to the point where a game that will be certainly be considered an incredible milestone in both AI development and Go History is listed as though it may be tomorrow's weather report.

I am almost led to wonder if you are a forum posting bot (=

[Charles Matthews]

Damn, Lee was playing so well at the beginning.

I want AlphaGo to win, but still pretty shocked.

I thought AlphaGo was slightly behind for much of the game, but it appears Lee misplayed the complex position on the right side.

Michael Redmond commentating thought AlphaGo showed good timing; but he didn't like the thick play upper left, tenuki from the lower left allowing a double approach. AlphaGo found a way to take sente there (apparently messy) and move over to the right, keeping complexity alive.

Interesting "living go", which is ironic with one player being silicon.

[mitsun]

I am sure we will get lots of human analysis of the game, but I would be interested in the Alpha Go analysis. I am pretty sure it can produce a graph of its expected win probability per move. It should also be able to say how early in the endgame it calculated the remaining play exactly.

[Cassandra]

I am sure we will get lots of human analysis of the game, but I would be interested in the Alpha Go analysis. I am pretty sure it can produce a graph of its expected win probability per move. It should also be able to say how early in the endgame it calculated the remaining play exactly.

Here is what two electronical "seniors" thought about the winning probability of their youthful buddy.

Please be careful while interpreting the diagrams.
I suppose that "EQUAL" is ABOVE 50 %, maybe something about 55 %, at least.

CS2012.jpg (98.51 KiB) Viewed 12254 times

MFGo12.jpg (97.37 KiB) Viewed 12254 times

[pookpooi]

Damn, Lee was playing so well at the beginning.

I want AlphaGo to win, but still pretty shocked.

I thought AlphaGo was slightly behind for much of the game, but it appears Lee misplayed the complex position on the right side.

Michael Redmond commentating thought AlphaGo showed good timing; but he didn't like the thick play upper left, tenuki from the lower left allowing a double approach. AlphaGo found a way to take sente there (apparently messy) and move over to the right, keeping complexity alive.

Interesting "living go", which is ironic with one player being silicon.

Demis Hassabis keep his promise to you

[CutFirstThinkLater]

Congratulations AlphaGo!

Some quick thoughts after game 1 (I watched the game on and off so this is not a thorough impression):

1. Lee clearly seemed to underestimate AlphaGo when he went strait to fighting very early in the game.

2. AlphaGo's evaluation skills seemed even better than Lee. It felt like it didn't miss a single weakness in both of its own and Lee's position, and incorporated it into its strategy. For example, the well timed probe at j8 in the middle game was quite impressive in itself, and the follow up after that - allowing Lee to create a huge, near-territory like moyo in the lower left, while building a long but useless-looking (at least territory wise) wall in the right side - looked a bit unreasonable at first, but it opened up so many possibilities in the right side so was not that bad after all (in fact, it was there where it was able to gain the lead in the game).

Its use of aji was very good throughout the game - a hint of how much it regards the value of hidden potentials, which I think is a sign of extraordinarily strong intuition formed from a vast amount of experience. Lee's strategy of making the game complicated didn't work out for him - because AlphaGo's strength in chaotic fighting was just as good (if not better) as Lee's. Especially AlphaGo's move 102 seemed to shock many pros, including Lee.
It also seemed to play a lot of unexpected but surprisingly effective moves. So much for human creativity.

3. The Korean pros commenting seemed pretty clueless of how computers tend to play. For example, during the game AlphaGo made a couple of "mistakes" according to them - because it seemed to lose points locally - but in the end those "lost points" were quickly recovered somewhere else, or didn't matter in the end. Anyone who played computers a lot will know they tend to start 'losing points' once they think they're ahead - right until they win precisely by half-point. It gave the impression that AlphaGo is a lot better than pros in positional judgement.

4. I think Lee held out well overall, but (according to pros) a single mistake in the lower right corner was enough to lose the game. If that's all it takes to lose, then AlphaGo's strength is scary indeed. I expect a 5-0 victory for the bot.

[CutFirstThinkLater]

Lee's comments after game 1:

Initial feelings after the game:
"I was so surprised because I didn't think I would lose. About the game, I messed up in the fuseki stage and wasn't able to recover throughout the game. I didn't imagine its play would be this perfect. Very surprised indeed."

About Deepmind Team:
"The Deepmind Team expressed their thanks for me playing the game, but I take my hats off to them for creating such an excellent program."

What surprised him the most:
"I was surprised by 2 things. First, its ability to play such smoothly in the opening stage, because I thought the A.I. will have a hard time when so many things are wide open. Second, just when I think it's difficult for both sides, it plays 'shaking moves (勝負手; gambling moves that try to turn the game around by making it chaotic when you think you're losing)'. Moves that humans normally don't play. It's a Go related thing so I can't explain clearly (He seems to be talking about AlphaGo's ability to play unexpected but sharp, 'gambling' type of moves. I think he particularly has move 102 in mind)."

About the following matches:
"It's true the result is a bit of shock. But I enjoyed the game very much, and really looking forward to the remaining games as well. I'm not sure what the outcome will be. Today I messed up to much in the opening. If I am more careful next time, it may go well for me."
"If AlphaGo wasn't able to play aforementioned 'shaking moves' I would have said I have a better chance to win, but since it apparently can, I'd say it's a 50:50." (I think he means he underestimated AlphaGo before the match, but now sees it at least as his equal.)

Source: http://sports.media.daum.net/sports/gen ... 9174608093
(in Korean)

(edited for wrong grammer)

[Charles Matthews]

Demis Hassabis keep his promise to you

Actually, going back to the 1990s, he told me he would be a "good soldier" for go. But who would have guessed this?

[pookpooi]

Demis Hassabis keep his promise to you

Actually, going back to the 1990s, he told me he would be a "good soldier" for go. But who would have guessed this?

A little bit sad that this side story is hidden within only lifein19 it deserved to be in the news as well

[Charles Matthews]

Demis Hassabis keep his promise to you

Actually, going back to the 1990s, he told me he would be a "good soldier" for go. But who would have guessed this?

A little bit sad that this side story is hidden within only lifein19 it deserved to be in the news as well

Suits me if it is known to those who will understand the context.

[sybob]

A little bit sad that this side story is hidden within only lifein19 it deserved to be in the news as well

So, let's cherish Lifein19.

[trout]

2nd game;

AlphaGo defeated Lee Sedol by resign.

AlphaGo vs Lee Sedol 2:0

[wineandgolover]

I guess this marks a new era. Well I suppose that officially happens Friday, but the writing is on the wall.

"Game over, man!"