Page 2 of 4 FirstFirst 1234 LastLast
Results 16 to 30 of 49

Thread: Chess

  1. #16
    Join Date
    Jan 2006
    Posts
    2,649

    Default

    I think the reason that Go is so hard to do with a computer is that it has a great deal of long term strategic thinking and comparatively less short term tactical planning. Most good chess moves either have i) a positional impact that can be roughly evaluated immediately (gaining material, putting a rook on an open file, getting the king out of the centre, creating a strong pawn structure) or ii) a tactical impact that will become apparent in the next half dozen or so moves (ie it'll enable moves that have type i impact). Computers are generally very good at identifying ii) which makes up for them only being so-so at evaluating i).

    In contrast, as I understand it playing Go effectively requires much more subtle positional understanding - "if I put that stone there rather than one place to the right, will I be slightly overreaching myself so I can't properly secure the territory later on" - or possibly it has less tactical complications, so it's harder to make up for a slight lack of positional sophistication by using ferociously effective tactical calculation.

  2. #17
    Join Date
    Jun 2006
    Posts
    16,147

    Default

    Quote Originally Posted by Slothrop View Post
    In contrast, as I understand it playing Go effectively requires much more subtle positional understanding - "if I put that stone there rather than one place to the right, will I be slightly overreaching myself so I can't properly secure the territory later on" - or possibly it has less tactical complications, so it's harder to make up for a slight lack of positional sophistication by using ferociously effective tactical calculation.
    This sounds reasonable. Although it's also related to the much greater effective playing area in Go - nearly six times as many places for pieces to be placed - combined with the fact that most chess pieces can move across the board in one move, whereas Go stones only affect the points adjacent to them. So the difference between placing a stone on one point rather than an adjacent point could be pretty subtle, as you say, whereas in chess having a certain piece on this square, or that square next to it, can mean the difference between life or death.
    Doin' the Lambeth Warp New: DISSENSUS - THE NOVEL - PM me your email address and I'll add you

  3. #18
    Join Date
    Oct 2004
    Posts
    8,582

    Default

    There was a Go thread on Dissenssus a while back - ended up playing a few games online vs Sufi andd a couple of other guys from here. Headfuck of a game...I like it, but prefer chess by quite a distance, less headaches, more beautiufl combinations. Go defintiely more complicated though.

  4. #19
    Join Date
    Jan 2006
    Posts
    7,918

    Default

    "I think the reason that Go is so hard to do with a computer is that it has a great deal of long term strategic thinking and comparatively less short term tactical planning. Most good chess moves either have i) a positional impact that can be roughly evaluated immediately (gaining material, putting a rook on an open file, getting the king out of the centre, creating a strong pawn structure) or ii) a tactical impact that will become apparent in the next half dozen or so moves (ie it'll enable moves that have type i impact). Computers are generally very good at identifying ii) which makes up for them only being so-so at evaluating i)."
    I think that's roughly what I was trying to say above isn't it? We're in agreement I mean.

  5. #20
    Join Date
    Oct 2004
    Posts
    8,582

    Default

    The number of moves worth considering is so much higher in Go- in chess there're usually dozens of moves that can be eliminated almost instantaneously by a human player, but which will still eb stuied by a computer going methodically through everything (obvisouly computer design has improved greatly in recent years, so presumably there are shortcuts through this now). In Go it's so much more unclear, making the computer's ability to methodically do hundreds of claculations a second, way more valuable.

    Which is essentially (ok, pretty exactly, having re-read it) what is said above about the weight of strategy in Go, albeit in a slightly different way.

    All I really know is that Go makes my head hurt, and I'm not sure being a better player would make my head hurt any less.

    Also, as said above, to me Go has no equivalent to the beauty and variety of the combinational play in chess. After a few games i lose interest, and that's not so with chess. Best chess book I ever had was one by John Nunn where he broke down 24 (as I recall) of his best games...

    http://www.amazon.co.uk/Secrets-Gran...364922&sr=1-16 ah yes, this one. Perfect combination of tactics (he is/was an extremely tactical player, making for very lively games) and strategic insight, and a lot of real beauty to be seen in his play. great advert for chess, in the final analysis.
    Last edited by baboon2004; 29-06-2011 at 05:31 PM.

