Computer Go

Discuss anything you like about chess related matters in this forum.
MartinCarpenter
Posts: 2604
Joined: Tue May 24, 2011 10:58 am

Re: Computer Go

Post by MartinCarpenter » Sat May 27, 2017 9:43 am

I'm not sure if anyone has but the change has clearly been utterly enormous.

Computer go programs will definitely get better from here but the sort of really fancy stuff they're doing with AlphaGo won't, I think, scale as directly with extra processing power as early chess computers did.

NickFaulks
Posts: 5626
Joined: Sat Jan 02, 2010 1:28 pm

Re: Computer Go

Post by NickFaulks » Sat May 27, 2017 10:24 am

Christopher Kreuzer wrote:Has anyone looked in detail at just how much computers changed chess at all levels from amateur to the very top?
I do strongly suspect that computers have been substantially responsible for the stretching of the rating list over the past 25 years ( an effect which has been, and sometimes still is, erroneously described as inflation ). Simply put, training with computers can make you play chess better, but the better you are to begin with the more it helps.

John McKenna
Posts: 4137
Joined: Tue May 17, 2011 2:02 pm

Re: Computer Go

Post by John McKenna » Sat May 27, 2017 1:48 pm

MartinCarpenter wrote:I'm not sure if anyone has but the change has clearly been utterly enormous.

Computer go programs will definitely get better from here but the sort of really fancy stuff they're doing with AlphaGo won't, I think, scale as directly with extra processing power as early chess computers did.
There may be a quantum leap in processing power to come in the foreseeable future -

http://spectrum.ieee.org/computing/hard ... -computing

Who knows what that will do for AI?
To find a for(u)m that accommodates the mess, that is the task of the artist now. (Samuel Beckett)

Roger de Coverly
Posts: 18643
Joined: Tue Apr 15, 2008 2:51 pm

Re: Computer Go

Post by Roger de Coverly » Sat May 27, 2017 1:52 pm

Christopher Kreuzer wrote:Has anyone looked in detail at just how much computers changed chess at all levels from amateur to the very top?
Aren't we still using the methods used by the best prepared players of fifty years ago, Botvinnik, Fischer and Portisch springing to mind? That is to know what your opponents have played, what other people have played and to analyse your pet systems to exhaustion. Finding out what's been played before requires next to no legwork and analysis is far quicker with the aid of an engine. Tablebases that solve some endings for you are also an assistance.

The difference from the pre computer era perhaps is that whilst only a handful of players had the time and inclination to study chess to the depths of Botvinnik and others, nowadays there would be several thousand or tens of thousand.

Unlike what's suggested in Go, it's not so obvious that the engines have discovered whole new strategies. The recent use of "anti-theory" systems by the top players is only the same as used by Em Lasker and in the case of GMs using the Colle and London, only the same as used by amateur players.

User avatar
Michael Farthing
Posts: 1903
Joined: Fri Apr 04, 2014 1:28 pm
Location: Morecambe, Europe

Re: Computer Go

Post by Michael Farthing » Sat May 27, 2017 5:44 pm

John McKenna wrote:
There may be a quantum leap in processing power to come in the foreseeable future -

http://spectrum.ieee.org/computing/hard ... -computing

Who knows what that will do for AI?
Rather curiously, such a computer would lend itself to the Go problem rather more readily than the chess - the opposite to the experience with traditional digital machines. The real problem, however, may not even be the technology, but the development of the programming paradigm needed to use the technology.
Rejoiner

User avatar
David Shepherd
Posts: 855
Joined: Fri Nov 23, 2007 3:46 pm

Re: Computer Go

Post by David Shepherd » Wed Oct 18, 2017 8:14 pm

http://www.bbc.co.uk/news/technology-41668701 AlphaGo Zero (with just the rules and 72 hours to work out how to play optimally) beat AlphaGo 100-0 :o

MartinCarpenter
Posts: 2604
Joined: Tue May 24, 2011 10:58 am

Re: Computer Go

Post by MartinCarpenter » Wed Oct 18, 2017 8:46 pm

Now that is really depressing. Its one thing to beat us up, its another thing to rub our noses in it by showing precisely how much 'use' our centuries of hard won knowledge are!

