AlphaGo vs. AlphaGo with Michael Redmond 9p: Game 8

Michael Redmond 9p, hosted by the AGA E-Journal's Chris Garlock, review the 8th game of the amazing AlphaGo vs. AlphaGo selfplay games. Sgf file here: www.usgo.org/ne...
"In this game we're going to see two 3-3 invasions; when AlphaGo jumps in to the 3-3, the other AlphaGo does as well," says Redmond. "This is something that happens a lot, and I have my own theories about why that might be. We're also going to see the follow-up moves in both of those corners, so we'll see some options about possible follow-ups. And then there's going to be some spectacular fighting inside of Black's moyo."

[@trucid2]

Pizza........................✓ Beer..........................✓ Friday Night.............✓ Redmond Review....✓ If life can get better than this, I don't know how.

[@Rubrickety]

Everyone who enjoys these should watch the 3 new videos from DeepMind where Ke Jie and Fan Hui analyze the Ke Jie / Alphago games, with realtime analysis and win % from AlphaGo itself. They're fascinating. The analysis is in Chinese, but with excellent English subtitling. I hope Michael will have some comments on these once his new videos start being taped after those were posted.

[@kmartine622]

Thank you Michael and Chris for these fascinating, informative, entertaining, and enriching videos.

[@hippophile]

Was that the dog panting at 53:15? :-)

[@Rubrickety]

Ever since Alphago hit the scene, I've wondered what it would think of Shusaku's Ear Reddening move. Is it actually as brilliant as history has made it? Or would Alphago say "eh, I have a few better options".

[@letMeSayThatInIrish]

So many interesting questions like this could have been answered if google made alphago available to the go community. * Sigh *

[@peterschumer924]

Perhaps we should refer to an ear-reddening move by AlphaGo as a "circuit breaking" move.

[@whendrik]

For the Russian speakers, or users of translation software, Alexander Dinerstein (Александр Динерштейн) has the same fascination about that corner move, read his view on rusgo.org/magazine18/

[@hippophile]

On the variation after O17, where Alphago plays P17: The attachment at O18 seems to have been reinstated by humans after a new variation was found with N18 P17 O16 and the push through and cut followed by hane and connect at S16 S15 S17 R14 then white has a splendid move at M16! I have seen a couple of pro games with this. The latest, with Ke Jie taking black in the position and losing to An Sung Joo.

[@hippophile]

The continuation: Black M17, White M16, black P18 then White throws in the perfectly timed Q15 atari before squeezing with the throw-in at N17. Result is ko in the corner, which is pretty tough.

[@renevillarreall.r.3503]

Thanks Michael and Chris. About the S17 hane... I think it was played because it reduces the aji of White playing at S14 and makes Black thicker. It seems to me that it's the same idea as not playing the hanetsugi at S16 to avoid making Black thicker, just in the opposite way. What do you think about that? Is it valid or maybe a different reason?

[@whendrik]

For a clean archive of AlphaGo games, with SGF and reviews, look at www.alphago-games.com

[@ConsciousBreaks]

Another excellent and informative video! I wonder if it would be possible at some point (and I think this may have been mentioned already) to make a comprehensive video on the current opinions and pro research done on AlphaGo moves. That would be very interesting.

[@bernardfinucane2061]

I'm getting into the habit of staying late (Central European Time) to see these!

[@DiapaYY]

Bernard Finucane same here :)

[@mikeyoung9810]

Whoa. I became a go fan watching alpha go play and defeat Lee Sedol and have truely been missing Mr Redmond and Go since then.

[@seventus]

Thanks again for these golden videos. They're pretty much priceless content and I'm sure everyone appreciates the hell out of what you're doing.

[@Mdmathemovie2016]

When is Michael going to play against AlphaGo?

[@benjaminschooley3108]

Michael is a 9 dan pro at go and a 9 dan pro at commentary.

[@derpepe0]

44:05 Does anyone have any details about this famous move at a dame point?

[@vesago]

Honinbo Genjo vs Yasui Chitoku, April 1812. This is featured in the book "Appreciating Famous Games" but here's a link to Sensei's Library senseis.xmp.net/?ABrilliantWasteOfAMove

[@DontbtmeplaysGo]

Thanks a lot for the commentary. The game seems to make so much sense with it! ^_^

[@aresistar8285]

I think Fan Hui said in one of the three videos with Ke Jie that AlphaGo didn't "play nice/safe" before it was above ~70% win margin? Can this mean that the theory that one of the AlphaGo's "cooperate" to make the game 0.5p false?

[@DiapaYY]

Ares iSTAR I understood it as even more than 70% but maybe I'm wrong

[@BluesPiano100]

Fan Hui also claims that AlphaGo doesn't care about shape but only about efficiency while Redmond sais different

[@Quickshot0]

If Alphago plays for efficiency and Redmond has the impression that it's playing shape moves a lot. Then that probably means that often times shape moves are efficient. Which would certainly be a useful tip to remember then. Though of course there are many an exceptions as the self play games demonstrate. Also I understood that it doesn't even really like to slack even if it has passed 80% already, as I think was demonstrated in one of the alternate scenarios that Ke Jie tried out in one of the videos. And AlphaGo despite having well over 80% lead in its assessment answered with a rather merciless response.

[@seventus]

It would certainly do that were it an older version of Alphago like the master version, but I think you're right since afaik the self play games were on a post Wuzhen version.

[@steve4427]

After these A vs A games, is there a new view on what KOMI should be?

[@ANSIcode]

Alpha has komi built into it so it's trained to play with a fixed komi. It's not clear how to make it to play equally well with different komi.

[@steve4427]

Thanks for responding. I'm wondering though if komi should be set differently dependent on the 'rank' of combatants, eg if one is giving a stone or two should komi vary? Thanks again

[@ANSIcode]

Standard practice is to reduce komi to 0.5 points (Jap. rules) as the first stage of handicap and only then starting to give handicap stones while maintaining komi 0.5.

[@steve4427]

Thanks, I'm a fan of Redmond!