https://www.youtube.com/watch?v=l-GsfyVCBu0

Lee Sedol takes white, AlphaGo takes black.

The AGA stream doesn't seem to be up yet.

AGA stream expects to start one hour in.

_________________
- Brady
Want to see videos of low-dan mistakes and what to learn from them? Brady's Blunders

And DeepMind took the second game, also by resignation though this time both were in extra time periods at the end.

I couldn't catch it from the stream, but does the person who's sitting in for AlphaGo at the board also input Lee Sedol's moves for AlphaGo to analyze? Or does someone else handle that?

Also, does anyone know if the agreement stipulates if the DeepMind team can modify AlphaGo between matches or not?

I heard something like it's learning by itself all the time, without direct human interfere. But don't quote me yet, I think after the match end everything will become clear

They are using the same version of AlphaGo for the entire match. One of the devs said this when interviewed near the start of the game 2 DeepMind commentary.

So, what are your thoughts about game 2? When I caught the stream (maybe around move 150), Kim 9p on the AGA stream was saying Lee had not made mistakes, apart maybe not starting a ko at the top. Kind of scary if this is the case and he still lost.

And apparently Alphago pulled a shoulder hit on a 4th line stone. I saw some comments mentioning a move that made Gu Li and other Chinese pros laugh, calling it a 4-5p level move. Is this the one?

If so, maybe this shows how important and misunderstood the center still is...

Aja Huang (5 dan) is the person from the DeepMind team who's playing the moves. I believe that is all he's doing, as apparently it's quite stressful enough just to do that.

I don't know who's pressing the clocks either, but Lee isn't pressing his.

Other commentators and Lee himself haven't identified any real errors he made so far, unlike the first game.

I think Gu Li and others may have to re-evaluate their criticism of that move though...

Imagine if he realized mid game that he actually misplaced one stone :O

Takemiya 9p played 6th line reductions forcing 4th line territory or 5-5 against 4-4 when building a huge center, so IMO he would approve the move.

I bet! Imagine having to sit for 5 hours long and having to promptly respond to every move within seconds.

I've been impressed with Aja's poker face, I don't think I could resist a slight smile of pride if my baby played such brilliant moves as the r10 invasion in the first game or the p10 shoulder hit in the second.

Actually, privately a few pros have expressed to me that Aja could be very distracting for Lee, because his emotions completely misread the board position.

E.g., in game 1, Aja was very excited to play the 3-3 point in the upper left, really smacks the stone down, although the move itself is not a good one. He likened it to having someone say something brilliant while their body language told you they were a simpleton, it would be very disorienting. Said he would certainly prefer to play against a robot arm, or just play against a computer -- far less distracting.

Interesting point seigenblues. Btw, why is the 3-3 not good? Is it the wrong local move (iron pillar normal right?) or the wrong timing? I was thinking the 3-3 might leave Black with fewer ko threats, and also by being closer to c14 it means Black's moves on the left side are slightly less powerful and the yose there will be a bit worse. Maybe that was why Lee didn't do anything on the left side on move 147 which surprised me. I would have wanted to play c13 but maybe that is an overplay, and the 3-3 vs iron pillar certainly makes a difference to the fighting if white doesn't answer c13 but pincers and tries to kill the whole resulting group. The idea might not be to take gote to reduce that area but to play some 'screening kikashi' so that white cannot so cleanly take that area with the 2 space extension as in the game. So if the 3-3 makes that plan less effective AlphaGo is a genius , or am I reading too much into it? I think AlphaGo could well be capable of such long-term long-distance interaction planning (improving on the poor use/understanding of aji Myungwan Kim highlighted from October).

I think the move was good. In his commentary, Redmond anticipated the iron pillar, but did not make anything of the slightly different placement. I believe he also thought the timing was correct.

Yeah, I thought the 3-3 was good, and quite probably actually better than the iron pillar due to the reasons above.

From Facebook, re: move 37



<edit>
OK, Lee left the room before it was played: But imagine how strong a move must be if it does that to him even before it was played?!
</edit>

_________________
“The only difference between me and a madman is that I’m not mad.” — Salvador Dali ★ Play a slooooow correspondence game with me on OGS?

3-3 isn't bad and it isn't great, it's totally normal. the idea was that he was too excited to play it, when really it's just a very ordinary move...

In this translated article from Li Zhe 6p (the one studying philosophy at university) he said 3-3 was good, and better than the iron pillar (due to some reading that he didn't share):

https://massgoblog.wordpress.com/2016/0 ... -weakness/ (a bit over half way down under "HAS THE AI READ THROUGH?")