Predict AlphaGo on Future of Go Summit in Wuzhen, China

dhu163 · **#101**

the chinese commentators (gu li, etc) have been impressed by ke jie and are actually quite optimistic

dhu163 · **#102**

the feeling is maybe the top right was wrong for ke jie, but actually demis said it was perfect play for the first 30 moves. Next, they criticised Ke Jie for playing on the left and not protecting on the lower side. Alphago played some very strange and crude moves, but then that it seemed W was in serious trouble. But it was a huge complicated mess with the life and death of 6-10 groups unclear.

Then Ke Jie played a brilliant move on the upper side preparing for the ko.

Then on, they can't find a way out for alphago from the complications due to the huge ko on the lower left that could kill B's group, and so they've been optimistic for quite a while though it is too complicated to say. And they still think Ke is playing very well, finding better moves than they've noticed.

dhu163 · **#103**

Gu Li just said he thinks alphago's winning percentage is probably higher for Ke Jie!

djhbrown · **#104**

long ago, i predicted that Ke Jie would win on points in the second round, to keep the audience enthralled, only to be crushed by an uppercut in the third. let's see what happens on Saturday

dhu163 · **#105**

unsurprisingly Gu Li now notices good moves for alphago.

he predicts several different options that might happen since often W might be behind by a ko threat.

Or B ends the ko, W captures the top right, seals in the centre B group for a semeai where W loses the lower right. And W tries to kill the entire top side.

dhu163 · **#106**

now he thinks ke jie made a big mistake with too small a ko threat

xiayun · **#107**

I very much appreciate that Ke Jie is trying all different strategies in these games. AlphaGo may not have thought the mid-game was all that complicated, but it was pretty exciting to see for human players.

djhbrown · **#108**

Michael and Andrew agreed that it looks like this Alphago is making less reliance on Monte-Carlo than last year - but i'm not so sure... the biggest change in her is the introduction of TPUs, and it's possible that this enables her convolution inputs to scope bigger quantities of pixels - possibly even the whole board. that would make a huge benefit - but it's also possible that the TPUs can do with hardware what used to be done with software insofar as search is concerned, so it could be that her Monte-Carlo search is much broader than before as well. Stephanie mentioned her deep reading... however, my prediction for Round 2 has just gone wrong so i could be wrong about this too

https://books.google.com.au/books?id=AT ... lo&f=false
PS oh, no, wait... they were using TPUs last year as well
https://cloudplatform.googleblog.com/20 ... -chip.html
i guess the photo in the article is the sort of thing they mean when they say Alphago is running on a single machine - try balancing that lot on your laptop knees

dhu163 · **#109**

yes, ke jie resigned

but the game was a spectacular one

djhbrown · **#110**

their post-game analysis looks even more dramatic than the game; hope someone is recording it, with sound

dhu163 · **#111**

hassabis did comment that they've never seen alphago put under so much pressure, the first 15 moves alphago thought were perfect and up to move 100 it was extremely close. Ke Jie said he thought he was winning around move 100 but said he got too nervous and hasty.

pookpooi · **#112**

Ke Jie request to play as white in the next game (he cite white has more winrate) so that he can give exciting match and Demis Hassabis said yes

Uberdude · **#113**

For discussion of the games themselves without off-topic chit-chat: viewtopic.php?f=13&t=14250.

Sneegurd · **#114**

Not much to see here. Sooner than expected, but still. AlphaGo is to Go what was DeepBlue to Chess. Hopefully, we'll see engine development speed up, hopefully we'll see engine tournaments on a (semi) professional level (like e.g. TCEC in chess).

Pippen · **#115**

Sneegurd wrote:

Not much to see here. Sooner than expected, but still. AlphaGo is to Go what was DeepBlue to Chess. Hopefully, we'll see engine development speed up, hopefully we'll see engine tournaments on a (semi) professional level (like e.g. TCEC in chess).

...hopefully we'll see better conditions for humans, e.g. 2-3 days for a game like in the old days, so that a player can really study a position or a database of 100.000 AG-self-games or at least some games before a competition.

Galation · **#116**

Sneegurd wrote:

AlphaGo is to Go what was DeepBlue to Chess.

I liked better the following sentence (given by Demis Hassabis I think)

Quote:

where Deep Blue have been the end, AlphaGo is the beginning.

In the first game, Ke Jie played so many of the new moves learned by AlphaGo games and many Pro players confirmed that the AI impact on Go theory have been huge in just two years, compared to the millennia that Go existed.
I am only a beginner, but I enjoy this game and I am happy to share this opportunity.
Galation

Charlie · **#117**

Sneegurd wrote:

Not much to see here. Sooner than expected, but still. AlphaGo is to Go what was DeepBlue to Chess. Hopefully, we'll see engine development speed up, hopefully we'll see engine tournaments on a (semi) professional level (like e.g. TCEC in chess).

I don't think AlphaGo is comparable to DeepBlue. I think AG is far more generalised and the difference in ability between AlphaGo and top humans is way, way more marked. After all, DeepBlue won but it was nowhere near a whitewash - it took ten years for chess AIs to progress to that point and I think AG is already there.

Pippen wrote:

...hopefully we'll see better conditions for humans, e.g. 2-3 days for a game like in the old days, so that a player can really study a position or a database of 100.000 AG-self-games or at least some games before a competition.

I agree with the desire for longer time for the humans but did humans "in the old days" have the option to do research during that time? I think they could only use their own mental faculty - not databases. I also wonder whether the modern style would really benefit from extremely long time controls.

Personally, I have absolutely zero interest in a game where a human player is using a database and playing against an AI. To my mind, that would be the most un-Go-like thing you could have. I'd hate to see a future where sequences and moves have to be "legitimised" by databases or bots before humans are prepared to try them out.

pookpooi · **#118**

Stephanie Yin calls AlphaGo slack move 'Announcement Move' and Hajin Lee like this name very much. Pro team, however...

pookpooi · **#119**

Bonobo · **#120**

(Emphasis added)

DeepMind wrote:

While AlphaGo is stepping back from competitive play, it’s certainly not the end of our work with the Go community, to which we owe a huge debt of gratitude for their encouragement and motivation over the past few years.

We plan to publish one final academic paper later this year that will detail the extensive set of improvements we made to the algorithms’ efficiency and potential to be generalised across a broader set of problems. Just like our first AlphaGo paper, we hope that other developers will pick up the baton, and use these new advances to build their own set of strong Go programs.

We’re also working on a teaching tool - one of the top requests we’ve received throughout this week. The tool will show AlphaGo’s analysis of Go positions, providing an insight into how the program thinks, and hopefully giving all players and fans the opportunity to see the game through the lens of AlphaGo. We’re particularly honoured that our first collaborator in this effort will be the great Ke Jie, who has agreed to work with us on a study of his match with AlphaGo. We’re excited to hear his insights into these amazing games, and to have the chance to share some of AlphaGo’s own analysis too.
[..]
We’re now publishing a special set of 50 AlphaGo vs AlphaGo games, played at full length time controls, which we believe contain many new and interesting ideas and strategies.

https://deepmind.com/blog/alphagos-next-move/

Predict AlphaGo on Future of Go Summit in Wuzhen, China

Who is online