Talk:Computer Go

9x9 win against a pro

I don't know if this is worth adding to the article -
In April 2008, the Go-playing program MoGo, running on a cluster, won an even 9x9 game against 5-dan pro Catalin Taranu [1]. (A 19x19 win against a pro is still a long way off, imho.) Maproom (talk) 13:58, 18 April 2008 (UTC)

Maybe he had a bad day (sick?). Before you go tooting MoGo's horn, this experiment should be replicated several times. What is its winning percentage against professional players? JRSpriggs (talk) 15:39, 18 April 2008 (UTC)

He showed no signs of being sick, and made no complaint. He went on to play against it again.

This was a three-game series, the pro won the other two games, though in the first game MoGo lost its connection with the cluster and played most of the game on an ordinary PC.

The only other 9x9 games I know of between MoGo and a pro were on KGS last June, against Guo Juan 5p. It won 12 games, she won 7. But it is not clear that she was trying to win, she may just have been experimenting with it. Whereas the series against Catalin Taranu was widely publicised and had a large audience, he was definitely trying to win. Maproom (talk) 16:44, 18 April 2008 (UTC)

I heard that the games against Guo Juan were with Guo taking white without komi? HermanHiddema (talk) 19:22, 18 April 2008 (UTC)

Sorry, you are right. Most of these games were played with only half a point komi. Of the three played with 7.5 points komi, Guo Juan won two, MoGo won one. (For anyone who wants to check, look for the 9x9 games on this page: [2].) Maproom (talk) 10:04, 19 April 2008 (UTC)

"People think"

I see that Kibiusa has deleted the section "Why Humans are successful at Go", with the reasons "weasel words ("people think"); POV; original research".

I don't accept the "weasel words" argument at all. Yes, those words can be used as weasel words, but they aren't in this case. There is indeed a widespread belief that computers are bad at Go compared with humans, as expressed in the article itself by the section heading (of which this was a subsection) "Obstacles to high level performance".

The POV/Original research argument carries more weight. But as David Fotland has expressed this view, I think it should be accepted. I wish I could find where he stated it; his web pages about Go programming no longer exist. Maproom (talk) 09:16, 11 May 2008 (UTC)

Gunnar Farnebäck (principal current developer of GNU Go) stated this view in his public talk on Computer Go at the 2008 European Go Congress. I propose reinstating this section. Maproom (talk) 23:14, 17 September 2008 (UTC)

Citation for bots' strengths

The article now reads "Currently, the best Go programs are ranked as (1-3 kyu)[citation needed]." I am wondering what form a citation could take. The facts are not is dispute, anyone can verify them by logging in to KGS (you don't need an account there) and looking at the current strengths of the leading bots. Would a link like this do? Maproom (talk) 08:40, 8 July 2008 (UTC)

This is true, but one needs to know the names of the robots to verify this. Perhaps a link to the KGS rank graph page for crazystone or mogo would be suitable.--ZincBelief (talk) 08:45, 8 July 2008 (UTC)

These are the 1-3k bots on KGS as of 2008-12. All of these are Monte Carlo searchers. Even ManyFaces is using an experimental MC engine. This progress is also being tracked here. --IanOsgood (talk) 23:15, 15 August 2008 (UTC)

• 1k CrazyStone
• 1k LeelaBot
• 1k ManyFaces2
• 2k ManyFaces
• 2k MoGoBot3
• 2k MoGoBot2
• 2k MoGoBot1
• 2k Fuego
• 3k AyaMC
• 3k CzechBot is release 3 of MoGo

These sufficient? HermanHiddema (talk) 11:16, 8 July 2008 (UTC)

KCC Igo prominent?

To User:IanOsgood: I see that to Computer Go you have added, at the head of the list of "Prominent go-playing programs", Silver Star/KCC Igo. Have you any reason for choosing this program, which was derived, illegally, from Chen Zhixing's Handtalk, to head the list? The list does not mention the stronger MoGo and Crazy Stone.

I suggest deleting KCC Igo, and adding MoGo, CrazyStone, and maybe Leela. Maproom (talk) 09:57, 15 August 2008 (UTC) Maproom (talk) 09:57, 15 August 2008 (UTC)

Given that KCC Igo was originally plagiarized from Handtalk in 1998. (A whole paragraph could be given to this controversy, since it effectively killed computer Go sponsorship after the Ing Prize, and may even have resulted in the execution of the KCC Igo project manager!) Nevertheless, KCC Igo has evolved since then, winning all four of the Gifu Challenges. Silver Star has been the strongest Go program available for purchase for a number of years. (Evidence: these matches of commercial programs against GNU Go by the author of Aya.)

