Go has fallen to computer domination?

[Isaac]

5 games have been played between a new go program called Alphago (from google deepmind) against the top European champion. No handicap stone. The program won the 5 games.
Sources: http://www.nature.com/nature/journal/v5 ... 16961.html
Games can be downloaded at http://www.usgo.org/news/2016/01/alphag ... i-advance/.

In March it is said that Lee Sedol (regarded as the strongest go player in the world by many) will play Alphago.

What's next?

[PK]

It's a great achievement for computer go, but the difference between top European players and top Asians is probably like a difference between an IM and a GM. The best Polish player, Mateusz Surma, who has gained the newly created title of "Europaean professional" recently (there are only 4 or 6 of them) was far from the best in the Korean go school.

This being said, I looked at the game and its flow was surprisingly natural.

[matthewlai]

5 games have been played between a new go program called Alphago (from google deepmind) against the top European champion. No handicap stone. The program won the 5 games.
Sources: http://www.nature.com/nature/journal/v5 ... 16961.html
Games can be downloaded at http://www.usgo.org/news/2016/01/alphag ... i-advance/.

In March it is said that Lee Sedol (regarded as the strongest go player in the world by many) will play Alphago.

What's next?

Videos available on DeepMind site:
http://deepmind.com/alpha-go.html

[chetday]

And another story on this at http://www.wired.com/2016/01/in-a-huge- ... ame-of-go/

[towforce]

It is very important to note that Alphago is a knowledge-based system: in contrast to IBM's Deep Blue, Google have used deep neural networks.

Firstly, they learned from a database of existing games, then the system played millions of games against itself, steadily improving.

Human level intelligence is quickly coming into sight...

[Laskos]

Fan Hui is 2p professional. 5-0 suggests at least 4p for AlphaGo. Also, it beats normal MC programs like Crazy Stone (say 5 dan on i7) by 99.8%, which suggests 8-10 more stones, also above 4p. This thing runs on 170 GPU cards and 1,200 standard processors, and assuming quadrupling for a stone, it has 5 stones advantage over normal hardware on KGS (say a cluster of 24 regular cores). Lee Sedol is above regular 9p. Let's see.

[Bloodbane]

Also, it beats normal MC programs like Crazy Stone (say 5 dan on i7) by 99.8%, which suggests 8-10 more stones, also above 4p.

This was the most shocking part in my opinion.

[Laskos]

Also, it beats normal MC programs like Crazy Stone (say 5 dan on i7) by 99.8%, which suggests 8-10 more stones, also above 4p.

This was the most shocking part in my opinion.

Yes, that's amazing.

I have a better estimate, using this site: http://senseis.xmp.net/?EGFRatingSystem . 99.8% against 6d Crazy Stone seems close to 500 ELO points (or whatever it's called). It means close to 5 stones. If 1p is ~ 8d , then 5 stones above Crazy Stone is about 4p. It seems to me AlphaGo will lose to Lee Sedol (9p).

[PK]

IIRC professional go grades are not one handicap stone apart. More like 1/3 of a stone.

[Laskos]

IIRC professional go grades are not one handicap stone apart. More like 1/3 of a stone.

Thanks, I think that's correct. Then if AlphaGo is 5 stones above Crazy Stone, it should be close to 7-8p.

Wonder how much of this 5 stone advantage is due to those 170 GPU cards.