Computer Go

Discuss anything you like about chess related matters in this forum.
MartinCarpenter
Posts: 3044
Joined: Tue May 24, 2011 10:58 am

Re: Computer Go

Post by MartinCarpenter » Thu Mar 10, 2016 11:42 am

I've no idea about the relative levels of perfection for Go vs chess. iirc rather more levels in the human playing strength heirarchy in Go than chess?

The learning definitely won't be the same, or as easy, as it was in chess. Of course, it will still be another source of very (super humanly in a few years) strong games to study and to try and learn from, but it'll take much more effort.

In chess the engines worked incredibly differently to human brains,. Also, ask a chess engine why it wants to play something? Lots of lines, numerical evaluations etc. Also very concrete explanations for mistakes - especially the very frequent tactical ones!

This is all data we can analyse, use and usefully guide/learn from. And even improve on, as per correspondence chess.

The operation of this go engine is buried in deep learnt neural nets, and getting explanations out those? It'll just shrug and tell you the move 'felt' right :)

User avatar
Christopher Kreuzer
Posts: 8806
Joined: Fri Aug 06, 2010 2:34 am
Location: London

Re: Computer Go

Post by Christopher Kreuzer » Thu Mar 10, 2016 11:54 am

Can you not suggest variations and 'learn' by having a 'conversation' (analysis) with the neural net? How do go players analyse anyway? I never understood the game enough to really work out what was going on, let alone being able to play back through a game. Maybe online go would help there if I ever go back to playing it.

Clive Blackburn

Re: Computer Go

Post by Clive Blackburn » Thu Mar 10, 2016 12:02 pm

Go is hugely popular in Japan but I have no idea what the media set-up is there.
Do they have live broadcasts of some of the big matches on the internet?
Is there live commentary with experts endeavoring to explain the moves to fans?

User avatar
Christopher Kreuzer
Posts: 8806
Joined: Fri Aug 06, 2010 2:34 am
Location: London

Re: Computer Go

Post by Christopher Kreuzer » Thu Mar 10, 2016 12:15 pm


MartinCarpenter
Posts: 3044
Joined: Tue May 24, 2011 10:58 am

Re: Computer Go

Post by MartinCarpenter » Thu Mar 10, 2016 12:21 pm

Think it'd be roughly equivalent to talking a largely intuitively based chess player who didn't offer any verbal explanations to try and explain things :) Suspect the learning process will probably have to go through human intermediaries analysing the games first.

The other thing to remember is that it isn't very likely that this engine will survive after this match - see deep blue/Hydra for instance. Its a useful test bed yes, but mostly a publicity exercise. Maybe a few more matches after this one to emphasise the point/let the humans try and find weaknesses etc.

It could be a very long time before computers every reach these levels again. The initial publicity hit will be gone of course, so it'd either need a commercial market to develop or to come down to the level that something like this sort of performance can be matched as a keen hobby project.

John McKenna

Re: Computer Go

Post by John McKenna » Thu Mar 10, 2016 2:26 pm

Clive Blackburn wrote:Go is hugely popular in Japan but I have no idea what the media set-up is there.
Do they have live broadcasts of some of the big matches on the internet?
Is there live commentary with experts endeavoring to explain the moves to fans?
There are regular Go (and separate Shogi) TV programmes on NHK - the Japanese equivalent of the BBC - every Sunday morning. They each last about an hour, or so, and are aimed at serious players - with master players playing and analysing.

Here's a review of an anime (Japanese for cartoon) about the perfect Go game -

http://www.tv.com/shows/hikaru-no-go/

NB: Lee Sedol, the Korean master, who is playing Alphago turned pro at twelve years of age. Perhaps he was possessed by the spirit of Go - like the boy in the cartoon - but he's now being dispossessed, literally, by the deus ex machina.

User avatar
Christopher Kreuzer
Posts: 8806
Joined: Fri Aug 06, 2010 2:34 am
Location: London

Re: Computer Go

Post by Christopher Kreuzer » Thu Mar 10, 2016 4:34 pm

Side-topic on chess masters who play(ed) go:

https://en.wikipedia.org/wiki/Edward_Lasker

Any others?

Also, John Nunn quote from here:

http://en.chessbase.com/post/computer-b ... ofessional

"I suspect that the 'hard for computers to master Go' thing is a bit of a myth. If the same effort had been put into computer Go as was put into chess, they would probably have reached this point a long time ago."

MartinCarpenter
Posts: 3044
Joined: Tue May 24, 2011 10:58 am

Re: Computer Go

Post by MartinCarpenter » Thu Mar 10, 2016 4:46 pm

I imagine most of the Chinese/Japanese (and that general geographical area) people have played it at least a bit :)

I share some of Nunn's scepticism, but it might be a bit misplaced with specifically Go. The stuff they're using in AlphaGo is all quite recent developments.

Its definitely true for stuff like Shogi, Xainqui and even Bridge etc.

David Robertson

Re: Computer Go

Post by David Robertson » Thu Mar 10, 2016 5:28 pm

Not sure I share the astonishment of people here. Sure, I can see the significance for Go players. And sure, it's another bruise on the amour propre of humanity. But an AI will win 5-0 at some point soon, possibly by the end of the week.

