AlphaGo vs Ke Jie Game discussion

Uberdude · #1

There are several threads on the Computer Go subforum about the current match (plus rather a lot of drivel), so here's a thread for discussing just the games themselves (off-topic posts will be removed).

Game 1 was won by AlphaGo as white by 0.5 points, but it was further ahead and played a soft/safe endgame. A fairly peaceful game, with Ke Jie trying Master's early 3-3 against itself, see my comments.
DeepMind stream: https://youtu.be/Z-HL5nppBnM
AGA stream: https://youtu.be/rFNgHXjIJo4
BIBA stream: https://youtu.be/xY6m4pIWOqk

Game 2 was won by AlphaGo as black by resign. Amusingly it played an opening 3-3, are we sure it's not sentient and trolling Ke Jie to show him how to use a 3-3?! [In the 60 Master games it was in the position before move 3 here 3 times, and played the facing 3-4 twice and twisted 3-4 once] It was a much more complicated fighting game with many weak groups. In post-match press conference Demis said (according to AlphaGo) it was even for the first 50 moves, and very close for 50-100, and Ke Jie explained his chest holding as nerves and excitement as he thought he could see some chances in the complications but missed them. In post-mortem with Tang Weixing it seems he regretted his ko threat and wanted to cover in the centre, it does seem his threat was not enough and from then on AlphaGo simply had too much territory.
DeepMind stream: https://youtu.be/1U1p4Mwis60
AGA stream: none, game over too fast (3 hours)
BIBA stream: https://youtu.be/dkwr0vDNJ1A

For Game 3 Ke Jie requested to play white, acknowledging that machine view is white is slightly better with big komi (and famously he didn't lose as white for almost a year), hoping to create an exciting game with many capturing races. Demis agreed.

pookpooi · #2

First game: 4 hours and 20 minutes from timeclock start to beginning of counting (timeclock stopped)
Second game: 3 hours and 6 minutes from timeclock start to resign
(Both game I count from DeepMind length of video instead of adding up time use from timeclock)

While Ke Jie play slower than AlphaGo, I think he's still considered fast, not entering byoyomi period.
This may be due to the fact that he has been preparing very hard, especially the early part (opening to mid) of the game. So that's there's not many things to think about (even though Lee Sedol encourage him to use more time)

In second game, from nicovideo stream, DeepZenGo also think that Ke Jie play very well. In fact, in the early part, winrate for Ke Jie shoot up as high as 60% (this may be due to high komi, but still impressive), not until after the first 100 moves that DeepZenGo begin to favor AlphaGo. This is in contrast to the first match that DeepZenGo favor AlphaGo since the beginning and progressively loving it more and more, 80+% midgame and close to 100% at end game.

I only observe the difference in style from this version and Master in new year season, but sadly no difference (from these two games). Maybe I need to see more game but can someone spot the difference as of now? 3-3 opening is not so 'new' to AlphaGo' this time, avoiding (unnecessary) ko seems normal since like the lee sedol version, avoiding fight? hard to judge.

Also, can I discussed pair go and 5vs1 match here? (despite the thread name is AlphaGo vs. Ke Jie)

Uberdude · #3

Thanks for the timing/win-rate info. Ke Jie is known for playing fast vs humans, but I think Lee Sedol's suggestion to slow down could be good: for example in game 1 attachment at L5 he played in just a few minutes. It looks kinda cool so wins kudos points, but as it was such an important decision as to how the lower side continued it seems more planning could have helped: based on the continuation and Ke's body language it seems like he got what he wanted: some sabaki in white's area, but AlphaGo was happy to go along with it as the loss of the 2 stones was actually quite big (and more than initial judgement of Michael Redmond) so not a satisfactory result. Beforehand Myungwan was saying can't give up those stones and looking at other ways of running out (which didn't seem great either tbh). Quite a lot of game 1 had that feeling to me, AG giving Ke what he wants (e.g. trade on left upper and left middle, also cut at top right) because its judgement was the result is good enough for AG to win. Game 2 though it had to be fiercer and hence the fireworks.

pookpooi wrote:

Also, can I discussed pair go and 5vs1 match here? (despite the thread name is AlphaGo vs. Ke Jie)

No thanks, new thread.

