Talk:AlphaGo versus Lee Sedol

This is the talk page for discussing improvements to the AlphaGo versus Lee Sedol article. This is not a forum for general discussion of the article's subject.

Put new text under old text. Click here to start a new topic. New to Wikipedia? Welcome! Learn to edit; get help. Assume good faith Be polite and avoid personal attacks Be welcoming to newcomers Seek dispute resolution if needed Article policies Neutral point of view No original research Verifiability

Find sources: Google (books · news · scholar · free images · WP refs) · FENS · JSTOR · TWL

A news item involving AlphaGo versus Lee Sedol was featured on Wikipedia's Main Page in the In the news section on 12 March 2016.

This article is rated B-class on Wikipedia's content assessment scale. It is of interest to the following WikiProjects:

Go High‑importance This article is within the scope of WikiProject Go, a collaborative effort to improve the coverage of the game of Go on Wikipedia. If you would like to participate, please visit the project page, where you can join the discussion and see a list of open tasks.GoWikipedia:WikiProject GoTemplate:WikiProject GoGo articles High This article has been rated as High-importance on the importance scale. Computing Low‑importance This article is within the scope of WikiProject Computing, a collaborative effort to improve the coverage of computers, computing, and information technology on Wikipedia. If you would like to participate, please visit the project page, where you can join the discussion and see a list of open tasks.ComputingWikipedia:WikiProject ComputingTemplate:WikiProject ComputingComputing articles Low This article has been rated as Low-importance on the project's importance scale. Google Low‑importance This article is within the scope of WikiProject Google, a collaborative effort to improve the coverage of Google and related topics on Wikipedia. If you would like to participate, please visit the project page, where you can join the discussion and see a list of open tasks.GoogleWikipedia:WikiProject GoogleTemplate:WikiProject GoogleGoogle articles Low This article has been rated as Low-importance on the project's importance scale. WikiProject Google To-do: edit history watch purge Here are some tasks awaiting attention: Article requests : Articles for most of the other products listed here and here. Assess : All articles in the Category:Unknown-importance Google articles and Category:Unassessed Google articles using the project's assessment scale Expand : Google Mapathon, Google Talkback Maintain : This WikiProject Merge : Google mobile services into List of Google products Stubs : Category:Stub-Class Google articles and Category:Google stubs Update : List of features in Android and Gmail interface#Product integration. Update logos of Google Marketing Platform products Other : Add more stuff to this to do list if you like! (click here...) create: Help the Google article for a good article status Improve the Outline of Google Get more members using : {{subst:Wikipedia:WikiProject Google/Invite Members}} Infobox Images with transparent areas needing a different background color

Is "W+resign" standard go language? I'd expand to white wins, black resigned. Also is this match on tv somewhere in asia? -Koppapa (talk) 16:19, 9 March 2016 (UTC)Reply

"W + Res" was used in the "Final Scores" table on Youtube's live video. I think it can also be used in this article. Many Chinese websites broadcast the live video of this match on line although no television station in China broadcast it. --Neo-Jay (talk) 19:08, 9 March 2016 (UTC)Reply

Yes, that is standard notation. If White had won by by a certain amount of points, one would write W+4.5 (i.e., at the end white has 4.5 points more than black). If black resigns, then "white wins by resignation", and one would write either of W+R, W+Res, W+resign (or similar) --188.96.85.12 (talk) 01:36, 10 March 2016 (UTC)Reply

I feel "W+resign" is MOS:JARGON and that we should spell out "Lee Sedol resigned." Most people visiting the article will be in the lay audience and also are more interested in who won/lost. While "white wins by resignation" is plain English it also means the reader then needs to figure out who was white. --Marc Kupper|talk

For consistency in the article, should Go ranks be hyphenated? For example, should we write "9 dan" or "9-dan". Currently, the article contains both versions. MLabrum (talk) 17:07, 12 March 2016 (UTC)Reply

