[Computer-go] 9x9 result in Olympiad

Hideki Kato hideki_katoh at ybb.ne.jp
Sat Aug 17 19:55:23 PDT 2013


Darren Cook: <520F71B5.5030806 at dcook.org>:
>> 9x9 tournament had 10 participants, and Zen won.
>> 1st Zen     8  -1
>Congratulations to Zen on a clean sweep of the go tournaments!


>I'm sure feeling vindicated is one of the baser emotions, but I feel it
>anyway: I remember 10 or so years ago (*) people telling me 9x9 was a
>dull game, not worthy of research effort, and I was saying, no, there is
>a lot of depth here.

Now most Japanese professionals know 9x9 is still a huge space to 
explorer, I strongly believe.

>And I claim there is still a lot of depth left to study. I notice Zen
>lost a game. When I spoke with Hideki Kato (who also shares my
>excitement at the remaining challenge of 9x9) about the loss he said it
>was a bug or difficult position, as Zen's evaluation went from
>90-something to 0 in the space of a move.

I'll study what caused this later.

>Yet this imperfect Zen beat everyone else. And this was played with a
>komi of 7, so what we should have been seeing is draws everywhere.

Unfortunately, other strongest programs on 9x9 such as CrazyStone and 
Steenvreter didn't participate and no draws happened.

>Hideki also mentioned that most pros thought they could beat the strong
>computer programs at 9x9. I think he was saying it is 19x19 they are
>worried about, because it is too big and hard to play for humans. But on
>9x9 they can read it out.  [Hideki, please correct me if I'm mis-quoting
>you or the pros there.]

Hirofumi Ohashi 6p told me after the 6 wins in a row on 9x9 last year 
that all professionals can read-out (not from the beggining, of course 
:-) on 9x9 and should have advantages against stochastic programs.  
I was pretty impressed.  Current MC Go programs have some errors 
of (at least) a few percent on L&D, ie, a dead group lives in some 
percent of the simulations.  Can we make it zero in MC framework or we 
need some other devices?


>*: I remember 15+ years ago telling someone I thought I could *solve*
>9x9 with a year of concerted effort (but unfortunately I was busy with
>19x19). Ah, the arrogance of youth :-)
>Computer-go mailing list
>Computer-go at dvandva.org
Hideki Kato <mailto:hideki_katoh at ybb.ne.jp>

More information about the Computer-go mailing list