Uberdude · #4

One thing I found interesting near the end of the game (I didn't watch the commentary for the beginning/middle yet and it's rather complicated!) is how AlphaGo indirectly settled the marked left side stones. Michael Redmond had been talking for a while about how when the fighting resolves he expected black to play something like below which makes a base and also aims at some weakness of white at a/b:

Click Here To Show Diagram Code: [go]$$ Bc Possible move for black on left when all the lower through middle to upper fighting finishes. $$ --------------------------------------- $$ | . . . . . . . . . . . . . . . . . . . | $$ | . . . . . . . . . . . O O . X X X . . | $$ | . . . . X . . . . X . X O X O X O O . | $$ | . . O , . . . . . , . X O X O O X O . | $$ | . . . . . . . . . . . O . X O X . . . | $$ | . O . O . . . . . . . . . X O O O . . | $$ | . # O . . . . . . O . X . . X . . . . | $$ | . . # . # . . . . . . . O O O O . . . | $$ | . . . . . . . . . . X X O X X . . . . | $$ | . . 1 , . a O . O , . . X O . X . . . | $$ | . . . . O . . . . O O X . . O X . . . | $$ | . . O b . . X . . O X X . O . X . . . | $$ | . . . . O . . . . . O X . . O X . . . | $$ | . . X X X X . X . . O X . . O X . . . | $$ | . O X O . . X O O O . X O X O O X . . | $$ | . X . X O O X O X O O X . O X X X . . | $$ | . O X O . O O X X O X O O . O X . . . | $$ | . . . . . X X . . X X . . . . O X . . | $$ | . . . . . . . . . . . . . . . . . . . | $$ ---------------------------------------[/go]

However, when the fighting had settled white had just played 1 below. Black's approach at the top left is now weaker so it makes sense to want to do something with it, and with the corner attach and bulge AG ended up sacrificing it in exchange for settling the left in style.

Click Here To Show Diagram Code: [go]$$W Moves 146 to 155 $$ --------------------------------------- $$ | . . . . . . . . . . . . . . . . . . . | $$ | . . 5 . . . . . O O . O O . X X X . . | $$ | . 3 2 7 X . . 1 X X . X O X O X O O . | $$ | . . O 4 . . . . . , . X O X O O X O . | $$ | . 0 6 . . . . . . . X O O X O X . . . | $$ | . O 8 O . . . . . X . . . X O O O . . | $$ | . X O 9 . . . . . O X X . . X . . . . | $$ | . . X . X . . . . . . . O O O O . . . | $$ | . . . . . . . . . . X X O X X . . . . | $$ | . . . , . . O . O , . . X O . X . . . | $$ | . . . . O . . . . O O X . . O X . . . | $$ | . . O . . . X . . O X X . O . X . . . | $$ | . . . . O . . . . . O X . . O X . . . | $$ | . . X X X X O X O X O X . . O X . . . | $$ | . O X O . O . O O O . X O X O O X . . | $$ | . X X X O O . O X O O X . O X X X . . | $$ | . O X O . O O X X O X O O . O X . . . | $$ | . . . . X X X . . X X . . . . O X . . | $$ | . . . . . . . . . . . . . . . . . . . | $$ ---------------------------------------[/go]

Ke Jie resigned after 10 above, but if he continued maybe it could continue as below with black settling on the left with nice profit from capturing the marked white stone, and still aji of a if white extends at 3. And maybe black could even connect for 2 as the corner has troubles, there's a crane's nest incoming if white tries to cut on the 2nd line.

Click Here To Show Diagram Code: [go]$$W $$ --------------------------------------- $$ | . . . . . . . . . . . . . . . . . . . | $$ | . . O . . . . . O O . O O . X X X . . | $$ | . O . O X . . O X X . X O X O X O O . | $$ | . . O X a . . 3 . , . X O X O O X O . | $$ | . X X 1 . . . . . . X O O X O X . . . | $$ | . @ X O . . . . . X . . . X O O O . . | $$ | . X O O . . . . . O X X . . X . . . . | $$ | . . X . X . . . . . . . O O O O . . . | $$ | . 2 . . . . . . . . X X O X X . . . . | $$ | . . . , . . O . O , . . X O . X . . . | $$ | . . . . O . . . . O O X . . O X . . . | $$ | . . O . . . X . . O X X . O . X . . . | $$ | . . . . O . . . . . O X . . O X . . . | $$ | . . X X X X O X O X O X . . O X . . . | $$ | . O X O . O . O O O . X O X O O X . . | $$ | . X X X O O . O X O O X . O X X X . . | $$ | . O X O . O O X X O X O O . O X . . . | $$ | . . . . X X X . . X X . . . . O X . . | $$ | . . . . . . . . . . . . . . . . . . . | $$ ---------------------------------------[/go]

AnthonyC · #5

I don't understand why Ke Jie didn't choose a different ko threat for movie 136 (F5). Both H2 or J2 seems bigger?

jeromie · #6

Game 2 was fascinating. Ke Jie played so well in the opening, and his approach to dealing with his weak stones was really interesting. It looks like he was intentionally building a board with many unsettled groups and lots of complicated fighting. Is there a consensus on exactly where he went wrong?

