[computer-go] Re: Strength of Monte-Carlo w/ UCT...
Hi there. I do agree with your point Robert Waite. I have yet seen no such paper as one that would prove that there is such thing as scalability based on any mathematical proofs. So all your points at criticizing the mathematical certainty of the scalability, is probably 100% right. There is no such things as mathematical certainty there. It can be modelized easily, as you already did : what if the the evaluation function is giving on purpose wrong data. How would one mathematically prove that it doesn't ? You would at a minimum have to know WHAT the evaluation function ACTUALLY exactly is ... In fact all the evidences that we have gathered about the scalability may rather been surprising to some persons : why in hell does all that works so well ? But, it's a proven fact that it does indeed works well so far. So that it seems perfectly natural to speak such phrases as there are evidences that given the hardware we got in twenty years, human will be beaten by current algorithms. I don't see how those evidences can be qualified with the term mathematical, but they are here (hiding among us !). Now if someone has the feeling that maybe there is a roadblock, it has to be considered for what it is : a personal intuition. What is this intuitions precisely based on ? Why are you trying to share it with us in the first place. For myself, i believe that what you are trying to do, is to begin to analyses all the data the community has gathered so far, trying to understand why indeed it worked so well that it even beaten out a pro with a 9 stones handicap and with as few as 1.7 million evaluations/second (running on some 800 hundreds cores). To the point that the pro felt he had no chances of wining at all with that much of a handicap. Your are trying to understand this, and are probably right on track for that goal. The term mathematical is very valuable to you, and you'll find it that it has a much wider use (on this list) than what you would like it to. But now, mathematics as proven to be of little use in the context of go programming lately. It's more of a physician world. You make up a (mathematical) model. You test it again reality via experimentations. You then get empirical certitudes that the model is indeed correct. There is no way of mathematically proving that light speed would still be constant if i chose to dance naked on the champs-Elysée some day. You'll definitely find no paper on that. Yet to speak of it as mathematically certain, is probably not as wrong as it sound. But as it is, i'm playing the devil advocates here. I'm totally agreeing with you. I found your way to fight irrationnality very interesting indeed. It's been very refreshing. - Robert Waite has wrote : I would really like to see what paper you are referring to. Do you mean Bandit based Monte-Carlo Planning? Please post the name of the paper which you are referring to. I do not think that the empirical evidence is overwhelming that it is scalable in a practical way for the problem of beating a human. Now the topic has moved to scalable to beat a human and I disagree with the interpretation of the data. We are both interpreting data. Your data doesn't count as a theory.. where you reduced my theory to one that has no data. We are both interpreting the same data. Diminishing returns was just an example of something that could be a roadblock. I was questioning how this necessarily scales to humans. It seems more data is needed from MC-programs vs. humans to make a rigorous theory of scalability. So far.. the only scalability that seems proven is a case for solving the game... not beating humans. There is some point between that would most likely in my opinion lead to humans being beaten.. some amount of calculation before you solved it.. but the shape of this curve is something I am unsure of. It doesn't seem that unreasonable to question if there is a practical scalability. _ Retouchez, classez et partagez vos photos gratuitement avec le logiciel Galerie de Photos ! http://www.windowslive.fr/galerie/___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
[computer-go] Re: mogo beats pro!
How long will it be until a computer system reach pro level play ? (answering Bob Hearn question) Maybe that rather than taking the raw speed of hardware as a reference, we could use the raw number of simulation (per second) as a base of speculation. Assuming it's a fixed game time with the same setting than the 9 stone game that was played. + So first question is : how much more 1.7 millions simulations can yield in strength than what it does in today's mogo version ? Mogo has been in a very intense state of development lately. I have no figures about the scaling of the efficiency per simulation, but it has to be taken into account for trying to guess what it will all look likes in ten years. + the second question is : how much more will the algorithm be tuned to the hardware (or the hardware to the soft). I guess the theoretical throughput of the system mogo was run onto is MUCH more than 1.7 millions simulations per second (taking into account all the tree logic). Given the exact same property of the simulation that were used, i'd think that in 10 years times, given nothing more to do ... it could easily reach 17 millions simulations per second on the same system. + the third question is : will moore law olds ? So far moore law was linked to mono-threading paradigm. It may be true that super-computer didn't improved by much (i don't know about that to much). But there are also evidences that with the explosion and democratisation of multiprocessors, the moore law will not held in it's current form. (GPU card have had there efficiency increased but much more than x2 every two years) + the fourth question being : will there be much more efficient go-solving method birthing in the next few years. And also, will not mogo get access to more and more powerfull hardware as it'll close up to pro level ... (lately the top raw computing power accessible to mogo program seems to have increased by a lot ... was it running on a monocore 1ghz pentium 3 years ago ?) It's true that ten years is a short span of time indeed. It may seems a bit optimistic to hope for kim to fall in a fair even-game given 1day per move seting in ten years for now. But i wouldn't call that totally irealistic either. I guess i would easily put a bet on it, if the odds were about 1/20, me getting 16 times the amount of money i have bet if a go-program reaches pro-level on 19x19 toward the end of the year 2018. I'd probably go as far as 1/3 for the end of the year 2028. -- Another interesting question is : how much time before mogo can tackle the three-stone handicap game against a pro ? I have read somewhere that asked how much stones he would need for a sure win against god himself if he's life was at stake, the pro answered : three stones. Bob Hearn wrote : So -- quick back-of-the-envelope calculation, tell me where I am wrong. 63% win rate = about half a stone advantage in go. So we need 4x processing power to increase by a stone. At the current rate of Moore's law, that's about 4 years. Kim estimated that the game with MoGo would be hard at 8 stones. That suggests that in 32 years a supercomputer comparable to the one that played in this match would be as strong as Kim. This calculation is optimistic in assuming that you can meaningfully scale the 63% win rate indefinitely, especially when measuring strength against other opponents, and not a weaker version of itself. It's also pessimistic in assuming there will be no improvement in the Monte Carlo technique. But still, 32 years seems like a surprisingly long time, much longer than the 10 years that seems intuitively reasonable. Naively, it would seem that improvements in the Monte Carlo algorithms could gain some small number of stones in strength for fixed computation, but that would just shrink the 32 years by maybe a decade. How do others feel about this? I guess I should also go on record as believing that if it really does take 32 years, we *will* have general-purpose AI before then. _ Pendant tout l'été, consultez vos emails Hotmail sur votre mobile ! http://www.messengersurvotremobile.com/?d=hotmail___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] What's happening at the European Go Congress?