Does Lee Sedol have better chances of winning the next match against AlphaGo? --Pragyaverma92 (talk) 22:08, 9 March 2016 (UTC)Reply

Better than the previous game for two reasons: 1. It seems that Sed-ol played some weird moves while testing the machine a few times in this game, which he might not repeat in the next game. 2. He is better prepared to not be amazed at the machine's sudden moves. — Preceding unsigned comment added by Deveshbatra (talk • contribs) 22:15, 9 March 2016 (UTC)Reply

Please keep in mind that Wikipedia's talk pages are for improving the article content and not general discussion. See WP:TPG for more about this. --Marc Kupper|talk 20:36, 10 March 2016 (UTC)Reply

Challenge match

Game no.	Date	Lee Sedol	AlphaGo	Result	Moves
1	9 March 2016	0	1	Sedol resigned	186
2	10 March 2016	0	1	Sedol resigned	239
3	12 March 2016
4	13 March 2016
5	15 March 2016
AlphaGo leads: 2–0

Is this table better?-Koppapa (talk) 06:45, 10 March 2016 (UTC)Reply

Lee Sedol is a Korean name. If family name is used in this table, it should be Lee, not Sedol.--Neo-Jay (talk) 08:46, 10 March 2016 (UTC)Reply

Per WP:COLOR we should not use the color alone as a method of distinguishing which player had white vs. black. I personally don't like the color coding of the full cell background. It is distracting and it took me a moment to figure out why the table had the black cells. Once I realized the purpose of the colors my first thought was that we should use dots that look like the stones. I also moved the match score over to the right column and made it a cumulative score rather than showing 0 1 in the second row. Finally, I don't know what the tradition is for Go but in many sports the challenger is named first and so I flipped those columns. This also matches the name order in the article title.

Game no.	Date	AlphaGo	Lee Sedol	Result	Moves	Match Score
1	9 March 2016	white	black	Lee Sedol resigned	186	0–1
2	10 March 2016	black	white	Lee Sedol resigned	239	0–2

I feel the tan background is distracting but I could not find versions of the images without a background. Note, I trimmed the header/footer from the example table only for brevity as the example is intended only to show using stones instead of background colors and the addition of the match score column.

In comparing the proposes versions of the table against AlphaGo_versus_Lee_Sedol#Summary I believe the one in the main article is the better of the three other than that the result column. At present the result column has "W+resign" and "B+resign". Experienced go players can read and understand that but I feel it's falling into MOS:JARGON and that we should spell out "Lee Sedol resigned". --Marc Kupper|talk 20:01, 10 March 2016 (UTC)Reply

I agree we should try to present the match summary in a way that is easy to follow for non-Go players; many of the people interested in this match will, like me, be coming at it mainly from the AI perspective. Espresso Addict (talk) 01:49, 11 March 2016 (UTC)Reply

I think this is all overthinking things. A simple table used in whatever format the actual tournament uses is best. Clarity is good, but I think bolding the winner, and the "results" row, make it extremely clear what happened even if somebody is baffled by what "W+2.5" means or the like. SnowFire (talk) 09:41, 12 March 2016 (UTC)Reply

What about providing the games as KGS files? --Jobu0101 (talk) 18:41, 10 March 2016 (UTC)Reply

In huwiki article, I take a link in 'External links' section to GoGameGuru site, where game analysis and downloadable SGF files can be found. --Rlevente (talk) 14:10, 11 March 2016 (UTC)Reply

When I was looking at news articles yesterday I ran across something that said Lee Sedol would get paid an extra bonus for participating in all five games. I suspect that needs to be dug up as it's not clear from the article if AlphaGo versus Lee Sedol is five games or is a "match" where the first player who wins three games is declared the winner and they may end up not playing five games. --Marc Kupper|talk 20:45, 10 March 2016 (UTC)Reply