I presume that the first error came in how he dealt with the bottom / lower left corner, both because the game became visibly hard for white after that point and his play was "perfect" (according to AlphaGo) up to move 50 which was in the middle of the fighting on the bottom. By move 57 it looks to me like white has created two groups that are significantly weaker than black's groups on the bottom, though white using move 56 to threaten to settle his stones and/or take the corner looked like a brilliant way to keep things close. I'm obviously not strong enough to tell exactly which move was misplaced, though.

Uberdude · #7

AnthonyC wrote:

I don't understand why Ke Jie didn't choose a different ko threat for movie 136 (F5). Both H2 or J2 seems bigger?

I think a different ko threat could have been better: from his post-game review it looks like k14 would be better (answer j11 with j12 clamp, there's a throw in and ko). But j2 and h2 are bad ko threats because they increase black's eyeshape: if white wins the ko and then plays at e2 or m2 it is sente to kill the lower black group, with white simply killing at the other of e2/m2 to create the 'flying saucer' shape. In that shape depending how black tries to live white plays a different atari, but if you played atari as a ko threat then you can't kill the group. White being able to get m2 in sente (and possible m1 too) is very important for the life of the lower right white group.
P.S. from reddit discussion the Chinese commentators agreed it was a duff threat and a better one would mean the game continued to be close with maybe even some winning chance for Ke to their eyes. Not sure if Ke actually had enough threats though: it wouldn't surprise me if AlphaGo could exchange its way out to a favourable position somewhere. Hopefully we'll get some variations soon.

xiayun · #8

It's not just about that last ko threat by KJ. AlphaGo had two more ko threats at J7 and L5 as the follow-up to K6, so white needed to find 3 that are both big enough and not incurring large local loss.

It seems the best chance was to try to get into a liberty race with the middle black group using the ko as a leverage, but it's too complicated for me to calculate if that'll work or it'll even be enough assuming black captured the whole bottom-left and still having the right side.

wolfking · #9

On Fox Go Server, Tuo Jiaxi suggested that maybe Ke Jie should have started the ko fight at move 122 and reserve the real game 122 and the follow ups as ko threats.

wolfking · **#10**

After the game Tang Weixing and Shi Yue reviewed the game with Ke Jie. Shi Yue 9p explained what went wrong with Ke Jie's calculation in an interview with Sina.

At move 133, Ke Jie actually expected black to play at K7 as a ko threat instead of K6. Had black played K7, black would still have 3 ko threads locally, but white can also get 3 ko threats by first throwing in at M10 and then atari at N9, plus the push at M12. But in game play after K6 black still got 3 ko threats, and if white went ahead with his plan black would have one more ko threat at K10! So white would be short by 1 threat. This explained why Ke Jie was visibly frustrated after black K6.

fwiffo · **#11**

I can't help but feel that letting the lower right white group become so heavy was a mistake. After B89, I think he should have found a way to sacrifice it (or he should have saved it himself before then.) Granted, it would be a big sacrifice, as it was eventually separating two weak black groups. However, the left one was only weak if white can win the ko, and it was always unfavorable for him. The center black group only became weak after white added a lot of stones to his own.

This fits with the AlphaGo's assessment that the turning point came around move 100; white had just invested many more stones in that area. His attempt to make more ko threats was too costly; W100 started the ko and AlphaGo knew it was favored at that point.

IMO, this is the most exciting AlphaGo game we've seen. Ke Jie got what he wanted - he set the tone, creating a complicated fighting game. But AlphaGo had no trouble dealing with these complexities (and adding to them with its own cross-cuts and fighting moves) and was not afraid of the ko.

Uberdude · **#12**

wolfking wrote:

At move 133, Ke Jie actually expected black to play at K7 as a ko threat instead of K6. Had black played K7, black would still have 3 ko threads locally, but white can also get 3 ko threats by first throwing in at M10 and then atari at N9, plus the push at M12. But in game play after K6 black still got 3 ko threats, and if white went ahead with his plan black would have one more ko threat at K10! So white would be short by 1 threat. This explained why Ke Jie was visibly frustrated after black K6.

