AlphaGo to Zero: Revisiting AlphaGo vs Lee Sedol, Game 2, with Michael Redmond 9p

[@mikeyoung9810]

Thos days of Lee Sedol and you two doing the commentary were some of the most enjoyable times during that alphago match. I got really excited about Go for awhile. I've since cooled because I have no idea what I'm doing wrong and sadly I don't have a million man hours left in me to make much progress. But thanks again for a very interesting time during that match.

[@ConsciousBreaks]

That's a shame to hear. One of the best ways to keep Go interesting and to improve is to find a community(especially a local club-after coronavirus ofc) where you can play with and learn from. I would definitely recommend you to try and do this if you haven't already-it might rekindle your interest in Go! :)

[@esakoskinen5034]

I recently rewatched the documentary and this was extremely interesting to see afterwards. Thank you Chris and Michael for the review and sharing your perspective and experiences in more detail!

[@ewalltom]

Regarding D14, the Korean fellow doing the commentary for the channel while Michael and Chris were doing the commentary in Korea said if D14 was a good move, then the program was stronger than he was. He was very impressed with the move.

[@Maharani1991]

If anybody reads these old comments, I would absolutely love to see Michael revisit the AlphaGo vs Ke Jie games, too.

[@decidrophob]

1:32:18 D14 is the move Myungwhan Kim referred to as "if AlphaGo knows that this is working, then she knows the entire board, and hence Lee Sedol has no chance".

[@arekkrolak6320]

it doesn't know, probably it just figures out it wins 90 games out of 100 when this move is played

[@decidrophob]

@@arekkrolak6320 I guess the point Kim tried to make was not that AlphaGo "knew" the entire board but that she had "the global idea of the board". The emphasis is not on the "know" part, but rather on "the entire board" part. Very few commentators at the time had qualitative understanding of MCTS with deep neural networks, hence the commentators like Kim most probably had very little idea about how AlphaGo "thought" about the game. The only way they could explore AlphaGo was like Turing test towards others -- abductively inferring how AlphaGo thought via the moves played. And at least Kim seemed to realize the global thinking mode of AlphaGo with this particular move. I feel exactly the same regarding move 37, but at the time, he did not seem to be positively moved by move 37. Humans (at least at the time) would have devoted the entire intellectual resources to think about "the local" response to the top invasion. On the other hand, AlphaGo and other MCTS based programs assign some policy net related probabilities all across the board to start reading elsewhere as well. This probably made the moves more flexible than humans. In this sense, AlphaGo knew the entire board as Kim commented.

[@davidhunt7427]

??? Why did Myungwhan Kim refer to AlphaGo as _she?!_ Shall we now have a discussion about how many genders AlphaGo supposedly has??!? Bad jokes aside,.. I had much the same reaction to this move that Michael had. Like most kyu players I want to pounce upon G17 with a hammer,.. when the better players know you start by taking all the oxygen out of the room first.

[@Pintkonan]

@@davidhunt7427 its a funny thing with these ai's. in chess alphazero is referred to often as he, while "his little sister" leelachesszero is often referred to as she in videos where their games are analysed.

[@sinfinite7516]

Loved the revisit to these Alpha Go v. Lee Sedol videos!!!

[@MrStarchild3001]

Just ordered the book! Deeply appreciate the book and the free videos. Thank you!!

[@taxicabnumber1729]

Me: ooh an explanation from Redmond Chris Garlock: Go is a game where one player has black stones and the other player has white stones..

[@ghosthead7702]

Review starts at 25:42.

[@osuf3581]

MCTS with neural networks for either move prediction or evaluation has existed as an approach for many years prior to AlphaGo; no one had made it work well enough to beat the "strong-amateur" rule-based bots though.

[@Maharani1991]

+

[@ConsciousBreaks]

This video has the longest intro of all time; you can see Michael wanting to start the commentary already lol

[@XiaochengStephenHu]

Hi, this was a VOD from an event that was promoted on the Twitch front page, hence the format was 120 minutes instead of 75.

[@noeleom6394]

Thanks for the quality content as always!

[@decidrophob]

1:20:00 I understand that Lee Sedol's first regret in the post-game analysis was that he did not play F4 (1 in the variation) instead of E5. He possibly saw this move in real time as well, but he most probably judged the sequence from E5 cut was even clearly better for white, not allowing black to defend F6 cutting point in one way or another. In Go Premium Channel, Takao Shinji said at the time the actual sequence in the game in which white secured the bottom stones was clearly in favor of white, and he found it hard too see what AlphaGo was trying to accomplish in this local fight. I have also had trouble seeing that the sequence played in the game was in favor of black. Your explanation in this video and the book, the squeeze from the center (1:13:00) seems the most convincing verbal explanation for the positional judgment in favor of black. I think it was hard for the humans, including Lee Sedol himself, to assess this cut black group (D5 group) as thickness. At the time, I was like Takao Shinji feeling that black only created a heavy group in exchange for almost nothing.