In message [EMAIL PROTECTED], Gian-Carlo Pascutto [EMAIL PROTECTED] writes Hi all, there doesn't seem to be any news from the European Go Congress. Nevertheless, I see that partial results were posted: 19 x 19 Results 1stCrazy Stone 6/6 2ndLeela 5/6 3rdMany Faces of Go4/6 9 x 9 Results 1stLeela 4/5, SoDOS=13 2ndCrazy Stone 4/5, SoDOS=12 Sorry to have been taking so long over this. I am still working on my report. Also I see: Thursday August 7th about 19:00 (17:00 GMT)Demonstration 9×9 game between winning 9x9 program (Leela) and professional. This game should be played via KGS. What happened in this game?? First, I know that the 19x19 demonstration game between CrazyStone and a professional never happened. The pro showed a lack of enthusiasm, and did not turn up in the room at the time it was meant to happen. I left Leksand before the 9x9 game between Leela and a pro was scheduled, but I have seen no report of it, and suspect that it suffered the same fate. Nick -- Nick Wedd[EMAIL PROTECTED] ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
[computer-go] Re: What's happening at the European Go Congress?
Hello all the European Go Congress was a little short of organizers, it seems, as Sweden is a small country, so some of us who had planned to work on the web site were shifted to work with registration, info-desk and other vital tasks. This has led to some delays in reporting the results. I apologize. The results from 19x19: http://www.gokgs.com/tournEntrants.jsp? sort=sid=407 and from 9x9: http://www.gokgs.com/tournEntrants.jsp?sort=sid=408 It should come up on our website too, but I guess the KGS-pages will do fine until our webmasters have grabbed a 48-hour nap. :) Xiao Ai Lin, 1p vs LeelaBot This game did happen. It was not meant as a challenge, but as a friendly game to get an idea of what can be done to develop the leading programs on 9x9. It was relayed to the cinema-screen as a warm-up before MoGo's game. I will be back with the review as an SGF-file, that is what I managed to note from her review. Best Basti Weidemyr In message [EMAIL PROTECTED], Gian-Carlo Pascutto [EMAIL PROTECTED] writes Hi all, there doesn't seem to be any news from the European Go Congress. Nevertheless, I see that partial results were posted: 19 x 19 Results 1stCrazy Stone 6/6 2ndLeela 5/6 3rdMany Faces of Go4/6 9 x 9 Results 1stLeela 4/5, SoDOS=13 2ndCrazy Stone 4/5, SoDOS=12 Also I see: Thursday August 7th about 19:00 (17:00 GMT)Demonstration 9×9 game between winning 9x9 program (Leela) and professional. This game should be played via KGS. What happened in this game?? ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] Re: What's happening at the European Go Congress?
Xiao Ai Lin, 1p vs LeelaBot This game did happen. It was not meant as a challenge, but as a friendly game to get an idea of what can be done to develop the leading programs on 9x9. It was relayed to the cinema-screen as a warm-up before MoGo's game. I will be back with the review as an SGF-file, that is what I managed to note from her review. Thanks. I tried to analyze with Leela, but it thinks for a long time black still has chances and only starts dropping a bit after a long think. It would not have resigned in this position. Looking at the SGF I see white was about to lose on time. I have the nagging feeling Leela's operator resigned on behalf of the program to prevent the computer from winning on time in what was probably an objectively a lost position. -- GCP ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] Re: What's happening at the European Go Congress?
In message [EMAIL PROTECTED], Gian-Carlo Pascutto [EMAIL PROTECTED] writes Xiao Ai Lin, 1p vs LeelaBot This game did happen. It was not meant as a challenge, but as a friendly game to get an idea of what can be done to develop the leading programs on 9x9. It was relayed to the cinema-screen as a warm-up before MoGo's game. I will be back with the review as an SGF-file, that is what I managed to note from her review. Thanks. I tried to analyze with Leela, but it thinks for a long time black still has chances and only starts dropping a bit after a long think. It would not have resigned in this position. Looking at the SGF I see white was about to lose on time. I have the nagging feeling Leela's operator resigned on behalf of the program to prevent the computer from winning on time in what was probably an objectively a lost position. When I look at the game record, I see that at the end, the pro has 7:59 left, Leela 4:25. And Black is totally lost: White will capture the d4 group which only has two liberties, connecting her three groups which already have at least four liberties each, and leaving Black's b2 and b7 groups dead. Nick -- Nick Wedd[EMAIL PROTECTED] ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] mogo beats pro!
From: Bob Hearn [EMAIL PROTECTED] Now, my question. Sorry if this has already been beaten to death here. After the match, one of the MoGo programmers mentioned that doubling the computation led to a 63% win rate against the baseline version, and that so far this scaling seemed to continue as computation power increased. So -- quick back-of-the-envelope calculation, tell me where I am wrong. 63% win rate = about half a stone advantage in go. So we need 4x processing power to increase by a stone. At the current rate of Moore's law, that's about 4 years. Kim estimated that the game with MoGo would be hard at 8 stones. That suggests that in 32 years a supercomputer comparable to the one that played in this match would be as strong as Kim. This calculation is optimistic in assuming that you can meaningfully scale the 63% win rate indefinitely, especially when measuring strength against other opponents, and not a weaker version of itself. It's also pessimistic in assuming there will be no improvement in the Monte Carlo technique. But still, 32 years seems like a surprisingly long time, much longer than the 10 years that seems intuitively reasonable. Naively, it would seem that improvements in the Monte Carlo algorithms could gain some small number of stones in strength for fixed computation, but that would just shrink the 32 years by maybe a decade. How do others feel about this? I guess I should also go on record as believing that if it really does take 32 years, we *will* have general-purpose AI before then. I suspect that Mogo -- good as it is -- is far from being the optimal algorithm. In ten years time new methods will emerge which should yield considerable improvements. In addition, the 800-core supercomputer used was not today's state of the art; the Mogo team almost obtained a 3000-core supercomputer for this exhibition, which would be nearly 4x as large; as Computer Go becomes more exciting, we may be able to borrow still more impressive hardware -- current state-of-the-art is 65k or even 128k processors. Third, the 32 year figure is highly sensitive to one's expectation of Moore's Law. A doubling every 18 months would be a quadrupling every 36 months, which is three years; this factor alone shrinks the 32 years to 24. We may see a faster rate of growth - GPUs have been improving faster than general-purpose CPUs, and the coming multicore processors may have more in common with GPUs than with previous generations of X86 cores -- we may revert to simpler RISC cores, which use less silicon. In short, reaching the top of the pyramid would be a thousand-fold improvement in processing power -- about 4 to the 4th power, or half way to the goal. During the same period, the petaflops race and Moore's Law would continue to increase the power of the Top 500. Stir in some algorithmic improvements, and we should be within range in something closer to ten, not 32 years. If general purpose AI means an AI which can solve every problem at the expert level, that is probably not a prerequisite for solving one problem at an expert level. We're not asking for a program which can skillfully play a teaching game against a weaker player, as a human pro would, nor are we asking that it be able to dance the salsa; it just needs to beat a pro in an even game. We have just barely started optimizing the search. What do humans know that computers don't? How do pros manage to play well without the ability to examine trillions of playouts? ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] Re: What's happening at the European Go Congress?