In diagrams just showing ko threats:

Click Here To Show Diagram Code: [go]$$Bc Black's 3 threats, 6 at 1. $$ --------------------------------------- $$ | . . . . . . . . . . . . . . . . . . . | $$ | . . . . . . . . . . . O O . X X X . . | $$ | . . . . X . . . . X . X O X O X O O . | $$ | . . O , . . . . . , . X O X O O X O . | $$ | . . . . . . . . . . . O . X O X . . . | $$ | . O . O . . . . . . . . . X O O O . . | $$ | . X O . . . . . . O X X . . X . . . . | $$ | . . X . X . . . . . . . O O O O . . . | $$ | . . . . . . . . . . X X O X X . . . . | $$ | . . . , . . O . O , . . X O . X . . . | $$ | . . . . O . . . . O O X . . O X . . . | $$ | . . O . . . X . . O X X . O . X . . . | $$ | . . . . O . . . 3 4 O X . . O X . . . | $$ | . . X X X X . X 2 1 O X . . O X . . . | $$ | . O X O . . X O O O 5 X O X O O X . . | $$ | . X O , O O X O X O O X . O X X X . . | $$ | . O X O . O O X X O X O O . O X . . . | $$ | . . . . . X X . . X X . . . . O X . . | $$ | . . . . . . . . . . . . . . . . . . . | $$ ---------------------------------------[/go]

Click Here To Show Diagram Code: [go]$$Wc White's 3 threats, 4 at 1 $$ --------------------------------------- $$ | . . . . . . . . . . . . . . . . . . . | $$ | . . . . . . . . . . . O O . X X X . . | $$ | . . . . X . . . . X . X O X O X O O . | $$ | . . O , . . . . . , . X O X O O X O . | $$ | . . . . . . . . . . . O . X O X . . . | $$ | . O . O . . . . . . . . . X O O O . . | $$ | . X O . . . . . . O X X . . X . . . . | $$ | . . X . X . . . . . 6 5 O O O O . . . | $$ | . . . . . . . . . . X X O X X . . . . | $$ | . . . , . . O . O , 2 1 X O . X . . . | $$ | . . . . O . . . . O O X 3 . O X . . . | $$ | . . O . . . X . . O X X . O . X . . . | $$ | . . . . O . . . X O O X . . O X . . . | $$ | . . X X X X . X O O O X . . O X . . . | $$ | . O X O . . X O O O X X O X O O X . . | $$ | . X O , O O X O X O O X . O X X X . . | $$ | . O X O . O O X X O X O O . O X . . . | $$ | . . . . . X X . . X X . . . . O X . . | $$ | . . . . . . . . . . . . . . . . . . . | $$ ---------------------------------------[/go]

Black's extra threat afterwards that Ke Jie overlooked (empty triangle blindspot?):

Click Here To Show Diagram Code: [go]$$Bc Black's extra threat $$ --------------------------------------- $$ | . . . . . . . . . . . . . . . . . . . | $$ | . . . . . . . . . . . O O . X X X . . | $$ | . . . . X . . . . X . X O X O X O O . | $$ | . . O , . . . . . , . X O X O O X O . | $$ | . . . . . . . . . . . O . X O X . . . | $$ | . O . O . . . . . . . . . X O O O . . | $$ | . X O . . . . . . O X X . . X . . . . | $$ | . . X . X . . . . . X O O O O O . . . | $$ | . . . . . . . . . . X X O X X a . . . | $$ | . . . , . . O . O 1 X X X O . X . . . | $$ | . . . . O . . . . O O X O . O X . . . | $$ | . . O . . . X . . O X X . O . X . . . | $$ | . . . . O . . . X O O X . . O X . . . | $$ | . . X X X X . X O O O X . . O X . . . | $$ | . O X O . . X O O O X X O X O O X . . | $$ | . X O , O O X O X O O X . O X X X . . | $$ | . O X O . O O X X O X O O . O X . . . | $$ | . . . . . X X . . X X . . . . O X . . | $$ | . . . . . . . . . . . . . . . . . . . | $$ ---------------------------------------[/go]

Gu Li was also talking about the atari at a as a white threat, but it's lossy (push into bamboo is sad too) as it means black then has a ko to connect his big group that white was trying to attack.

Uberdude · **#13**

AlphaGo, playing black, won the 3rd game by resign. Ke Jie fought strongly, grabbing territory and making some sacrifices, but in the end one of his weak groups died following a trade.

mhlepore · **#14**

I only was able to catch the live stream of game 3, but just wanted to say I enjoyed Seigenblues and Michael Redmond - they were a perfect pair in my opinion.

AlphaGo vs Ke Jie Game discussion

Who is online