Nihon Ki-in announce World Go Championship including Zen AI

[Uberdude]

The Nihon Ki-in have just announced a new tournament for the top human players from Japan, China and Korea plus the AI Zen for March 21-24 2017.

http://www.nihonkiin.or.jp/news/release/ai.html (In Japanese, I'm using google-translate).

Japan's player will unsurprisingly be Iyama Yuta, it's not yet clear who will play for Korea/China. The obvious choices would be their #1 rated players Park Junghwan and Ke Jie, but maybe they will have a selection tournament. Round-robin format, time limits are 3 hours with what I presume is 5 1 minute byo-yomi periods, 6.5 komi Japanese rules. Total prize fund is 50 million yen with 1st getting 30 million (about £200k or $265k), then 10/5/5. Exciting times!

[Satorian]

Another high-profile tournament is good news in principle, but I find it a bit odd that they left out Europe and the US. Even without spectacular chances of winning, it would've been nice to acknowledge those communities and include them. I optimistically assume they chose Zen as AI entrant for being the latest UEC Cup winner.

[pookpooi]

Message from Hideki Kato

"Hi all,
Nihon Kiin created a new Go tournament, "World Go Championship", which will be held in March 21st to 23rd, in Osaka, Japan.
Top three professional players from Japan, China and Korea and one Computer Go program will attend. Yuta Iyama six-crown and Zen from Japan. China and Korea are in cordination. The prize money is 30 million yen."

Nihon Kiin recommend DeepZenGo as Go AI representative, but it's not set in the stone like Japan representative: Iyama Yuta. Maybe there'll be qualify process for Go AI? Is this where AlphaGo will come down from heaven?

Edit: Nihon Kiin already invite AlphaGo but DeepMind declined due to schedule, so that's mean next AlphaGo event will be in March too? Also, they'll hold World Go Championship for at least 3 years.

[Satorian]

Until then Zen should have seen enough improvement to give the top pros a run for their money, so I'm not too disappointed that AlphaGo isn't available.

[Uberdude]

I'm sure the Zen team are working hard to iron out these wrinkles, but despite Zen19L's impressive performance on KGS there do seem to be a few weaknesses that even I at my lower skill level can identify (not that I'd be able to hold my own in a game to exploit them):
- it is relatively weaker at endgame and can throw away winning games with endgame mistakes, examples being the games with Cho and Blackie. Ke Jie advised this.
- Cho's strategy in game 3 of making a more complex aji-full and fighting endgame instead of simple points grabbing looks to be a good plan.
- complex situations like sekis with kos, or a temporary seki on one side of a semeai (e.g. a game with zenkiler), can unbalance it.
- it might not correctly evaluate unfinished semeais left on the board (e.g. Cho game 1, and *), so if you can make (multiple of!) these it might have a too high winning probability estimate and play slackly/safely thinking it is losing not enough points to change the result but actually it does.
- it doesn't know the cut fails in micro-Chinese wedge shoulder hit etc. opening and plays it out. Still managed to make an impressive fight on the outside though!
- some KGS 3d tricked it with a 6-4 joseki (the one with belly attachment and then pole connection which leads to the triple hane). Maybe also vulnerable to some taisha tricks? It seems to avoid the avalanche in the games I've seen.
- it has some anti mirror-go strategy but not sure how effective against strong mirror play instead of the weak KGS troll mirror-play.
- still makes clear local mistakes (e.g. losing 2 points in that semeai with Cho, Bill's hypothesis looks plausible to me). Also the wasting ko threat stuff.

Maybe we will see Iyama Yuta and others testing it out on KGS? Or maybe they'll get a private test version. Anyway, I think the Zen team have shown a brave and commendable attitude in making it publicly available and thus vulnerable to weakness probing rather than shrouded in secrecy a-la AlphaGo. It might hurt their winning chances in the short-term, but hopefully in the long run they can address these problems and make a stronger and more robust bot.

*Edit: latest win for Blackie featured a semeai of a big group of Blackie with no eye beating Zen's one-eyed group in a critical fight. It resigned in late yose when a few more moves of the semeai were played out, so perhaps didn't realise it was dead before.

[John Fairbairn]

The most interesting thing about the announcement for me was the list of sponsors: long, varied and with some lapsed names resurfacing (NTT DoCoMo, Mitsui/Sumitomo Card, Daiwa Securities, Hanshin Dentetsu, Nihon Keizai Shinbun, Hanshin Hotels, JAL and Dowango). I sense the hand here of the new Nihon Ki-in Chairman, Dan Hiroaki. His previous post was as head of the Ministry of Post and Telecommunications.

So he's had an impressive start, but maybe this is just the beginning. Would it be too fanciful to surmise that he may be trying to position Japan as the leading oriental nation when it comes to computer go? And would that make go "cool" again in Japan?