In message [EMAIL PROTECTED], When I look at the game record, I see that at the end, the pro has 7:59 left, Leela 4:25. And Black is totally lost: White will capture the d4 group which only has two liberties, connecting her three groups which already have at least four liberties each, and leaving Black's b2 and b7 groups dead. Hi, this is another game! The game you posted and the one on KGS are totally different. In the one on KGS, black played with reduced komi and (as far as I can tell) held out a long time until white was about to forfeit on time. In the one you posted, the opponent doesn't appear to be a pro (sestir 2d instead of egc1p), no handicap/modified komi was used, and black lost quickly. -- GCP ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
[computer-go] Games vs professionals
- The review of Xiao Ai Lin vs Leela: http://www.weidemyr.com/egc/cg/ XiaoAiLin_Leela-review.sgf - Several people at the congress expressed worries to me about what would happen to the sport Go, if computer programs became stronger and threatened to defeat the strongest men. Go would lose its advantage over chess, they said, and people would feel redundant as computers could do it better. One man asked me repeatedly to quit running challenges between professionals and computers. The professionals themselves became very nervous when we asked them to play against a computer. It is not hard to imagine the bold headlines after losing, but it is hard to imagine them after winning. The game between MoGo and Kim Myung Wan was unique, since MoGo run on a large cluster and interesting to watch. (Congratulations MoGo team!) It was also a great way of showing people the progress that has been made in computer-go recently. However, maybe we do not need to use these kinds of challenges as a means of getting media attention. We would like to find a way to cooperate with the traditional go- community with little friction. What do you think? Best regards Basti Weidemyr kgs: sestir ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] Re: What's happening at the European Go Congress?
In message [EMAIL PROTECTED], Gian-Carlo Pascutto [EMAIL PROTECTED] writes In message [EMAIL PROTECTED], When I look at the game record, I see that at the end, the pro has 7:59 left, Leela 4:25. And Black is totally lost: White will capture the d4 group which only has two liberties, connecting her three groups which already have at least four liberties each, and leaving Black's b2 and b7 groups dead. Hi, this is another game! The game you posted and the one on KGS are totally different. In the one on KGS, black played with reduced komi and (as far as I can tell) held out a long time until white was about to forfeit on time. In the one you posted, the opponent doesn't appear to be a pro (sestir 2d instead of egc1p), no handicap/modified komi was used, and black lost quickly. sestir is Basti Weidemyr, who was in charge of arranging the challenge game. He has just posted to this list, so I hope he will explain what happened. Looking at LeelaBot's games on KGS since the tournament, I see only two: the one I posted, against sestir, and one against egc1p with 0.5 komi, which I cannot open, as it was not finished by the players and KGS is treating it as escaped. Nick -- Nick Wedd[EMAIL PROTECTED] ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] Re: What's happening at the European Go Congress?
Nick Wedd wrote: Looking at LeelaBot's games on KGS since the tournament, I see only two: the one I posted, against sestir, and one against egc1p with 0.5 komi, which I cannot open, as it was not finished by the players and KGS is treating it as escaped. Nick The link I sent yesterday works for me: http://files.gokgs.com/games/2008/8/7/egc1p-LeelaBot.sgf Rémi ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] Games vs professionals
On 11-aug-08, at 08:56, Basti Weidemyr wrote: However, maybe we do not need to use these kinds of challenges as a means of getting media attention. We would like to find a way to cooperate with the traditional go- community with little friction. What do you think? We come in peace! ;-) I must say I have little sympathy for people who think that computers are destroying their game. They should ask themselves, deep in their heart, why they're playing Go (or any other intellectual game). Most people play for fun. The fun of measuring themselves against other people and the fun of intellectual discovery that comes with it. Those things won't change. Some people play because of a certain social status it gives them, whether justified or not. Those people may feel threatened that this status will now be looked less upon when computers start challenging them. But it's no different from just about any human activity that gets trivialized by industrial or scientific progress. People will have to adapt to it, and they will. Desperately trying to shut it out does nobody any service and to me falls in the same category as people trying to force Galileo to recant his theories in the past. We get good things in return for progress. There may be a few small sacrifices in return. Maybe professionals feel threatened. But it's because as a group they live in an ivory tower if there ever was one. If they refuse to play computers it's like sticking their heads in the sand. It doesn't really surprise me, for example the Nihon Ki-in is nothing less than a medieval guild in modern times and sooner or later reality will pass them by. Chess is still played competitively and professionally. The same will be for Go for a long time to come. But yes, at some point it's likely some things will change. But most likely it will be for the better. Sorry if this comes across as a bit of a rant, I guess I'm a bit allergic to extreme conservatism. Mark ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] Re: What's happening at the European Go Congress?
On Mon, 2008-08-11 at 12:40 +0200, Gian-Carlo Pascutto wrote: Xiao Ai Lin, 1p vs LeelaBot This game did happen. It was not meant as a challenge, but as a friendly game to get an idea of what can be done to develop the leading programs on 9x9. It was relayed to the cinema-screen as a warm-up before MoGo's game. I will be back with the review as an SGF-file, that is what I managed to note from her review. Thanks. I tried to analyze with Leela, but it thinks for a long time black still has chances and only starts dropping a bit after a long think. It would not have resigned in this position. Looking at the SGF I see white was about to lose on time. I have the nagging feeling Leela's operator resigned on behalf of the program to prevent the computer from winning on time in what was probably an objectively a lost position. I hope that didn't happen. Otherwise, should have not played with clocks. - Don ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] Re: mogo beats pro!