@Marc Kupper: The first player who wins three games is declared the winner and can get the $1 million prize. But they should continue to play the rest two games even if one player wins the first three games. --Neo-Jay (talk) 15:34, 11 March 2016 (UTC)Reply

"Lee eventually assayed a complex ko from move 131.." - what exactly does "assayed" mean heere? "Attempted"? Could a better word be used? Bubba73 ^{You talkin' to me?} 01:51, 13 March 2016 (UTC)Reply

Might the word intended have been essayed? -- The Anome (talk) 02:43, 13 March 2016 (UTC)Reply

That makes more sense. Bubba73 ^{You talkin' to me?} 02:48, 13 March 2016 (UTC)Reply

AlphaGo seemed to mis-estimate the result of the capturing races. Perhaps the neural networks have difficulty distinguishing between two situations where almost all the board locations are in the same state, so they cannot correctly count the number of liberties remaining on both sides and decide who will win the race. What do you think? JRSpriggs (talk) 10:20, 13 March 2016 (UTC)Reply

It would be interesting to add commentary from experts in AI on this point; the material I've found has all been from the go side. Espresso Addict (talk) 03:11, 14 March 2016 (UTC)Reply

According to "Michael Redmond on AlphaGo, Lee Sedol and Honinbo Shuwa" by Chris Garlock at AGA's E-Journal, "... Michael Redmond 9P ... was convinced that Lee’s 'brilliant' move 78 — which had won the game — didn’t actually work. Somehow, though, it had prompted a fatal mistake by AlphaGo, which top members of the DeepMind team were still trying to understand, ...". JRSpriggs (talk) 01:12, 15 March 2016 (UTC)Reply

Apparently, move 78 was unexpected by AlphaGo (having a probability of less than 0.0001), so most of the search tree was pruned away leaving AlphaGo to build it up from scratch. The time control built into the program did not allow for this to be done before making move 79. So AlphaGo made a mistake because the tree was too small at that time. JRSpriggs (talk) 03:54, 15 March 2016 (UTC)Reply

Thus there is at least one bug in AlphaGo, namely, that it does not allow more time to rebuild the search tree when it is too small. There may be other bugs which were responsible for the estimated probability of move 78 being too small to begin with. JRSpriggs (talk) 11:33, 15 March 2016 (UTC)Reply

What do people think of (and possible names) for a category that both this article and Deep Blue versus Garry Kasparov? Naraht (talk) 11:23, 13 March 2016 (UTC)Reply

I've created Category:Human versus computer matches, and put both of these articles, and Brains in Bahrain, into it. Are there other matches that could be added to this category? -- The Anome (talk) 12:35, 13 March 2016 (UTC)Reply

Showing the game in a series of 100 move sections is using a lot of vertical space in the article. Would there be an objection to switching the display to horizontal/floating that wraps depending on the browser window width much like how text wraps? For example, here's game 2 of the match:

First 99 moves

Moves 100-199

Moves 200-211

If you change the browser window width the displays above will wrap and adjust to fit the available space. The change to the article is trivial. Use {| style="display:inline;" for each of the wrapper tables. --Marc Kupper|talk 19:22, 13 March 2016 (UTC)Reply

After saving the above I found the wrapping does not work in Chrome. I'm looking into that. It works in Firefox and Internet Explorer. --Marc Kupper|talk 19:25, 13 March 2016 (UTC)Reply

I believe the Chrome (and Opera) issue is fixed for the example above. I used {| style="display:inline; display:inline-table;". I know the display:inline; part is not needed for newer browsers but left it in place just in case there's a browser that's unhappy with display:inline-table;. What I'm looking for is feedback on if wrapping the tables like this is ok or if there's a preference for the vertical alignment the article currently has. --Marc Kupper|talk 19:45, 13 March 2016 (UTC)Reply