The Osaka connection is obvious because of Iyama, and the tournament will be played there. It will be a three-round round robin, which is a format that throws up a lot of ties. There's no mention of a tie-break format. I expect there will be no tie-break, especially as any artificial tie-break here would grossly unfair given the huge difference between 1st and 2nd prize.

I expect the people busiest from now to March will be a huge squad of Chinese coaches assembled to analyse the games of Iyama, Zen and whoever the Korean guy is. My impression is that all the computers seem to prefer a centre strategy, or at least don't wimp out of it the way humans do. It may be that Shin Fuseki needs to be revisited.

Note the 3-hour format - long enough to make the endgame a big factor. That might help the humans.

One picky thing: World Go Championship might be the literal translation but the meaning is just International Go Tournament.

[idontgetit]

Message from Hideki Kato

"Hi all,
Nihon Kiin created a new Go tournament, "World Go Championship", which will be held in March 21st to 23rd, in Osaka, Japan.
Top three professional players from Japan, China and Korea and one Computer Go program will attend. Yuta Iyama six-crown and Zen from Japan. China and Korea are in cordination. The prize money is 30 million yen."

Nihon Kiin recommend DeepZenGo as Go AI representative, but it's not set in the stone like Japan representative: Iyama Yuta. Maybe there'll be qualify process for Go AI? Is this where AlphaGo will come down from heaven?

Edit: Nihon Kiin already invite AlphaGo but DeepMind declined due to schedule, so that's mean next AlphaGo event will be in March too? Also, they'll hold World Go Championship for at least 3 years.

Due to "schedule" doesn't even make sense because unlike Ke Jie or Park, you can have many copies of AlphaGo. It can play in different events at the same time.

[Kirby]

Due to "schedule" doesn't even make sense because unlike Ke Jie or Park, you can have many copies of AlphaGo. It can play in different events at the same time.

Publicity is a factor, too. If Google wants to sponsor big events to get a lot of press, it makes more sense to spread them out.

[Satorian]

Due to "schedule" doesn't even make sense because unlike Ke Jie or Park, you can have many copies of AlphaGo. It can play in different events at the same time.

That would only work if you also had many copies of the hardware, you know, thousands of CPUs or GPUs or TPUs or whatever it is.

I would also assume that they need to book the processing power somewhat in advance, given its scale. Then again that might be different next year when AlphaGo resurfaces and might be restricted to more modest hardware similar to Zen's current incarnation.

[Uberdude]

Due to "schedule" doesn't even make sense because unlike Ke Jie or Park, you can have many copies of AlphaGo. It can play in different events at the same time.

Declining "due to schedule" can also be a polite way of saying "no thanks, we don't want to". Also given the care DeepMind have taken of AlphaGo's public appearances so far, they will want to closely manage the media and press situation. Demis Hassabis, David Silver, Aja Huang et al can't give post-match press interviews at the same time. Plus there are all the technical backend support engineers keeping AlphaGo running smoothly (see the Lee Sedol broadcasts).

[idontgetit]

Due to "schedule" doesn't even make sense because unlike Ke Jie or Park, you can have many copies of AlphaGo. It can play in different events at the same time.

Declining "due to schedule" can also be a polite way of saying "no thanks, we don't want to".

Right. What I was trying to say, is that if they wanted to participate, they would be able to, even if they have something else planned in March. Declining simply means they don't want to participate, it doesn't mean that they have something else planned at the same time.

[xiayun]

The Osaka connection is obvious because of Iyama, and the tournament will be played there. It will be a three-round round robin, which is a format that throws up a lot of ties. There's no mention of a tie-break format. I expect there will be no tie-break, especially as any artificial tie-break here would grossly unfair given the huge difference between 1st and 2nd prize.

From the sina article, it mentioned any ties after the round-robin will be decided by extra games, although it doesn't exactly say the format if we have a 3-person tie.

[Uberdude]

Following on from my post about weaknesses of Zen, in this one I collected a few of its losses, 2 blunders in the first game (missing a tesuji, missing simple sente for life and death) and 2 from being out-fought (which is rare to see now!).

[Uberdude]

There's no mention of a tie-break format. I expect there will be no tie-break, especially as any artificial tie-break here would grossly unfair given the huge difference between 1st and 2nd prize.

From the sina article, it mentioned any ties after the round-robin will be decided by extra games, although it doesn't exactly say the format if we have a 3-person tie.

Doesn't the original Nihon Ki-in announcement say something about tie-breaks? The machine-translation was such a mess I couldn't understand it clearly but something like main rounds were 21st-23rd with gold playoff on 24th, and "In 3 round round-robin league, conducting a championship playoff in the case of equal number of wins" from

総当りリーグ3回戦で、勝数同数の場合はプレーオフ（優勝決定戦）を実施 持ち時間3時間、残り5分から秒読み 対局開始は午前10時30分、休憩時間は設けず終局まで打ち継ぐ 総互先（黒番6目半コミ出し）日本ルール

[ez4u]

As xiayun wrote, it only mentions that there will be playoff but no further details.