![]() |
Short AI article
http://www.timesonline.co.uk/tol/com...cle2002699.ece
Surely article is wrong and poker is a more difficult game for a computer to win at. Bots can obviously win against bad players and full handed tables but could one ever beat a top pro heads up? I'm guessing that will be done sooner at GO than in poker |
Re: Short AI article
From what I know, Poki is a fairly good program capable of beating most amateurs. I don't think Go programs are at that level yet.
|
Re: Short AI article
If you can buy in short enough there should be no problem to write a bot along the lines of "The Mathematics of Poker"
I'm also pretty confident that HU can be played somewhat optimally with deep stacks vs. a pro. Go o the other hand, well I wouldn't even know how to start writing the bot :P |
Re: Short AI article
How many combinations does it have to work through and figure out the EV for? That's the end-all question for these game bots.
For Go, it's huge, huge, huge. We're talking about a 19x19 board. Even if you exclude plays that are known bad most of the time we can only drop that to about a 17x17, so you can see that the computer will have a lot to think about for just a few moves ahead. The choice tree of poker, particularly limit poker (but even no-limit once you exclude known-irrational pot sizes), is much, much, much smaller. Bet, raise, fold? And once the community cards come out, the combinations of possible opponents holdings become relatively small, too, because you can generally lump together most kinds of hands that missed the flop. |
Re: Short AI article
Surely you underestimate the complexity of Go.
|
Re: Short AI article
Additionally, for Poker it's relatively easy to calculate your equity at a given point in the hand, and compare that with the pot size and a given bet. You do have to put your opponent on a hand range, but that's something a human can tune for a bot pretty easily.
For go, evaulating a board position isn't so easy. The number of interactions from stone to stone, group to group, are ridiculous. |
Re: Short AI article
Computers don't have the STONES to play Go, HAHAHAHAHA.
|
Re: Short AI article
interesting article
any places to play Go (free) online? and good recourses to learn to play? |
Re: Short AI article
learn to play is http://playgo.to/interactive/ A+++
yahoo is actually a good place to play quick beginner games. I'm still a beginner myself. |
Re: Short AI article
[ QUOTE ]
interesting article any places to play Go (free) online? [/ QUOTE ] IGS Pandanet is the best place once you're past the total beginner stage. There are top pros who play there from time to time even. It's free if you don't live in Japan. [ QUOTE ] and good recourses to learn to play? [/ QUOTE ]For the rules, I agree with Amp's link. From there what you want to do is play games on a 9x9 board with beginners, in conjunction with study of beginner's go problems. gobase.org has some of those, as well as goproblems.com. I recommended the Graded Go Problems for Beginners book series, though. They helped me quite a bit. I really need to play more Go. I seem to be going the year off/year on plan of improvement. Curse you werewolf, heh. Oh, and an old go saying: It's best to lose your first 100 games as soon as possible. |
Re: Short AI article
The beauty of go is that players of different skill levels can play with "penaltys" for the better player and have a fair match.
I really like that, means we can run some go tourney some time, yay [img]/images/graemlins/smile.gif[/img] Will learn the rules now. |
Re: Short AI article
[ QUOTE ]
The beauty of go is that players of different skill levels can play with "penaltys" for the better player and have a fair match. I really like that, means we can run some go tourney some time, yay [img]/images/graemlins/smile.gif[/img] Will learn the rules now. [/ QUOTE ]The handicaps start to break down once you hit a 10 stone gap, and of course only work if everyone gets themselves rated in the same system *cough*IGS*cough*, but yes, they often work out. I'd love to have a tournament, heh. |
Re: Short AI article
that ladder problem on the link amp posted is tilting me, help
edit* n/m- im just retarded |
Re: Short AI article
so I've got the IGS opened and and was watching a game- can you explain to me the rankings?
|
Re: Short AI article
Heh, ladders are a very basic go concept. Good that you got it together!
As for the rating system, it's based on a traditional system of ranks that is still used today somewhat. Not sure how to describe it simply, but here it goes: There's a dividing line at what is believed to be a major skill point. The players just above that line are at the 1 dan level. Players who are good enough to give a one stone handicap to 1d players are 2d. 3d players give a one stone handicap to 2d players. It goes up from there. The players below that line are 1 kyu. Players bad enough that the 1k players will give them a 1 stone handicap are 2k, 2k players give a 1 stone handicap to 3k players, and so on down the line. Also, the line is worth one stone. So a 1d players gives one handicap to a 1k player. Total beginners are somewhere in the neighborhood of 30k. I'm at 12k on IGS. A top pro is significantly beyond this amateur scale, though they have their own rank system where the entry level pros are 1d (often written 1p in internet contexts to distinguish from the much more often discussed amateur ratings) and the top pros are 9p. |
Re: Short AI article
As Neil says, in addition to the branching factor, the lack of an easy 'evaluation function' to determine mid-game positions has held back Go programmes to date. My understanding is that in the last year or 2 or so, a new technique of monte-carlo simulations, playing a game out to the end randomly a few times to see who wins, is proving a bit of a leap in the technology.
I'm ok at go - probably over the peak of the bell curve at any rate, but I would expect that the best computers would be able to beat me in 10 years for sure, quite possibly 5. They seem have had got much better in the last 5 years - from really terrible to ok. |
Re: Short AI article
|
Re: Short AI article
are there computers that can be played online?
|
Re: Short AI article
[ QUOTE ]
are there computers that can be played online? [/ QUOTE ] if you play on KGS, there are quite a few accounts that are bots - and displayed as such. There certainly used to be free to download computer opponents.... |
Re: Short AI article
[ QUOTE ]
are there computers that can be played online? [/ QUOTE ]GNU Go can be played online or off. |
Re: Short AI article
[ QUOTE ]
[ QUOTE ] are there computers that can be played online? [/ QUOTE ] if you play on KGS, there are quite a accounts that are bots - and displayed as such. There certainly used to be free to download computer opponents.... [/ QUOTE ] ... http://www.britgo.org/gopcres/playit.html mentions a few few applications, I think |
Re: Short AI article
[ QUOTE ]
Sensei's Library is an interesting wiki on Go. [/ QUOTE ]I used to read SL but eventually got turned off by the obsession of the community there with using Japanese words as frequently as possible. There was a bit of a KGS invasion too, if I recall. KGS I do NOT recommend because of poor moderation/poisonous culture. Far too cliquish for someone just trying to have fun with the game. Anyway, SL is not a BAD resource but it's not the best I think. All the foreign words just get in the way. |
Re: Short AI article
hmmm, I play on KGS not IGS because when I last looked the software was much better (albeit 4+ years ago). I don't mind the culture, but I do kinda know what you mean.
Sensei's Library contains a lot of very interesting stuff, IMO, but it can be hard to find. Oriental (esp Japanese) language does get used unnecessarily a fair amount in Go, although I wouldn't say that Sensei's Library was particularly worse than anywhere else. (Full disclosure, I learned to play Go in Japanese, so perhaps I am part of the problem [img]/images/graemlins/frown.gif[/img] ) |
All times are GMT -4. The time now is 01:36 PM. |
Powered by vBulletin® Version 3.8.11
Copyright ©2000 - 2024, vBulletin Solutions Inc.