• Horizontal looks good to me (in Firefox) -- as long as it doesn't break the display altogether in Chrome. Espresso Addict (talk) 23:11, 13 March 2016 (UTC)Reply
  • Looks good to me in Chrome, and I agree that using less vertical space is a good idea. Thue (talk) 11:47, 14 March 2016 (UTC)Reply
•  Comment:: Wikipedia is not repository of source material, I think it's not necessary to put the full Kifu in this article. I think just put some comments from other grand masters is good enough.--Liaon98 (talk) 11:15, 15 March 2016 (UTC)Reply
• I don't play go and find the every-100-moves snapshots to be useful as it lets me see how the game flowed. As I don't know the rules of the game comments below the boards such as 154 at mean little to me but I assumed they are useful to those that know the game. FWIW, the Korean language version of the article is also using the same set of boards that we have here including the same commentary such as "154 자리에 ". I had not known the name for the game record is qipu. I found that a bit amusing as I was reading about quipu earlier today which were records stored as knots on strings in the Inca. --Marc Kupper|talk 08:17, 17 March 2016 (UTC)Reply

• To Marc Kupper: You cannot place a stone on a lattice point which is already occupied by a stone, so usually the moves will occur in different locations which allows one to make diagrams like these kifu. However, if a stone is captured, then it is removed from the board and its location is then available to be used by another stone. This is especially common in Ko fights.

For example, if move 113 is placing a black stone on location R11, then a black disk labeled "13" is shown at R11. If that stone is captured by move 120 which places a white stone at location S11, then location R11 becomes vacant again. When move 122 places a white stone at R11, this cannot be shown on the diagram because it already has a disk there for move 113. So instead a notation is made under the diagram that move 122 was at the location used by move 113, or "122 at 113" for short.

When the first move to a location was made in one of the earlier hundreds of moves and another move is made to that location one cannot identify the earlier move by a number, so a symbol (triangle or square) is used to distinguish it from other locations. JRSpriggs (talk) 01:26, 18 March 2016 (UTC)Reply

The dubious tag appears at the end of the following in the summary section "Under the official rules, it was intended that the colour assignments would be done at random. However during the press conference after the fourth match, Lee requested that he play black in the final match due to it being more of a challenge than winning with white. Hassabis agreed to his proposition."

I'm not sure what is dubious here - it matches my memory of the press conference and is supported by the cited wired article. Zabdiel (talk) 11:49, 15 March 2016 (UTC)Reply

Agree, I think tag should be removed. crandles (talk) 12:33, 15 March 2016 (UTC)Reply

Not to mention the Youtube video of the 4th match contains the press conference confirming this arrangement.--97.84.91.70 (talk) 14:01, 15 March 2016 (UTC)Reply

Removed crandles (talk) 14:25, 15 March 2016 (UTC)Reply

The normal practice for multi-game matches is to randomly choose who will be black for the first game and then alternate who is black for subsequent games. I have never heard of another method for even matches. See nigiri.

At the beginning of the first game, Lee Sedol was chosen to be black by nigiri. Since he had the bowl of white stones, they exchanged bowls. The announcer also stated that this had been done. This process did not occur in the later games.

Since Lee Sedol was already scheduled to take black in the fifth game by the above process, he would not have asked to have black. Perhaps he said that he was glad that he had black or that he wished he did not have it and this was mistranslated. JRSpriggs (talk) 18:14, 15 March 2016 (UTC)Reply

I was also puzzled about that Lee asked to play black. During the match I took a couple of shots, without success, to find any official rules about the match. I did it again tonight and still was not successful though I did find a picture of the honorary 9-dan certificate. Thus while bloggers, and even this WP article have mentioned "the official rules" they were either only released in Korean or are in print media handouts that none of the on-site reports then documented. JRSpriggs, brings up a good point in that it may be a translation issue. I checked the 알파고_대_이세돌 article (running it through the Google translator) and there's no mention of changing or agreeing to who would play white for game 5. I learned that someone there wants to move the article to something like "Google DeepMind Challenge Match" much like what happened to the English article. The talk page only has one thread, and it's a long one, about the propose change to the article title. --Marc Kupper|talk 08:01, 17 March 2016 (UTC)Reply