On 10-aug-08, at 17:24, Don Dailey wrote: Of course there is also the possibility of some exciting new hardware breakthrough around the corner that doesn't just extend Moore's law, but blows it out of the water. Of course there's that possibility. But I'm actually wondering if we wouldn't rather be seeing the opposite. Moore's law seems to have stalled for a few years, only to gain traction again with multiple core designs. But unlike previous advances in computing power, multiple processing is not as easily available to all software alike. People are already asking me (and themselves) what I need an 8-core computer for. Unless we also see some good progress in software development, 99% of people will have no use for a 1,000 CPU computer, either privately or professionally. Game developers already struggle to use the Playstation's cell architecture to its full potential. If that remains the case then the type of super-computers that MoGo ran on will stay in the domain of extreme scientific research isolated to very special purposes for a long time. So while I think it's definitely possible that massively parallel computing can still progress at a fast pace, the fact that it will become dependent on similar progress in the software field makes it a quite a bit less likely to happen at the same speed as in the previous decades, IMO. Because in the end it's the needs of the masses that drives the real progress of computing speed. This is all in the realm of speculation of course, and I'd just as happily be proven wrong on this. Mark ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] Games vs professionals
We would like to find a way to cooperate with the traditional go- community with little friction. What do you think? We should cooperate with the Go community as it concerns rules and whether computers are allowed to compete. We should never pressure players to play against computers. It's completely understandable that some will feel threatened as computers continue to improve. But it's really not a threat to the game in any way. If you look at other games, you will see that computers did not hurt them, if anything they helped advertise the game. You will see that the chess community, as well as the computer chess community is thriving. The checkers community was never huge, but it has not been damaged either. I also think they have to be reasonable. If one man asks you to stop, he cannot speak for the whole GO community. He can speak for himself. Everything is in the presentation. I have found there are ways to deal with resistance, ask any chess programmer who has been around more than a decade. Always remember that some will be fearful and show them respect and dignity. They have their reputation and they know they are fallible and mortal even when others exaggerate their greatness. Oh, and money never hurts. It shows respect for them because they are, after all, professionals which means it is how they make a living. Whether they cooperate or not, computers will continue to improve so it's not like anything they will do really affects that. - Don On Mon, 2008-08-11 at 13:56 +0200, Basti Weidemyr wrote: - The review of Xiao Ai Lin vs Leela: http://www.weidemyr.com/egc/cg/ XiaoAiLin_Leela-review.sgf - Several people at the congress expressed worries to me about what would happen to the sport Go, if computer programs became stronger and threatened to defeat the strongest men. Go would lose its advantage over chess, they said, and people would feel redundant as computers could do it better. One man asked me repeatedly to quit running challenges between professionals and computers. The professionals themselves became very nervous when we asked them to play against a computer. It is not hard to imagine the bold headlines after losing, but it is hard to imagine them after winning. The game between MoGo and Kim Myung Wan was unique, since MoGo run on a large cluster and interesting to watch. (Congratulations MoGo team!) It was also a great way of showing people the progress that has been made in computer-go recently. However, maybe we do not need to use these kinds of challenges as a means of getting media attention. We would like to find a way to cooperate with the traditional go- community with little friction. What do you think? Best regards Basti Weidemyr kgs: sestir ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/ ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] Re: mogo beats pro!
On Mon, 2008-08-11 at 10:19 -0300, Mark Boon wrote: On 10-aug-08, at 17:24, Don Dailey wrote: Of course there is also the possibility of some exciting new hardware breakthrough around the corner that doesn't just extend Moore's law, but blows it out of the water. Of course there's that possibility. But I'm actually wondering if we wouldn't rather be seeing the opposite. Moore's law seems to have stalled for a few years, only to gain traction again with multiple core designs. But unlike previous advances in computing power, multiple processing is not as easily available to all software alike. Yes, moores law isn't a smooth curve but a jagged one. Processors are still continuing to get faster, but at a slower rate. People are already asking me (and themselves) what I need an 8-core computer for. Unless we also see some good progress in software development, 99% of people will have no use for a 1,000 CPU computer, either privately or professionally. Game developers already struggle to use the Playstation's cell architecture to its full potential. If that remains the case then the type of super-computers that MoGo ran on will stay in the domain of extreme scientific research isolated to very special purposes for a long time. Of course you and I can never get enough. Give me 1 million processors and I would put them to work. How about testing? I could rate a variety of programs ACCURATELY in seconds with a million processors! I think software will catch up. This is all new and we are not at a point where very many people have 2 cores although it's now entry level. So in a year or two most people will have at least 2 cores. I think 4+ cores has to become really common before the pressure to build software to utilize it gains a lot of traction. There is also the issue that some algorithms are so serial in nature that they cannot benefit. So in a way it's bad news that Moores law has shifted to more processors. But in many cases the algorithms will have to change. There is already a very slow and very gradual shift to languages that can take advantage of parallel processing. A lot of people are now thinking about this so it will happen. But first the hardware has to be there and the pressure to do it will be overwhelming soon. - Don So while I think it's definitely possible that massively parallel computing can still progress at a fast pace, the fact that it will become dependent on similar progress in the software field makes it a quite a bit less likely to happen at the same speed as in the previous decades, IMO. Because in the end it's the needs of the masses that drives the real progress of computing speed. This is all in the realm of speculation of course, and I'd just as happily be proven wrong on this. Mark ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] Re: What's happening at the European Go Congress?
On Mon, 2008-08-11 at 14:26 +0100, Nick Wedd wrote: In message [EMAIL PROTECTED], Gian-Carlo Pascutto [EMAIL PROTECTED] writes this is another game! The game you posted and the one on KGS are totally different. In the one on KGS, black played with reduced komi and (as far as I can tell) held out a long time until white was about to forfeit on time. In the one you posted, the opponent doesn't appear to be a pro (sestir 2d instead of egc1p), no handicap/modified komi was used, and black lost quickly. In my curiousity to see the right game (which KGS would not let me do because it was treating it as escaped), I have done something foolish. I am admitting this here to get the record straight. I logged in to KGS using LeelaBot's account, and opened (and saved) the game. The game was still running, so there can have been no resignation. LeelaBot had over a minute left, I think less than 80 seconds but I don't remember exactly. The pro had three seconds left. This was foolish of me because I had resumed the game, and was allowing LeelaBot's time to pass. I have carelessly destroyed the evidence of LeelaBot's remaining time. There is now only my word (and perhaps the operator's) for my claim that LeelaBot had more than a minute left. I think you will be forgiven. To err is human. Nick ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] Re: What's happening at the European Go Congress?