I have nothing against adding CrazyStone, MoGo, and Leela to the list of prominent programs (only Leela is available to the public). Or we can remove the list entirely, since there is a more complete list in a later section. --IanOsgood (talk) 17:50, 15 August 2008 (UTC)

"Why Humans are better at Go" - citations

98.203.230.92 has requested citations for the section "Why Humans are better at Go". As I wrote above, I can't find any citable statements to this effect, but David Fotland (programmer of Many Faces of Go) and Gunnar Farnebäck (principal programmer of GNU Go) have both publicly stated this view. Maproom (talk) 16:13, 27 October 2008 (UTC)

Well a citation of their words ought to be a reasonably satisfying solution then. As two experts in the field, one could hardly question their agreement.--ZincBelief (talk) 16:16, 27 October 2008 (UTC)

I don't remember their exact words. I have emailed them both, asking. Maproom (talk) 18:25, 27 October 2008 (UTC)

I got a good laugh out of the second sentence. To rephrase in another context: "Perhaps the world's shortest midget isn't really short -- all other midgets are tall." This section adds nothing useful and should be deleted in its entirety. And there is merit to the argument that humans can read long sequences better than computers, in fact it is preposterous. Numerous studies (e.g. Chase and Simon) have shown conclusively that the difference between strong players and weak players is NOT the ability to read further; strong players, like weak ones, see four or five possibilities as soon as the opponent moves, but they are better possibilities. This intuitive feature of human thought accounts for human success against computers. kibi (talk) 14:29, 28 October 2008 (UTC)

This is a neglected article really, it ought to be up at GA status or A status by now. I would agree that some extra thought needs to be put into this section though. Humans are able to make whole baord judgements, which computers lack. Programming a joseki is easy, choosing which joseki is hard. There must be other theories as well.--ZincBelief (talk) 14:35, 28 October 2008 (UTC)

To kibi. I don't follow your midgets analogy. There is an observation that needs explaining: computers are perceived as playing chess much better than they play go. This section explains the perception, as follows. Computers don't really play chess much better than the play Go, but they are much better at beating humans at it, because humans play Go better than they play chess. Maproom (talk) 22:58, 28 October 2008 (UTC)

Here are David Fotland's words on the subject of this section (from an email to me). Maproom (talk) 22:58, 28 October 2008 (UTC)

I agree with this, although I have used slightly different words.

Compared to other games like chess, the pieces in go do not move. This makes is much easier for people to visualize the board after many moves have been played. So compared to other games, people can look further ahead.

Talking it out

This text currently appears:

The main problem is that Go playing software has no standardized interface to communicate in a dialog with its opponents. So if there is a disagreement about the status of a group of stones, there is no general way for two different programs to “talk it out” and resolve the conflict.

This passage lead to a recent discussion at Board Game Geek. It appears this is true of the Japanese rules and the "solution" is to use Chinese rules. How should we best explain this?  Randall Bart   Talk  04:39, 4 May 2009 (UTC)

Using Chinese rules does not solve the problem, the players can still disagree on the status of a group after the game stop. One effective solution is to use Chinese rules and require that, after a disagreement, the players play on and actually capture all groups that they think are dead. Maproom (talk) 11:36, 4 May 2009 (UTC)

A requirement to play out disagreements is formalized in rulesets like Tromp-Taylor and New Zealand rules. IMHO, for computer vs computer play, these are far more appropriate than more traditional rulesets. Evand (talk) 14:56, 4 May 2009 (UTC)

This computer vs. computer information must be updated, it sounds quite old. As Evand says, not only do Tromp-Taylor/NZ/Chinese+playout rules formalize workable computer vs computer play, but they are in fact in use 24/7/365 on the KGS and CGOS servers where computers play each other in somewhere between thousands and tens of thousands of games a month without any human intervention (in addition to human vs computer games on KGS). So, clearly, this is NOT a significant problem anymore. I am going to change the article, so please let me know if for some reason I am misunderstanding the current state of computer go. --Seventhpath (talk) 22:37, 1 March 2011 (UTC)

I have updated the page. To further clarify my position, the previous writing in this section of the article made it seem as though the lack of Japanese rules in computer vs. computer games was some critical flaw or huge obstacle to Computer Go when in fact it is nothing of the sort. Any rule set that allows for ambiguity is not going to work for machines, obviously (unless we're talking some super-Watson AI). By the same token, it is trivial to demonstrate that I can write a bot that will ALWAYS disagree with whatever is suggested by its opponent, no matter how correct, if given the choice. This really has nothing to do with the strength of or usefulness of computer Go programs for playing games of Go.--Seventhpath (talk) 02:16, 2 March 2011 (UTC)