And as far as the AI community is concerned, that box will have been ticked. That's all. Defeating the world's best Go player will change nothing in the AI community, even as it gives us some further reason to ponder our limitations. Indeed, beating the world's greatest chess player twenty years back led exactly nowhere. Quite simply, by the late 1980s, the AI community had already realised that, in search of Artificial Intelligence, chess had little to offer. Don't believe me? It's here, in a stupendously informative book by Nick Bostrom (pp. 7-8)

Bostrom's point is that activities like Chess or Go are trivially easy for AI entities, just as they are enormously complex for us. The reason? The nature of information processing: the human brain has evolved to be astonishingly effective and efficient at parallel processing, but with limited capacity for serial processing (as required by chess and Go). By contrast, AIs have evolved profound competence with the latter, but continue to perform poorly at the former. The task of the AI community these days is to improve parallel processing capacity - technically, hard. (There is an additional issue of speed in processing [slow nerve conductivity v. hyperfast AIs] which I'll put aside for the moment).

In exploring the emergence of superintelligent entities, Bostrom takes the discussion far beyond the trivialities of chess- or Go-playing AIs. He asks how we are going to manage/control AIs hugely more intelligent than humans - measured by whatever properties we assign to 'intelligence'. If the outer limit of human IQ is 200, what is to be expected of an AI with an IQ of 20,000? (!!). Rendered in chess metrics, since the best chess AI currently has an Elo of 3200, what would chess look like if played by an AI of Elo 30,000 (never mind 300,000)? Obviously such an AI would find better things to do with its time than waste it on chess or Go.

And that's why, despite all the hoopla about chess and Go going down to AIs, it really isn't that big a deal. But it will be a very big deal indeed the day an AI performs tasks as complex as those our brain performs the minute we open our eyes each morning, tasks a million times more complex than any game of chess. Or Go.

MartinCarpenter
Posts: 3044
Joined: Tue May 24, 2011 10:58 am

Re: Computer Go

Post by MartinCarpenter » Thu Mar 10, 2016 6:55 pm

No, its a bit of a bigger deal than that.

Had they simply brute forced their way through the closed system of Go like they did chess, yes it'd signify nothing. As things stand, while Go is a closed system like chess, it is big enough that in practical terms, with our current technology, it might as well not be.

In the terms you're using above, this software is basically is using parallel processing. Better than us too, on quite a hard sort of strategic decision making problem :)

Still way off true AI of course! Quite a few sorts of jobs very definitely under threat in the medium term though. What we'll do about it as a society I'm not sure.

NickFaulks
Posts: 8453
Joined: Sat Jan 02, 2010 1:28 pm

Re: Computer Go

Post by NickFaulks » Sat Mar 12, 2016 8:55 am

All over.
If you want a picture of the future, imagine a QR code stamped on a human face — forever.

User avatar
Christopher Kreuzer
Posts: 8806
Joined: Fri Aug 06, 2010 2:34 am
Location: London

Re: Computer Go

Post by Christopher Kreuzer » Sat Mar 12, 2016 9:07 am

Wow. Was still rooting for Lee Sedol. Sad that he didn't win one of the first three. Impressive by AlphaGo and Deep Mind.

Links for games 2 and 3:

https://online-go.com/demo/114754
https://online-go.com/review/115995

What do those who play go here think of the actual games themselves?

(Even some of the top professionals are finding it hard to understand some of the moves!)

Clive Blackburn

Re: Computer Go

Post by Clive Blackburn » Sat Mar 12, 2016 9:47 am

That is a crushing victory, it sounds as though the third game was the closest so far.

http://www.theverge.com/2016/3/9/111843 ... ind-result

User avatar
Christopher Kreuzer
Posts: 8806
Joined: Fri Aug 06, 2010 2:34 am
Location: London

Re: Computer Go

Post by Christopher Kreuzer » Sat Mar 12, 2016 9:58 am

I found these reports:

http://www.zdnet.com/article/alphago-is ... -go-match/

"following his second loss on Thursday, Lee met with four of his fellow professionals and spent the night listening to their advice and trying to formulate a plan to defeat AlphaGo."

Didn't help, but: "The million dollar winning purse will now be donated to Unicef and STEM", so that is good.

http://www.telegraph.co.uk/technology/2 ... l-be-used/

"Lee said that machines would never appreciate the artistic flair that you need to win at go. But after his loss, he admitted that the machine’s elegance had thrown him. Maybe he had seen a glimmer of “intelligence”."

"This shocking win did not, perhaps, signal the end of humanity, but the dawn of Humans 2.0. We humans are uniquely optimistic."

Wonder if AlphaGo and its ilk will end up driving our cars? :D (And more, of course.)

(The first job of true AI will be to design even better AIs.)

EDIT: An interesting difference between this and the Deep Blue win: AlphaGo's calculations are "running through Google's cloud computing with its server located in the United States" (the match is in South Korea).

Another report:

http://www.wired.com/2016/03/third-stra ... -go-champ/

I'll just quote the conclusion: "Machines have conquered the last games. Now comes the real world."

An article from The Economist here:

http://www.economist.com/news/science-a ... -milestone

MartinCarpenter
Posts: 3044
Joined: Tue May 24, 2011 10:58 am

Re: Computer Go

Post by MartinCarpenter » Sat Mar 12, 2016 12:45 pm

They will most definitely end up driving our cars :) The question is how much else!

Wonder if he'll sneak one of the last two? It doesn't sound like its quite as comically stronger as chess engines are just yet, so maybe if they froze it they might find weaknesses and get somewhere. Not a fixed target though!