In message [EMAIL PROTECTED], This was foolish of me because I had resumed the game, and was allowing LeelaBot's time to pass. I have carelessly destroyed the evidence of LeelaBot's remaining time. There is now only my word (and perhaps the operator's) for my claim that LeelaBot had more than a minute left. No worries :) I saved the game earlier today after Remi posted the link and before you resumed it. It is included in attachement. Leela scores this at about 30% winning chances for itself after a long think. I have no idea whether that's a reasonable assesement. -- GCP(;GM[1]FF[4]CA[UTF-8]AP[CGoban:3]ST[2] RU[Chinese]SZ[9]KM[0.50]TM[900] PW[egc1p]PB[LeelaBot]BR[2k]DT[2008-08-07]PC[The KGS Go Server at http://www.gokgs.com/]C[LeelaBot [2k\]: GTP Engine for LeelaBot (black): Leela version 0.3.14 ] ;B[ee]BL[863.668] ;W[ge]WL[876.158] ;B[ff]BL[799.072] ;W[ed]WL[819.939] ;B[dd]BL[739.277] ;W[ec]WL[786.111] ;B[cc]BL[683.916]C[sestir [2d\]: This game is with reduced komi - a handicap to the robot. ] ;W[cf]WL[617.097] ;B[ce]BL[632.695] ;W[df]WL[613.487] ;B[de]BL[585.351] ;W[fe]WL[562.183]C[sestir [2d\]: white: Xiao Ai Lin, 1p sestir [2d\]: black: Leela - the winner of yesterday's computer-go tournament on 9x9 ] ;B[ef]BL[541.529] ;W[dh]WL[554.55] ;B[gc]BL[518.749] ;W[fc]WL[535.634] ;B[hd]BL[499.271] ;W[gf]WL[399.993] ;B[gg]BL[461.913] ;W[hg]WL[389.159] ;B[eh]BL[427.467] ;W[be]WL[317.222] ;B[gh]BL[409.83] ;W[gb]WL[290.491] ;B[hb]BL[394.384] ;W[gd]WL[269.866] ;B[hc]BL[379.504] ;W[fb]WL[265.178] ;B[he]BL[365.168] ;W[hf]WL[262.259] ;B[bg]BL[351.413] ;W[bf]WL[229.446] ;B[ch]BL[325.177] ;W[di]WL[224.616] ;B[bh]BL[300.836] ;W[dg]WL[212.215] ;B[bd]BL[278.386] ;W[ag]WL[153.546] ;B[hh]BL[264.237] ;W[ha]WL[121.016] ;B[ie]BL[244.544] ;W[ih]WL[93.559] ;B[ah]BL[226.347] ;W[ae]WL[81.18] ;B[ga]BL[217.757] ;W[fa]WL[75.826] ;B[ib]BL[201.546] ;W[db]WL[63.589] ;B[cb]BL[186.553] ;W[ga]WL[50.871] ;B[da]BL[172.73] ;W[if]WL[44.983] ;B[ca]BL[159.825] ;W[eg]WL[41.384] ;B[fg]BL[153.781] ;W[ei]WL[38.3] ;B[gi]BL[142.393] ;W[fh]WL[34.496] ;B[bi]BL[131.84] ;W[fi]WL[29.929] ;B[eb]BL[126.443] ;W[ia]WL[25.454] ;B[ad]BL[119.513] ;W[ea]WL[22.432] ;B[dc]BL[111.198] ;W[ic]WL[19.499] ;B[af]BL[102.975] ;W[id]WL[15.899] ;B[hb]BL[99.063] ;W[ag]WL[11.939] ;B[bc]BL[91.754] ;W[af]WL[9.6] ;B[hd]BL[88.284] ;W[ii]WL[7.438] ;B[eb]BL[81.849] ;W[ig]WL[5.826] ;B[db]BL[75.919] ;W[hi]WL[3.82]C[sestir [2d\]: starts review RemiCoulom [5k\]: Well done, human player ! sestir [2d\]: indeed sestir [2d\]: review in demo ])___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] Re: What's happening at the European Go Congress?
Gian-Carlo Pascutto wrote: In message [EMAIL PROTECTED], This was foolish of me because I had resumed the game, and was allowing LeelaBot's time to pass. I have carelessly destroyed the evidence of LeelaBot's remaining time. There is now only my word (and perhaps the operator's) for my claim that LeelaBot had more than a minute left. No worries :) I saved the game earlier today after Remi posted the link and before you resumed it. It is included in attachement. Leela scores this at about 30% winning chances for itself after a long think. I have no idea whether that's a reasonable assesement. If I am not mistaken, bottom left is seki. This is probably what Leela misunderstood. And it may also be what you don't understand. The game look like an obvious win for W, starting from move 62. So it looks very fair that Leela did not win on time. Rémi ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
[computer-go] Re: What's happening at the European Go Congress?
Leela had 1 minute, 15 seconds and 919/1000 of a second left, according to the game-record. egc1p had 3.82 seconds left. What happened is still unclear and I do not know. It seems the professional had never played go on a computer before, at least not on KGS, so yes, we should probably have used longer time- settings, and explained that the robot would play plenty of unnecessary moves after filling dame. As time was running out and the robot played obstinate moves, I told the operator to kill it. However, it looked to me like he never touched the keyboard, so when a dialog appeared, stating that LeelaBot had resigned, I asked him if he had killed the robot, and he replied he did not. I assumed that it had resigned and we started the review. What would you have done in a case like this? :) - I recieved a correction from Gian-Carlo for the review ... I had guessed that Leela used an opening book, but it does not. /Basti ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] Re: What's happening at the European Go Congress?
On Mon, Aug 11, 2008 at 4:14 PM, Basti Weidemyr [EMAIL PROTECTED] wrote: What would you have done in a case like this? :) Inspect the log file. Erik ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] Re: What's happening at the European Go Congress?
On Mon, 2008-08-11 at 16:54 +0200, Gian-Carlo Pascutto wrote: As long as we're not there, these matches are a great promotion for the game of go. Just watch how much publicity the MoGo match got. And there's still lots of possibilities for the humans to take revenge, and for the computers to take counter-revenge... And I think we still have plenty of time before it gets to the point where computers are so clearly superior that it's pointless to play them. And well past that point in time computers can then be the one giving stones in handicap play. - Don ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] Re: What's happening at the European Go Congress?
Basti Weidemyr wrote: What would you have done in a case like this? :) You could not declare that game a win for the computer and survive. Rémi ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
[computer-go] Re: mogo beats pro!
Hi all, I'd like to say first Congratulations! to MoGo team. I have a question. Why do you all call the game as human vs. computer? It's obviously a match between Kim 8p and MoGo, a program developped by MoGo team, running on a supercomputer. As both MoGo and the supercomputer were developped by human, the game is clearly (a special type of) human vs. human. I'm afraid it may raise unnecessary emotional thoughts of against computers among people. It might be better to call such a game something of a style a professinal Goplayer vs. a program with its developper(s) to emphasize the program was created by human. -Hideki terry mcintyre: [EMAIL PROTECTED]: This is from the AGA newsletter: COMPUTER BEATS PRO AT U.S. GO CONGRESS: In a historic achievement, the MoGo computer program defeated Myungwan Kim 8P (l) Thursday afternoon by 1.5 points in a 9-stone game billed as Humanitys Last Stand? It played really well, said Kim, who estimated MoGos current strength at two or maybe three dan, though he noted that the program which used 800 processors, at 4.7 Ghz, 15 Teraflops on a borrowed European supercomputer made some 5-dan moves, like those in the lower right-hand corner, where Moyogo took advantage of a mistake by Kim to get an early lead. I cant tell you how amazing this is, David Doshay -- the SlugGo programmer who suggested the match -- told the E-Journal after the game. Im shocked at the result. I really didnt expect the computer to win in a one-hour game. Kim easily won two blitz games with 9 stones and 11 stones and minutes and lost one with 12 stones and 15 minutes by 3.5 points. The games were played live at the U.S. Go Congress, with over 500 watching online on KGS. I think theres no chance on nine stones, Kim told the EJ after the game. It would even be difficult with eight stones. MoGo played really well; after getting a lead, every time I played aggressively, it just played safely, even when it meant sacrificing some stones. It didnt try to maximize the win and just played the most sure way to win. Its like a machine. The game generated a lot of interest and discussion about the games tactics and philosophical implications. Congratulations on making history today, game organizer Peter Drake told both Kim and Olivier Teytaud, one of MoGos programmers, who participated ina brief online chat after the game. At a rare loss for words in a brief interview with the EJ after the game, Doshay wondered How much time do we have left? Weve improved nine stones in just a year and I suspect the next nine will fall quickly now. - reported by Chris Garlock, photo by Brian Allen Terry McIntyre [EMAIL PROTECTED] Wherever is found what is called a paternal government, there is found state education. It has been discovered that the best way to insure implicit obedience is to commence tyranny in the nursery. Benjamin Disraeli, Speech in the House of Commons [June 15, 1874] ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/ -- [EMAIL PROTECTED] (Kato) ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] Re: mogo beats pro!