  6. #21
    Join Date
    Jan 2006
    Posts
    7,918

    Default

    "In Go it's so much more unclear, making the computer's ability to methodically do hundreds of claculations a second, way more valuable."
    But the whole point is that that's obviously not the case cos no computer can get anywhere near a good human player in Go. Number crunching isn't the solution, the computer apparently needs to "think" in a different way and no-one's figured out how to make it do it yet

    From wikipedia:

    Go poses a daunting challenge to computer programmers. While the strongest computer chess programs can defeat the best human players (for example, the Deep Fritz program, running on a laptop, beat reigning world champion Vladimir Kramnik without losing a single game in 2006), the best Go programs only manage to reach an intermediate amateur level. On the small 99 board, the computer fares better, and some programs now win a fraction of their 9x9 games against professional players . Human players generally achieve an intermediate amateur level by studying and playing regularly for a few years. Many in the field of artificial intelligence consider Go to require more elements that mimic human thought than chess.

  7. #22
    Join Date
    Oct 2004
    Posts
    8,582

    Default

    yes, true. Getting confused, and I blame alcohol for that. Damnit.

  8. #23
    Join Date
    Jun 2006
    Posts
    16,147

    Default

    If computers do better against good human players on a much smaller board, that would seem to suggest combinatorics is at least part of the problem for AI.
    Doin' the Lambeth Warp New: DISSENSUS - THE NOVEL - PM me your email address and I'll add you

  9. #24
    Join Date
    Jan 2006
    Posts
    2,649

    Default

    This is relevant:
    http://en.wikipedia.org/wiki/Minimax...lternate_moves
    http://en.wikipedia.org/wiki/Alpha-beta_pruning

    By and large, the clever part is the construction of a decent evaluation function to plug in at the end. Also coming up with heuristics to ignore branches that 'clearly aren't going anywhere'.
    Last edited by Slothrop; 29-06-2011 at 06:15 PM.

  10. #25
    Join Date
    Jan 2006
    Posts
    7,918

    Default

    From reading further down the page, it is combinatorics that is the problem - for computers at least. Basically in Chess and Go humans are better at dismissing the vast variety of moves and concentrating on the few worthwhile ones. In both games the number of possible moves is hugely greater than the worthwhile ones and as you project several moves forward this discrepancy grows exponentially. However it grows more quickly in Go and although advances in computing have allowed computers to win at chess by sheer number crunching they are still a long way from that in Go. In fact the size of the numbers seems to imply that it would be better to work on building a computer that could figure out how humans did the first bit than to hold out for something that will make Deep Thought look like a toy calculator. It would also be more interesting cos it's always a bit boring to solve something by just going through all the possibilities - I think that's how they proved the 4 Colour Theorem much to many people's disappointment.

  11. #26
    Join Date
    Jan 2006
    Posts
    2,649

    Default

    Presumably (from my limited experience) in Go, not only is the branching factor (ie the number of possible moves at any given time) greater, but the number of moves required for the advantage of a particular move to become apparent is greater. Which works out even worse for the computers...

  12. #27
    Join Date
    Jan 2006
    Posts
    7,918

    Default

    Yeah, it said that as well - a given move could have an effect many many moves in the future.

  13. #28
    Join Date
    Jan 2006
    Posts
    7,918

  14. #29
    Join Date
    Feb 2007
    Location
    London
    Posts
    504

    Default

    We did some stuff on the differences between Go and Chess in my degree (AI and Computer Science). It's pretty much the number of possible moves leading to huge probability trees as other people have mentioned. Probably as a result of this, the way humans deal with the two problems varies quite a lot. A grand master is supposedly thinking ~12 moves ahead or something, I forget the exact figures (which were obviously fairly vague estimates any way), but the point is they are actually considering all the available moves and plotting the optimal one, though obviously with lots of clever heuristic pruning. So the computer can emulate that style of reasoning pretty well. With Go though that sort of thing hasn't got a chance, and human players rely heavily on pattern matching. Recognising situations that resemble ones in previous games and generalising them mainly. When it comes to that sort of thing I'm afraid the computers are just rubbish.

    To continue this branching factor idea, I read a good little anecdote about Marion Tinsley (Best checkers player ever it seems. No, I'd never heard of him either) the other day. He was playing against some computer program and announced "ha you're going to regret that!" after it played a move. His move which capitalised on this mistake was 26 turns later! So I guess that supports the idea that reduced branching allows much deeper search for humans as well.

    Oh and I'm shit at chess btw.

  15. #30
    Join Date
    Feb 2007
    Location
    London
    Posts
    504

    Default

    I've played that Mancala game as well, though the one my mate had used glass beads not seeds. I wasn't great at that either. Completely trivial for a computer though, there's only n possible moves per go, where n is the number of pits on each side, so 6 - 10.

Posting Permissions

  • You may not post new threads
  • You may not post replies
  • You may not post attachments
  • You may not edit your posts
  •