@JRSpriggs and Marc Kupper: AlphaGo and Lee Sedol exchanged black and white in the first four games. The fifth game should have randomly chosen again who would be black since this would be the final and odd-number game. This is the normal practice for multi-game Go matches. However Lee Sedol (bravely) requested that he play black. --Neo-Jay (talk) 22:11, 18 March 2016 (UTC)Reply

"failing to recognise black's "tombstone squeeze" tesuji "

I think this should be removed as AlphaGo sacrificed the stones in the lower left to cover the cut above 40. After AlphaGo's sequence a cut can be captured in a geta rather than relying on the ladder. This might be a bad judgement call, but I see no evidence it was a misread. — Preceding unsigned comment added by 149.155.192.89 (talk) 21:04, 16 March 2016 (UTC)Reply

Details on contract signing and match organization are explained in the following video:

• AlphaGo Match Stories and Game Highlights — Preceding unsigned comment added by 93.103.244.180 (talk) 17:13, 26 March 2016 (UTC)Reply

The main political parties in South Korea tried to capitalize on the recent AlphaGo-versus-Lee Se-dol go match, by placing an engineer, a mathematician and a scientist on top of their respective lists of proportional representation candidates for the April 13 general election. It is certain that they all will be given Assembly seats. ^{[1] [2]} — Preceding unsigned comment added by 93.103.244.180 (talk • contribs) 17:34, 13 April 2016 ‎(UTC)Reply

Several go software vendors announced new projects that will compete with Alphago. (The following is an abbreviated version of ^[1].)

On March 1 a "Deep Zen Go Project" was announced between the developers of the computer go program Zen (Yoji Ojima, Hideki Kato), telecommunications and media company Dwango and a deep learning research team at Tokyo University (developers of Ponanza - a shogi AI that beat all the human pros). Japanese Go Association is also pledging their support. Their goal is to beat AlphaGo in 6 months to 1 year. ^[2]

Also, in other news, Crazy Stone developer Remi Coulom is also working very hard on his own deep learning version of Crazy Stone. ^[3] — Preceding unsigned comment added by 93.103.244.180 (talk • contribs) 20:53, 28 March 2016 ‎(UTC)Reply

The following articles originally written by Lee Sedol were published between March 18 and March 23 on Donga.com. ^{[1] [2] [3] [4] [5]} English translation is available here.^[6] — Preceding unsigned comment added by 93.103.244.180 (talk • contribs) 17:23, 13 April 2016 ‎(UTC)Reply

Alphago's confidence charts for all five games are available here: ^[1] They were were presented by David Silver (Alphago team) at his UCL talk in late March. — Preceding unsigned comment added by 93.103.244.180 (talk) 21:47, 20 April 2016 (UTC)Reply

Having reviewed the previous discussions, and (I now realize mistakenly) moved the page once back and forth myself, I now propose the following:

• that this page keep its descriptive name of AlphaGo versus Lee Sedol, following the convention of other similar articles such as Deep Blue versus Garry Kasparov.
• that, nevertheless, the match be referred to as the Google DeepMind Challenge Match within the article itself, because that was its official name

-- The Anome (talk) 09:05, 1 May 2016 (UTC)Reply

I dunno. The majority of journalism I read on the topic never used "Google DeepMind Challenge Match" (which doesn't say much) and the articles that did mentioned the phrase once, then moved on. I'm not averse to mentioning that title somewhere, but some variation on "AlphaGo / Lee Sedol" seemed to be by far the most common way to talk about the match, and I don't see why Wikipedia shouldn't mimic that. Mention the official title once then never again. SnowFire (talk) 18:23, 2 May 2016 (UTC)Reply