On 10-aug-08, at 17:24, Don Dailey wrote: Of course there is also the possibility of some exciting new hardware breakthrough around the corner that doesn't just extend Moore's law, but blows it out of the water. From: Mark Boon [EMAIL PROTECTED] Of course there's that possibility. But I'm actually wondering if we wouldn't rather be seeing the opposite. Moore's law seems to have stalled for a few years, only to gain traction again with multiple core designs. But unlike previous advances in computing power, multiple processing is not as easily available to all software alike. True, not all software can utilize 100s or 1000s of processors effectively -- but there is at least one Go program which happily scales to hundreds of processors. The folks who do HPC will eat up manycore chips by the bushel. It is reasonable to ask If the average consumer sees no benefit, will manycore chips be produced? One can argue that millions of chips must be sold to recover the costs of developing higher-resolution processes, new architectures, and so forth. I suspect that applications will be developed which harness those chips. We can't say exactly what they'll be. Games, no doubt, will soak up lots of CPU cycles. Business desktops will be asked to do far more complex data mining. Web services will demand lots of CPU cycles. HPC isn't just for universities, oil, and financial firms; Google and Amazon and other search firms will be asking for more computer power in more compact spaces using less energy. Will my Great Aunt Tilda have 256 or 1024 cores on her desktop in ten years? Probably not; but enthusiasts like those on this list will. Small research departments will have even bigger clusters; they'll consider an equivalent to the 800 Power6 cores used last week to be a rather modest investment. This does not require much extrapolation; several quad chips are widely available; GPUs use hundreds of processor cores; Sun's Niagara chip has 8 cores which do 8 threads apiece, and claims that a 16 core times 16 thread version (Rock) will be available in late 2009. Cavium already ships 16-core MIPS-based processors. Cisco has a 188-core Metro network processor, and Tilera produces a 64-core chip, the Tile64. Sicortex ships a 6-core MIPS-based chip -- in tightly-coupled clusters with up to 5832 cores in a single box. There was a time when million-dollar supercomputers were used for research in Chess programming. Here's hope that the same will happen for Go programming. ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] Re: What's happening at the European Go Congress?
On Mon, Aug 11, 2008 at 4:54 PM, Gian-Carlo Pascutto [EMAIL PROTECTED] wrote: She was also a bit unlucky in the sense that Leela did not understand it was dead lost. I use quotes because had it understood better it was losing, it would have put up more of a fight :-) If Basti is correct that Leela resigned that would suggest that 'she' actually did understand. For the final position in the game record any strong human player will tell you that the game is clearly over. No points are left to be gained and the result is obvious. If Leela had persisted in attempting to push the opponent through the clock, then I guess any EGC referee would have considered that 'unsportsmanlike' behavior (but it would of course be nice to know for sure). As time was running out and the robot played obstinate moves, I told the operator to kill it. However, it looked to me like he never touched the keyboard, so when a dialog appeared, stating that LeelaBot had resigned, I asked him if he had killed the robot, and he replied he did not. The KGS server should have recorded the resignation instantly, but there is no sign of it in the game record. Some time ago I observed that kgsgtp does not tell my program that the opponent has resigned (which is a bit annoying because it then keeps pondering when the game is already over). It's a long shot but maybe this behavior somehow also goes the other way around? Erik ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] Re: What's happening at the European Go Congress?
On Aug 11, 2008, at 12:02 PM, Erik van der Werf [EMAIL PROTECTED] wrote: On Mon, Aug 11, 2008 at 4:54 PM, Gian-Carlo Pascutto [EMAIL PROTECTED] wrote: Some time ago I observed that kgsgtp does not tell my program that the opponent has resigned (which is a bit annoying because it then keeps pondering when the game is already over). KgsGtp should send kgs-game_over in such cases. ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
[computer-go] Re: mogo beats pro!
Yes, but exhausitve search does not improve your player by 63% (eg.) for a doubling in CPU time. This part was done in an empirical scalability study. Please check the archives of the list. In the (inifinite) limit minimax+evaluation-function would find the perfect move too, but UCT/MC already find good moves before the limit. Yes... I agree... UCT/MC seems to find the good moves before the limit and from statistics.. seems that the good moves come out long before we have exhaustively searched the tree. I was questioning the rate at which we approach perfect play. This term seems silly to me... as it would imply actually solving the game. The whole idea of playing vs. god and drawing or winning only means one thing to me... and that would be actually knowing every possible path to determine the best path. The results of the MC statistics simply say that this move appears to be better given the sample size. To me.. I don't think anyone could say that you could beat god without actually knowing the whole tree. That would be conjecture at least at this point. And having God in the equation already moves us to mysticism (or some sort of statement that the game has a solution). As far as the 63% gain... I feel that there are certain additional descriptors needed there. We did not see a statistical increase in ability vs. human players. We saw a 63% gain when putting programs against programs. This is hardly the same problem. It is valuable information and I am not discounting it at all. I just feel that this evidence DNE what it seemed to be used for in previous discussions. ...Why are you trying to share it with us in the first place. For myself, i believe that what you are trying to do, is to begin to analyses all the data the community has gathered so far... Well.. things certainly got heated and as I looked at the list.. I started feeling guilty that I kind of took over. The list seems primarily used for coordination between you guys and perhaps at times theoretical discussion. I apologize for the rants that have perhaps shown up suddenly. The background reason I came in here was that I love go and have loved it ever since I learned to play about 5 years ago. I am also a developer and long ago had read many articles on computer go. At the time.. and perhaps up to now.. there have been many go players, computer scientists and lay people who have worried that perhaps the greatest strength of the computer, fast computation, would not be such a great help with playing go. There were taunts from this side saying that computers couldn't really beat children who were decent. After reading and hearing these sorts of discussions... I started to fall into that group. My personal feeling was that AI now is akin to a human taking a lot of time trying to create a particular algorithm. Then this algorithm would work in a particular scenario. This seems difficult for go as each of these heuristics are focused and meanwhile, you have a human who is constantly changing his heuristics during their years of learning. I feel that to have what movies consider AI or what the general public expects from AI, we will need a new paradigm where computers learn to solve problems by themselves through experimentation and learning. This does not necessarily apply to go, but is possible. The reason I brought up complexity theory is not to confine computer go to a particular complexity class... but to discuss the fact that our current model of computing machines do appear to solve many important problems.. but that there some classes of problems that we are not so certain can be solved with the computer model we all have at our desks or in our datacenters. When I read the article by the DeepBlue guy called Cracking Go, I was very skeptical. I felt that he was assuming too much. When I read that Mogo was going to get a nice big cluster.. I was very excited and couldn't wait to watch the game. When Mogo started to turn around... I had completely swtiched from skeptic to cheering it on. I think the Mogo team and many people on here have done a great job. So then I jumped into conversation here and perhaps had not fully researched previous topics and breakthroughs... but I felt that I was cut down pretty quickly with the phrase proven to be scalable to perfect play. The phrase itself was used to completely nullify my argument. That is perhaps where it started to get out of hand. Don's Duck does not really seem to be clearly a duck. In his analogy... his duck is almost an axiom and I am some crazy freak who thinks the world is flat. I felt it was a bit condescending and did feel I had to try to clear the logic up. At this point.. I have read the Bandit paper and am pretty sure where he got this phrase. In the paper it is phrased differently. I am probably at fault here because I have just jumped in here and have not been a part of much previous discourse. Perhaps that phrase has a different meaning here and people would
Re: [computer-go] Re: What's happening at the European Go Congress?