5x5 solved link

..is no longer available. Betaben (talk) 13:30, 30 October 2009 (UTC)

You have not discussed the far greater financial investment made into chess programs as a factor in determining their superior play against human professional players compared to Go software, e.g. IBM made the development of Deep Blue a major project with huge investment. As a result I think your conclusions could be distorted and misleading. — Preceding unsigned comment added by 79.47.49.25 (talk) 12:44, 27 August 2012 (UTC)

Wording of games with Handicap stones

It might be a good idea to reword some references to games played with handicap stones to be clear which side was being given an advantage: "in March 2013, Crazy Stone beat Yoshio Ishida in a 19×19 game with four handicap stones" doesn't indicate if Crazy Stone or Yoshi Ishida is the better player.

Yes, it's easy for those familiar with the field to forget that this won't be obvious to all readers. I've made several changes towards increased clarity. Maproom (talk) 19:53, 22 May 2013 (UTC)

investment and software/hardware comparison with chess engines

Hi. I have recently added a note on the general go page regarding the fact that there is no information regarding investment of resources and money as a factor in the superior performance of chess engines against professional players, i.e. IBM, Chessbase have made a significant investment in developing chess engines over a long period of time and there is no evidence to suggest the same sort of investment of money and resources for go. There is also no technical data about ply depth etc compared to chess engines. While recognizing that GO presents specific problems for software developers, it is clear that the comparatively small investment that has been made is an important factor to discuss, which I believe would greatly improve the quality of the article. — Preceding unsigned comment added by 79.43.48.105 (talk) 18:21, 18 May 2014 (UTC)

There is no technical data about ply depth because the currently successful Go engines do not use alpha-beta search. Their developers instead use "roll-outs per second" and "rollouts per move" as measures of performance. Maproom (talk) 07:39, 19 May 2014 (UTC)

Also, are you referring to investment in hardware, or in software? The system currently used by Zen has 50 cores (Deep Blue had 30), each running at 3GHz or more (Deep Blue's ran at 120MHz). So it is not lack of computing power that is holding Go engines back. Maproom (talk) 07:52, 19 May 2014 (UTC)

The investments by IBM and Chessbase came late in the chess engine development. Chessbase started work on fritz in the early 1990s. The IBM investment started in 1989 and stopped in 1997, a short period. In 1988, a computer defeated a grandmaster. Chessbase or IBM did not use their resources to reach the grandmaster level. A grandmaster is comparable to a go professional. The computer has not yet reached the go professional level. Corporate investments did not help the computer reaching grandmaster level. Similarly, corporate investments did not help the computer reaching go professional level. So far, corporate investments in chess and go are similar. Mschribr (talk) 15:55, 19 May 2014 (UTC)

cleaning up recent results

The "recent results" section looks like it has grown quite a bit over time. I have marked it as needing cleanup because, in my opinion, it seems overly repetitive. We do not need to have an exhaustive list of all the wins of computer go programs against humans, but key ones. For example, we could document the first on a 5x5, 9x9, 13x13, 19x19 board, with, without handicap, against amateur then professionnal players, and structure the narrative as such. Also, I have put a note in the main go talk page about this section, because it looks like the statement that go software can't play against professional players is starting to erode. In other words, a summary of the cleanup should also end up in the main go page (in Go_(game)#Software_players specifically). --TheAnarcat (talk) 13:48, 22 October 2014 (UTC)

Gongo

I was chatting about Go on IRC. A friend wondered if anyone had written a program in Go that plays Go. Yes! https://github.com/skybrian/Gongo — tbc (talk) 20:53, 30 September 2015 (UTC)

Short lead

The lead is very short. It should be easy enough to summarise the rest of the article in an additional couple of sentences. Tayste (edits) 01:19, 15 April 2016 (UTC)

Mick's Computer Go page

Link www.reiss.demon.co.uk/webgo/compgo.htm went dead sometime after 22Aug2015. — Preceding unsigned comment added by NanooGeek (talk • contribs) 19:36, 12 June 2016 (UTC)

Milton N. Bradley

Has written extensively on Go, such as "The Game of Go - The Ultimate Programming Challenge?", 1978 , p.89. — Preceding unsigned comment added by NanooGeek (talk • contribs) 19:49, 12 June 2016 (UTC)

References