Whichever name we use, once we're in the article, it would generally mostly be described as "the match", and by other similar indirect references, so that shouldn't be a problem. -- The Anome (talk) 19:14, 2 May 2016 (UTC)Reply

"Google DeepMind Challenge Match" can be mentioned in the lead section. But "AlphaGo versus Lee Sedol", as the article title, should be used in the infobox, and also be used in the lead section. --Neo-Jay (talk) 03:35, 17 May 2016 (UTC)Reply

Hello fellow Wikipedians,

I have just modified one external link on AlphaGo versus Lee Sedol. Please take a moment to review my edit. If you have any questions, or need the bot to ignore the links, or the page altogether, please visit this simple FaQ for additional information. I made the following changes:

• Added archive https://web.archive.org/web/20130324221549/http://sankei.jp.msn.com/life/news/130320/igo13032020420000-n1.htm to http://sankei.jp.msn.com/life/news/130320/igo13032020420000-n1.htm

When you have finished reviewing my changes, you may follow the instructions on the template below to fix any issues with the URLs.

This message was posted before February 2018. After February 2018, "External links modified" talk page sections are no longer generated or monitored by InternetArchiveBot. No special action is required regarding these talk page notices, other than regular verification using the archive tool instructions below. Editors have permission to delete these "External links modified" talk page sections if they want to de-clutter talk pages, but see the RfC before doing mass systematic removals. This message is updated dynamically through the template {{source check}} (last update: 5 June 2024).

• If you have discovered URLs which were erroneously considered dead by the bot, you can report them with this tool.
• If you found an error with any archives or the URLs themselves, you can fix them with this tool.

Cheers.—InternetArchiveBot (Report bug) 05:44, 8 May 2017 (UTC)Reply

  On Nov. 5th the PBS series Frontline did an episode called "In The Age of AI" where Lee Sedol's match against AlphaGo is flagged as a pivital moment in the history of AI. Maybe someone could figure out where this should be mentioned in the article. thanx  JP  — Preceding unsigned comment added by 71.255.35.61 (talk) 07:00, 7 November 2019 (UTC)Reply 

I am sorry, but it seems U do not understand the problem. AlphaZero is NOT a "program", the very idea of AI is, that we do not know, how it thinks! U can not compare this match to DeepBlue vs. Kasparov because DeepBlue was a program: we know exactly, what this program is doing, we can improve it etc. We cannot do that with AI. AlphaZero cannot explain us, how it reasons exactly like I cannot explain to my dog how I win in chess. The precise idea of constructing AI U can find in works of the late Stanislaw Lem, the famous Polish SF writer. JKM - jkmjanusz@o2.pl — Preceding unsigned comment added by 195.242.138.20 (talk) 19:21, 1 January 2021 (UTC)Reply

At least on some browsers and systems, the AlphaGo versus Lee Sedol page is extraordinarily slow to load, and it appears to be needlessly resource-hungry. The issue appears to be that the Goban template—which is used multiple times on the article page—renders 19x19=361 pictures for each go board, of which there are twelve in the article, for a total of 4332 images. Granted, many of these are the same image appearing multiple times, but still, some browsers/systems may struggle under that needless burden. And it is to some extent a needless burden. Yes, in a way it's very clever someone managed to use and extend MediaWiki's capabilities to create the Goban template, but it would be far less burdensome if those twelve boards got included as screenshots/single images. The template info used to create those respective images could still be included in the source, but commented out. Actual article-space use of the Goban template is too clever by half. Given a choice between an uber-clever template or sticking to the KISS principle, I suggest the latter. Yes, the former sounds alluring, but that's a Siren song. There is real merit in not wasting everybody's system resources and not being more resource-hungry than necessary. I appreciate that someone will probably claim the 19x19 tiles approach was less resource-hungry than twelve screenshot images, but that ignores the overhead. Asking a browser to deal with literally thousands of images on a page is asking a lot. —ReadOnlyAccount (talk) 00:36, 11 December 2023 (UTC)Reply