On Mon, Aug 11, 2008 at 6:17 PM, Jason House [EMAIL PROTECTED] wrote: On Aug 11, 2008, at 12:02 PM, Erik van der Werf [EMAIL PROTECTED] wrote: On Mon, Aug 11, 2008 at 4:54 PM, Gian-Carlo Pascutto [EMAIL PROTECTED] wrote: Some time ago I observed that kgsgtp does not tell my program that the opponent has resigned (which is a bit annoying because it then keeps pondering when the game is already over). KgsGtp should send kgs-game_over in such cases. Hmm, guess I missed that command. I had solved the issue by setting an upper bound on the ponder time, which also works well for playing manually. Thanks, Erik ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] Re: What's happening at the European Go Congress?
Erik van der Werf wrote: On Mon, Aug 11, 2008 at 4:54 PM, Gian-Carlo Pascutto [EMAIL PROTECTED] wrote: She was also a bit unlucky in the sense that Leela did not understand it was dead lost. I use quotes because had it understood better it was losing, it would have put up more of a fight :-) If Basti is correct that Leela resigned that would suggest that 'she' actually did understand. Why you are arguing with me about this? I am the author, I have the binary. It does not understand the seki at any level. Now, if the Leela binary would somehow have gained a better understanding of seki on the trip to Leksand, that would just be _scary_ :) Some time ago I observed that kgsgtp does not tell my program that the opponent has resigned (which is a bit annoying because it then keeps pondering when the game is already over). It's a long shot but maybe this behavior somehow also goes the other way around? I hope not. This would mean the opponents in those games would have to sit out the remaining time. It cannot have happened anyway - in your case either Leela or kgsGtp would have to have popped up the mysterious window, and neither has that ability as far as I know. -- GCP ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
[computer-go] 3x3 patterns
How are folks constructing their 3x3 pattern databases? How are they being used? If they are being used for playout biases, then I don't think examining games is the right way to gather data. 90% of the moves considered in a game of Go are unplayed; the tactical analysis that is required to determine whether the moves actually played are sound. This seems to be what the playouts represent. 3x3 is all about contact, which mostly is about fighting, tesuji, joseki, semeais, life-and-death, connectivity, yose, and finalizing boundaries. So it seems to me that 3x3 patterns should bias sente and urgent moves (hane, extend, shoulder hit, attach, block, peep, push, connect, turn, ko, ladders) and prevent local mistakes (filling eyes, bad shape). My own studies show that the empty 3x3 pattern is by far the most used (and I suspect crucial), followed by hane, attach, block, shoulder, and extend. The probability of each connection and blocking pattern with many stones is low, because there are more possible stone combinations that are essentially the same situation; the likelyhood of any one situation showing up in a game is small. Do folks have sparser pattern databases for empty space move selection in playouts (one point jump, keima, two point jump, corner enclosures, loose connections, wall extensions, etc)? Have you seen other surprising biases in your generated 3x3 pattern databases? Also, has anyone used the small diamond pattern instead of 3x3 patterns? This is gives you one-point jumps, kos, and more sensitivity to edge effects. Ian Terms (for a move in the center by O, '?' means maybe add one O): . ... .. .. small diamond pattern ... . OX. . . hane ??? .X. ? . attach ?.. XX? O ? block ??? X.. . . shoulder ??? XO. XOX . . . . extend ... ... OX. O . turn ?.. ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] Re: What's happening at the European Go Congress?
Erik van der Werf wrote: For the final position in the game record any strong human player will tell you that the game is clearly over. No points are left to be gained and the result is obvious. Actually there's one point left to gain in the seki, since the game is played with Chinese rules. ;-) /Gunnar ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] Re: What's happening at the European Go Congress?
On Mon, 2008-08-11 at 17:26 +0200, Rémi Coulom wrote: Basti Weidemyr wrote: What would you have done in a case like this? :) You could not declare that game a win for the computer and survive. Yes, and I really hate this. You have a situation where the actual winner has to resign the game in order to not be ridiculed as being petty. And is the human player supposed to feel good about his victory? - Don Rémi ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/ ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] Re: What's happening at the European Go Congress?
On Aug 11, 2008, at 2:06 PM, Don Dailey [EMAIL PROTECTED] wrote: On Mon, 2008-08-11 at 17:26 +0200, Rémi Coulom wrote: Basti Weidemyr wrote: What would you have done in a case like this? :) You could not declare that game a win for the computer and survive. Yes, and I really hate this. You have a situation where the actual winner has to resign the game in order to not be ridiculed as being petty. I hate absolute time limits for this reason. Even a small byo yomi prevents wins for such a stupid reason. Certainly, humans can't have 10 millisecond response times like a computer. And is the human player supposed to feel good about his victory? Nobody should be happy with a game decided by time in late yose. Of course, rules are rules. I just don't play games with absolute time - Don Rémi ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/ ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/ ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] Re: What's happening at the European Go Congress?