That's some awfully clever algorithms they're using to do that learning from nothing - its really easy to get caught up in all sorts of local optima doing that.

Roger de Coverly
Posts: 18643
Joined: Tue Apr 15, 2008 2:51 pm

Re: Computer Go

Post by Roger de Coverly » Wed Oct 18, 2017 9:43 pm

David Shepherd wrote:with just the rules and 72 hours to work out how to play optimally
You wonder what would happen if you let it loose on chess. Would it agree with Fischer that 1. e4 was "best by test"?

NickFaulks
Posts: 5626
Joined: Sat Jan 02, 2010 1:28 pm

Re: Computer Go

Post by NickFaulks » Wed Oct 18, 2017 10:08 pm

Roger de Coverly wrote:You wonder what would happen if you let it loose on chess.
I was wondering that. Surely it would have already been tried.

Roger de Coverly
Posts: 18643
Joined: Tue Apr 15, 2008 2:51 pm

Re: Computer Go

Post by Roger de Coverly » Wed Oct 18, 2017 10:17 pm

NickFaulks wrote: Surely it would have already been tried.
Perhaps not with the same first principles approach. Programmers of chess engines from the very earliest have tried to use evaluation functions of various levels of sophistication and learn openings by look up.

I'm aware that there was an engine that confronted with the moves 1. e4 e5 2. Nf3 Nc6 3. Bb5 a6 was apt to play 4. Bc4 if left to its own devices. This for a while worked well against other engines whose programmers had not warned their creations about such sneaky tricks. The point being that whilst the engines had been told how to meet 1. e4 e5 2. Nf3 Nc6 3. Bc4, they didn't know what to do if the move .. a6 had been sneaked in.

A fundamentalist program would know for itself what the best move was in the opening.

MartinCarpenter
Posts: 2604
Joined: Tue May 24, 2011 10:58 am

Re: Computer Go

Post by MartinCarpenter » Thu Oct 19, 2017 8:12 pm

The problem with doing chess this way is more that the brute force approach is so very, very good at the game. Not much incentive to try and get fancy.

User avatar
IM Jack Rudd
Posts: 4043
Joined: Tue Apr 17, 2007 1:13 am
Location: Bideford

Re: Computer Go

Post by IM Jack Rudd » Thu Oct 19, 2017 8:38 pm

Well, the incentive is to learn about your AI programming, rather than to learn about chess.

MartinCarpenter
Posts: 2604
Joined: Tue May 24, 2011 10:58 am

Re: Computer Go

Post by MartinCarpenter » Thu Oct 19, 2017 9:44 pm

Sure, but it takes a good chunk of money & time to do this so they prefer to get publicity as per Alpha Go rather than 'we machine learned something that's nearly as good as Rybka'......
(That or applying to 'real' challenges.).

Roger de Coverly
Posts: 18643
Joined: Tue Apr 15, 2008 2:51 pm

Re: Computer Go

Post by Roger de Coverly » Thu Oct 19, 2017 11:29 pm

MartinCarpenter wrote:'we machine learned something that's nearly as good as Rybka'......
Agreed, but it would be interesting to see evaluations, particularly of openings which weren't influenced by the prejudices or prior opinions of its programmers. Up to a point contemporary GMs supply this, by playing ideas that engines allege to be dubious.

Peter Shaw
Posts: 185
Joined: Wed Apr 11, 2007 8:22 pm
Location: Wakefield

Re: Computer Go

Post by Peter Shaw » Wed Dec 06, 2017 11:51 am

AlphaGo learns chess and after 4 hours playing against itself it is good enough to beat Stockfish +28 =72 -0 :shock:

https://twitter.com/chess24com/status/9 ... 2951357441

In the 2 example black games it plays the Berlin defence. :lol:

Would love to see the early games. How long does it take for it to go from making random moves to something that looks like chess?

Post Reply