Takemiya versus Zen19 on March 17

yoyoma · **#21**

Zen won the second game with 4 stone handi against Takemiya Masaki 9p. Edit: I fixed a mistake, I now get B+20, same as they did.

EdLee · **#22**

Post-game commentary:

illluck · **#23**

My own transcriptions, the first game I probably have quite a few major order errors (mostly related to the ko on top left). The second should mostly be fine, though in both games the endgame definitely have minor errors.

EdLee · **#24**

Bye bye...

Javaness2 · **#25**

That's a pretty scary, seeing a pro being taken down on 4 stones by Zen. I don't think I would have a chance on 4 stones against the mighty T.

uPWarrior · **#26**

They should have kept going to 3H!

shapenaji · **#27**

Javaness2 wrote:

That's a pretty scary, seeing a pro being taken down on 4 stones by Zen. I don't think I would have a chance on 4 stones against the mighty T.

The thing is, Zen's positional awareness is so good that while it may not see the consequence of some of it's aji, it makes very good (and consistent) judgement calls. It's style is very well suited to handi go.

yoyoma · **#28**

Official SGFs:

Quote:

On Sat, Mar 17, 2012 at 12:29 PM, Hiroshi Yamashita <yss@bd.mbn.or.jp> wrote:
All SGF are now available from organizer page.

Takemiya 9p vs Zen 5 handicaps, Zen won
http://entcog.c.ooco.jp/entcog/event/20 ... 9x19-1.sgf
Takemiya 9p vs Zen 4 handicaps, Zen won
http://entcog.c.ooco.jp/entcog/event/20 ... 9x19-2.sgf

Ohashi 5p vs Zen 9x9 Komi is 7.0, Ohashi won
http://entcog.c.ooco.jp/entcog/event/20120317/9x9-1.sgf
Ohashi 5p vs Zen 9x9 Komi is 7.0, Zen won
http://entcog.c.ooco.jp/entcog/event/20120317/9x9-2.sgf

Cognitive Science and Entertainment
http://entcog.c.ooco.jp/entcog/event/20 ... vscom.html

Regards,
Hiroshi Yamashita

topazg · **#29**

Wow, that's truly impressive. I'm going to have to eat my hat, Computer Go is improving more quickly at the mid dan ranks than I suspected would be possible before at least 2020.

Pippen · **#30**

Now, we got a problem chess has for some years now: players that cheat in running a programm besides their play on the internet. I don't see a effective solution other than blitz games.

I also think one should match a pro vs. Zen in 2013 to play an even game with a price tag of 10K, sponsored by some firm. I think it's time that Zen can legitimately look to beat pro's, esp. in rapid games.

illluck · **#31**

I think Zen is nowhere near enough to beat pros even. I also suspect that Takemiya could give 5 stones or even more if he didn't care about pretty game records :p

jts · **#32**

Pippen wrote:

Now, we got a problem chess has for some years now: players that cheat in running a programm besides their play on the internet. I don't see a effective solution other than blitz games.

Zen is using a lot of power to deliver these results, isn't it? The Monte Carlo bots have made huge strides (well, at least Zen has!). That has forced a lot of people to reconsider their guesstimates about how strong Zen is, how quickly it's improving, and things like that. But how strong would Zen be if it were running on my laptop?

speedchase · **#33**

I am pretty impressed, however I think people are expecting result too quickly. MCTS has a limit because of size of the Branching factor. Also I'd be interested to know what the time limits were like for these games.

uPWarrior · **#34**

jts wrote:

Pippen wrote:

Now, we got a problem chess has for some years now: players that cheat in running a programm besides their play on the internet. I don't see a effective solution other than blitz games.

Zen is using a lot of power to deliver these results, isn't it? The Monte Carlo bots have made huge strides (well, at least Zen has!). That has forced a lot of people to reconsider their guesstimates about how strong Zen is, how quickly it's improving, and things like that. But how strong would Zen be if it were running on my laptop?

Some versions of Zen are running on standard hardware (though upper range, nonetheless) or small clusters (4-8 PCs).
Given the natural hardware evolution (hail Moore's law!), that's how powerful new laptops will be in two to four years.

daal · **#35**

I would really like to read Takemiya's comments on the game. Considering that both Takemiya and Zen are known for their use of influence, I wonder what he thinks of Zen's unorthodox style, and whether he plans on studying the strategy that the computer employs.

Philip Traum · **#36**

The article on GoGameGuru: http://gogameguru.com/zen-computer-go-program-beats-takemiya-masaki-4-stones/

daniel_the_smith · **#37**

speedchase wrote:

MCTS has a limit because of size of the Branching factor.

Hm? MCTS is a way of minimizing the effect of huge branching factors...

mschribr · **#38**

Who was making the moves on the board for Zen?

speedchase · **#39**

daniel_the_smith wrote:

Hm? MCTS is a way of minimizing the effect of huge branching factors...

Sorry, what I was trying to say was that MCTS is only useful to an extent because the huge branching factor prevents MCTS from seeing all the moves. to play like a pro, I think an MCTS bot will have to run so many simulations that it seems all of the possible moves many moves deep, and then still run a considerable number of moves for each possible branch

EdLee · **#40**

mschribr wrote:

Who was making the moves on the board for Zen?

I think Mr. Kato Hideki of team DeepZen.

Takemiya versus Zen19 on March 17

Who is online