Who's playing in Tygem? -- It was AlphaGo!!

[clemi]

Regarding Game 55 - Chen Yaoye, there's a funny anecdote to it.
It was in fact Chen Yaoye and Meng Tailing playing both against Master (although the decision was always Chen's to make) during a live session at Weiqi TV.

Here's the first part where Chen Yaoye and Meng Tailing talk about their respective games and their feelings when waiting for Nie Weiping's game to finish :
https://www.youtube.com/watch?v=WUFZLxj_u70

And here's the backstage as the match was going on :
https://www.youtube.com/watch?v=BknZMjTpqmY

During the game, around minute 8, there is a funny part that translates like this after Master plays the keima in response to the pincer :

Meng Tailing : and what about the kosumi ? is it playable (shi bu shi qi = (literally) is it go) ?
Chen Yaoye : (muttering) no, that's not playable
Meng Tailing : (after Chen Yaoye protects the corner) and now, if white presses you down, do you play it ?
Chen Yaoye : i don't know

and later in the game (forgot where) :
Meng Tailing : (again) why not play this move ?
Chen Yaoye : no
Meng Tailing : what's your refutation ? how does white respond ?
Chen Yaoye : i don't know... doesn't feel like the right move...
Meng Tailing : (laughing) oh sh** !

[Bonobo]

German IT news site “Golem” links to this thread in L19 in an article about AlphaGo

http://www.golem.de/news/deep-mind-alph ... 25415.html

[DrStraw]

German IT news site “Golem” links to this thread in L19 in an article about AlphaGo

http://www.golem.de/news/deep-mind-alph ... 25415.html

So if you linked back to it then we have a circular link. Will the internet crash?

[xiayun]

Meng Tailing reviewed his first game against Master:

Expect him to review the rest in future episodes, although he did say he performed the best in this first game and attributed part of that to not knowing it's a computer AI yet. The exchange in the upper-right was a highlight, where he believed he got much closer after falling behind in the first 50 moves. Seeing what happened at the 33:19 mark was pretty crazy too.

[Cassandra]

According to CrazyStone's analysis, this game was one of a very few "untypical" ones of the series, as it was "undecided" until late into the game (approx. move 180).

Usually, the "point of no return" is much earlier in the games (please refer to viewtopic.php?p=215477#p215477 for diagrams).

[Cassandra]

Please find attached CS's result of its analysis with "depth=high".

[Monkey]

According to CrazyStone's analysis, this game was one of a very few "untypical" ones of the series, as it was "undecided" until late into the game (approx. move 180).

Usually, the "point of no return" is much earlier in the games (please refer to viewtopic.php?p=215477#p215477 for diagrams).

Which game are you referring to?

[Cassandra]

"This" game, which is the basis for the diagrams above, is game #9.

Please find attached a diagram that shows the averages of "P (B wins)" for AlphaGo as Black and as White.

EDIT: Adjusted the diagram following ez4u's suggestion below.

"1-White" is the mirrored curve for "AlphaGo as White". It is above "AlphaGo as Black", which can be interpreted that AlphaGo plays better with White than with Black (this was also said at the Lee-Sedol-time).

[Bonobo]

Yet another IT Web site links to this thread—this time it’s Ars Technica!

[..] Over the last week, a number of forum threads have popped up to discuss this mystery debutante who has been thrashing the world's best players. [..]

[ez4u]

"This" game, which is the basis for the diagrams above, is game #9.

Please find attached a diagram that shows the averages of "P (B wins)" for AlphaGo as Black and as White.

"1-Black" is the mirrored curve for "AlphaGo as Black". It is above "AlphaGo as White", which can be interpreted that AlphaGo plays better with White than with Black (this was also said at the Lee-Sedol-time).

I think you should invert White rather than Black to make your comparison. You would then have lines for the probability AG wins as Black versus the probability AG wins as White.

In any case, we see that the winning probability is higher playing White from the very beginning. The immediate interpretation would seem to be that komi is too large and gives White an advantage. This is consistent with many players' opinion based on human vs human play. The lines seem very consistent over the first 150+ plays. If AG were better playing White, the gap between the lines should diverge move by move. This does not seem to happen to any significant extent.

How should we interpret the changes that occur around move 165? Is it AG changing its evaluation process as the winning probability grows too large? Is it desperation appearing in the opponents' play?

Anyway, very interesting information. Well done!

[Cassandra]

Please find below the results of CS's analysis (depth="high") for some of the "longer" games.
Scaling for both axis is identical in all the diagrams.

I am not strong enough to correlate this "bot view" with the judgement of a strong human player.

However, one idea might be that the diffences in the graphs are affected by the number of territories during the games ?

Please note that the "score" in game #9 did never exceed 10 points, which seems to be quite unique here (#36 was in this range for most of the game, especially in the phase before the "point of no return").
On the other hand, Gu Li's large moyo strategy in game #60 was not "praised" by CS.

[Cassandra]

...

[Cassandra]

For those who are interested in, please find attached the exports of CS's analysis of the five games mentioned above.

[uPWarrior]

In any case, we see that the winning probability is higher playing White from the very beginning. The immediate interpretation would seem to be that komi is too large and gives White an advantage. This is consistent with many players' opinion based on human vs human play. The lines seem very consistent over the first 150+ plays. If AG were better playing White, the gap between the lines should diverge move by move. This does not seem to happen to any significant extent.

The only thing we can conclude is that Crazy Stone prefers white. We know nothing about AlphaGo's perceived winning probability.

[Cassandra]

The only thing we can conclude is that Crazy Stone prefers white.

Yes, indeed, it seems that there is an offset. I did not consider this.