[@Maharani1991]

+

[@shabadsebastian]

Thank you!!!

[@quackivonquackenstein2398]

I looked up the "triple take" from the shoulder hit. This video, starting at 1:18:00: kzbin.info/www/bejne/ol6qpJmvi6h1q5I It is really a nice little weird scene, I have to say. I like it! Michael is completely absorbed by the move, and Chris confused by Michaels absentmindedness.

[@renevillarreall.r.3503]

Around 50:00 ... About this peep, I think that it is somewhat important that Black plays it without the extension on the lower side. Maybe if he does play the extension the group would be strong enough by itself that the peep is pointless and does not contribute enough to justify erasing the aji on the right side. My point is that maybe the reason that the older masters didn't like it is that they hadn't considered playing away from the lower side and so they only had that shape with a strong group and strengthening it with the peep was a net loss in that context. What do you think about that reasoning?

[@decidrophob]

I think this early peep is very closely connected with the famous shoulder hit kind of global strategy, to play from the center connecting to the bottom black than worrying about right bottom yose. As far as my memory goes, even when AlphaGo played kaketsugi, which in itself got almost extinct in Zero, I remember it very rarely played the early peep after that. Thus, the early peep is estimated not to be as much a part of right bottom joseki as a part of the global strategy unique to this particular game. As Michael and others explain, in various other games, though right bottom white is too thick and is not big for black to approach there, you still leave significant possibility that you want to play other than peep as black. The older masters probably did not tend to foresee the global strategy as clearly as AlphaGo did, hence wanted to have more open options.

[@seapanda-117]

I know it's been said already, but thanks for the material. I tried to watch the recording of you guys doing the live commentary of the games, but they are all a little jittery and jumpy. Were they corrupted or were they edited like that? Also Steven, thanks for the work editing these. Michael and Chris acknowledge you every video, but I don't think we do. I can't watch these live, so I appreciate you editing them!

[@XiaochengStephenHu]

Hi Ryan, not sure why "all of them were jittery" on Twitch - might be a peculiar issue with the video player. In any case we try to edit all of our Twitch VODs, and post them here eventually (with the exception of AlphaGo vs. The World, which is a series designed for KZbin). But it seems that this video was okay? Thank you for your support!

[@armyguy196]

Just make 2 videos. We don't want to wait every couple minutes for an elementary aside.

[@mareksicinski3726]

?

[@mareksicinski3726]

You an skip whenever. Also don’t write we when you’re talking about yourself

[@Vonsen]

I see your book is currently available only in pdf/ebook form. Will you be publishing it as a print book eventually?

[@RatatRatR]

I'd like to see it become a physical book too.

[@jonathan4055]

I have posted subtitles in French for the game, they will ask for you to review could you please approve them :D

[@USGOWeb]

Published!

[@sinfinite7516]

Was game 5 ever covered?

[@mareksicinski3726]

24:42 but there were issues no?

[@mareksicinski3726]

34:32 well most different stuff is new

Chris Garlock's forceful voice jars the ear, his crude conversational style disturbs the mind. Please stay behind the scenes Mr. Garlock!

[@sinfinite7516]

Your being rude ngl.

[@半道-y1e]

很有趣，没想到能看到西方版本。 基于西方更重视技术层面的理解，围棋在东方更近乎于道！ 至于胜负，反而没那么重要。 博弈当中的取舍、选择、判断、计算，才是围棋的本质。 需要胜负的只是人而已……

[@Neurotoxin-q2r]

2:17 Michael redmond chanses his color, I knew that he is reptiloid, that's why 9 dan

[@Metalsonar]

Chris is so annoying.

I want Chris Garlock gone from videos. #sorrynotsorry

[@procreatineGO]

let me help you 25:41 start to be honest, not to be rude but why bother reviewing old game like this? ppl are using katago or lizzie to review their game already, how about reviewing current matches like japan Honinbo or Meijin it would be more interesting than doing a 2 hours on a game that no one talk about it. you have a 9p there.. use it

[@brett4748]

This game has historical significance to the game of Go. This series with Sedol is why people are using AI to review games everywhere. Not everything in the world needs to be a discussion of what just happened. Just because a match is current doesn't make it important. If you don't like these videos, you could always go elsewhere; it's not like the title mislead you. Also, you are completely wrong that people aren't talking about this game. The Deep Mind documentary came out only a few months ago and interest in these games are resurfacing.

[@vegahimsa3057]

The last human to defeat top AI will permanently be among the most important matches in go history.

[@mareksicinski3726]

Because it is a historic and important game? ‘That no one talk about it’ what is that grammar, anyway it’s probably the most talked about game kf go ecer Use ‘it’? Also did you not watch the video? Do you not know who they are or why they are doing this? Stupendous ignorance. Also, people talk about games even from hundreds of years ago, like the ‘ear-reddening game’