World #1 Go Player Ke Jie accepts Google Alpha Go Match..

[supersharp77]

Go to have the same fate as human chess as Google "Deep Mind Team"
moves "secret artificial intelligence project" forward...but to what end?

http://arstechnica.com/information-tech ... h-details/

[Dirt]

Reading the link, I see this announcement is very preliminary. It may not happen.

I'm not sure what Google hopes to achieve, but we need more games against top players. One match may have been a fluke, especially since it was 4-1 and not 5-0.

[PK]

Yes, more games are needed. But since a single game of go lasts much longer than a game of chess (in terms of the number of moves), we should translate AlphaGo's score to 80% out of 20 games.

[whereagles]

This is great news. Hope the match can be organized.

Probably the last of AlphaGo's performance... afterwards it will probably be dismantled and reassembled to solve some traffic engineering problem

[Dirt]

Yes, more games are needed. But since a single game of go lasts much longer than a game of chess (in terms of the number of moves), we should translate AlphaGo's score to 80% out of 20 games.

No, I don't think so. It's not important, though.

[Harvey Williamson]

Go to have the same fate as human chess as Google "Deep Mind Team"
moves "secret artificial intelligence project" forward...but to what end?

http://arstechnica.com/information-tech ... h-details/

Really interesting and nice that we can discuss this match in the main section.

[Laskos]

Reading the link, I see this announcement is very preliminary. It may not happen.

I'm not sure what Google hopes to achieve, but we need more games against top players. One match may have been a fluke, especially since it was 4-1 and not 5-0.

About that 4-1, and the single loss of AlphaGo, it seems the mistake AlphaGo committed in game four: Black 79. K10 aftetr the "divine move" of Lee Sedol as White 78. L11 is not a fundamental flaw of MCTS approach and pruning as some said. I got the latest "Crazy Stone Deep Learning" which attempts to go on the steps of AlphaGo, and it finds the apparently correct answer 79. L10 which was suggested by top Go pros:



Then the game proceeds as following (played 2 long time control games from 78. L11 "divine move" of Lee Sedol as White) with the Black narrow wins:



Therefore, the most flagrant flaws in AlphaGo can be corrected, and for the next march with Ke Jie I expect a 5-0 result.

[Laskos]

Reading the link, I see this announcement is very preliminary. It may not happen.

I'm not sure what Google hopes to achieve, but we need more games against top players. One match may have been a fluke, especially since it was 4-1 and not 5-0.

About that 4-1, and the single loss of AlphaGo, it seems the mistake AlphaGo committed in game four: Black 79. K10 aftetr the "divine move" of Lee Sedol as White 78. L11 is not a fundamental flaw of MCTS approach and pruning as some said. I got the latest "Crazy Stone Deep Learning" which attempts to go on the steps of AlphaGo, and it finds the apparently correct answer 79. L10 which was suggested by top Go pros:



Then the game proceeds as following (played 2 long time control games from 78. L11 "divine move" of Lee Sedol as White) with the Black narrow wins:



Therefore, the most flagrant flaws in AlphaGo can be corrected, and for the next march with Ke Jie I expect a 5-0 result.

It seems to turn out that tesuji 78. L11 "divine move" of Lee Sedol (White) is really special, and both AlphaGo and CS Deep Learning miss it, the main mistake being earlier Black's (AlphaGo) 77. before the "divine move", 79 being a smaller mistake. Playing 10 long time control games from 77, in all of them CS misses as White the "divine move" 78 and the ensuing tsumego (life and death problem), and the average of the outcomes is clear victory of Black by 13.5 points. If the "divine move" is made, then 79. L10 still wins for Black, but by a narrower margin of 4.5 points. Missing L10 too, and playing 79. K10 as AlphaGo did, loses for Black by 2.5 points on average. If all this means something (CS Deep Learning is still much weaker than AlphaGo and Lee Sedol), then Black should play something different at 77 to prevent the "divine move" and keep the serious advantage, and it's probably a good test position for AlphaGo, and generally all strong programs.

[Laskos]

It seems to turn out that tesuji 78. L11 "divine move" of Lee Sedol (White) is really special, and both AlphaGo and CS Deep Learning miss it, the main mistake being earlier Black's (AlphaGo) 77. before the "divine move", 79 being a smaller mistake. Playing 10 long time control games from 77, in all of them CS misses as White the "divine move" 78 and the ensuing tsumego (life and death problem), and the average of the outcomes is clear victory of Black by 13.5 points. If the "divine move" is made, then 79. L10 still wins for Black, but by a narrower margin of 4.5 points. Missing L10 too, and playing 79. K10 as AlphaGo did, loses for Black by 2.5 points on average. If all this means something (CS Deep Learning is still much weaker than AlphaGo and Lee Sedol), then Black should play something different at 77 to prevent the "divine move" and keep the serious advantage, and it's probably a good test position for AlphaGo, and generally all strong programs.

I add the diagram up to Black (AlphaGo) move 79. K10? with numbered moves:

[Dirt]

The match is now scheduled to take place from May 23 to May 27 of 2017. Only three games are in the main match but there will also be matches between professionals alternating moves with computers as well as other activities.