On Mon, 2008-08-11 at 18:02 +0200, Erik van der Werf wrote: On Mon, Aug 11, 2008 at 4:54 PM, Gian-Carlo Pascutto [EMAIL PROTECTED] wrote: She was also a bit unlucky in the sense that Leela did not understand it was dead lost. I use quotes because had it understood better it was losing, it would have put up more of a fight :-) If Basti is correct that Leela resigned that would suggest that 'she' actually did understand. For the final position in the game record any strong human player will tell you that the game is clearly over. No points are left to be gained and the result is obvious. If Leela had persisted in attempting to push the opponent through the clock, then I guess any EGC referee would have considered that 'unsportsmanlike' behavior (but it would of course be nice to know for sure). But is it really? Now instead of clearly defined rules, you enter the domain of judgment calls and these should be minimized. How clear does it have to be there is a win? Who decides where the gray area is? In chess it's been an important part of the game. You can get great positions if you spend a lot of time thinking and it's clear that is true in GO too.The longer I think, the better on average my position will be. But if I am less honest than my opponent about managing my time, why should I be given a free pass? I think the best thing is to use a Fischer clock with 1 or 2 seconds added per move and be religiously strict about honoring the rules. The rules I'm talking about, by the way, are the rules that you agreed to play by, before starting the game. The Fischer clock will protect you from unexpectedly long end games. Maybe it's just me, but I don't want my games judged. I don't want anybody saying that you lose even though my opponent used too much time. If you want to grant wins to the time loser, then instead of requiring someone to judge the result spell out the kinds of positions where the game should be stopped. If you cannot spell it out, then you have to judge it. - Don As time was running out and the robot played obstinate moves, I told the operator to kill it. However, it looked to me like he never touched the keyboard, so when a dialog appeared, stating that LeelaBot had resigned, I asked him if he had killed the robot, and he replied he did not. The KGS server should have recorded the resignation instantly, but there is no sign of it in the game record. Some time ago I observed that kgsgtp does not tell my program that the opponent has resigned (which is a bit annoying because it then keeps pondering when the game is already over). It's a long shot but maybe this behavior somehow also goes the other way around? Erik ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/ ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] Re: mogo beats pro!
On Mon, 2008-08-11 at 12:23 -0400, Robert Waite wrote: Yes, but exhausitve search does not improve your player by 63% (eg.) for a doubling in CPU time. This part was done in an empirical scalability study. Please check the archives of the list. In the (inifinite) limit minimax+evaluation-function would find the perfect move too, but UCT/MC already find good moves before the limit. Yes... I agree... UCT/MC seems to find the good moves before the limit and from statistics.. seems that the good moves come out long before we have exhaustively searched the tree. I was questioning the rate at which we approach perfect play. This term seems silly to me... as it would imply actually solving the game. The whole idea of playing vs. god and drawing or winning only means one thing to me... and that would be actually knowing every possible path to determine the best path. The results of the MC statistics simply say that this move appears to be better given the sample size. To me.. I don't think anyone could say that you could beat god without actually knowing the whole tree. That would be conjecture at least at this point. And having God in the equation already moves us to mysticism (or some sort of statement that the game has a solution). You don't need to know the whole tree, you only need to know some of the tree and it's a very small fraction of the whole. That's what alpha/beta pruning is all about. - Don As far as the 63% gain... I feel that there are certain additional descriptors needed there. We did not see a statistical increase in ability vs. human players. We saw a 63% gain when putting programs against programs. This is hardly the same problem. It is valuable information and I am not discounting it at all. I just feel that this evidence DNE what it seemed to be used for in previous discussions. ...Why are you trying to share it with us in the first place. For myself, i believe that what you are trying to do, is to begin to analyses all the data the community has gathered so far... Well.. things certainly got heated and as I looked at the list.. I started feeling guilty that I kind of took over. The list seems primarily used for coordination between you guys and perhaps at times theoretical discussion. I apologize for the rants that have perhaps shown up suddenly. The background reason I came in here was that I love go and have loved it ever since I learned to play about 5 years ago. I am also a developer and long ago had read many articles on computer go. At the time.. and perhaps up to now.. there have been many go players, computer scientists and lay people who have worried that perhaps the greatest strength of the computer, fast computation, would not be such a great help with playing go. There were taunts from this side saying that computers couldn't really beat children who were decent. After reading and hearing these sorts of discussions... I started to fall into that group. My personal feeling was that AI now is akin to a human taking a lot of time trying to create a particular algorithm. Then this algorithm would work in a particular scenario. This seems difficult for go as each of these heuristics are focused and meanwhile, you have a human who is constantly changing his heuristics during their years of learning. I feel that to have what movies consider AI or what the general public expects from AI, we will need a new paradigm where computers learn to solve problems by themselves through experimentation and learning. This does not necessarily apply to go, but is possible. The reason I brought up complexity theory is not to confine computer go to a particular complexity class... but to discuss the fact that our current model of computing machines do appear to solve many important problems.. but that there some classes of problems that we are not so certain can be solved with the computer model we all have at our desks or in our datacenters. When I read the article by the DeepBlue guy called Cracking Go, I was very skeptical. I felt that he was assuming too much. When I read that Mogo was going to get a nice big cluster.. I was very excited and couldn't wait to watch the game. When Mogo started to turn around... I had completely swtiched from skeptic to cheering it on. I think the Mogo team and many people on here have done a great job. So then I jumped into conversation here and perhaps had not fully researched previous topics and breakthroughs... but I felt that I was cut down pretty quickly with the phrase proven to be scalable to perfect play. The phrase itself was used to completely nullify my argument. That is perhaps where it started to get out of hand. Don's Duck does not really seem to be clearly a duck. In his analogy... his duck is almost an axiom and I am some crazy freak who thinks the world is flat. I felt it was a bit condescending and did feel I had to try to clear the
Re: [computer-go] Re: What's happening at the European Go Congress?
I agree with you Jason. I advocate the more modern Fisher clock, where some fixed amount of time is added to each move and remains yours to keep. Even 1 or 2 seconds per move is enough since you can build up time. - Don On Mon, 2008-08-11 at 14:18 -0400, Jason House wrote: On Aug 11, 2008, at 2:06 PM, Don Dailey [EMAIL PROTECTED] wrote: On Mon, 2008-08-11 at 17:26 +0200, Rémi Coulom wrote: Basti Weidemyr wrote: What would you have done in a case like this? :) You could not declare that game a win for the computer and survive. Yes, and I really hate this. You have a situation where the actual winner has to resign the game in order to not be ridiculed as being petty. I hate absolute time limits for this reason. Even a small byo yomi prevents wins for such a stupid reason. Certainly, humans can't have 10 millisecond response times like a computer. And is the human player supposed to feel good about his victory? Nobody should be happy with a game decided by time in late yose. Of course, rules are rules. I just don't play games with absolute time - Don Rémi ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/ ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/ ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] Re: What's happening at the European Go Congress?
In message [EMAIL PROTECTED], Don Dailey [EMAIL PROTECTED] writes On Mon, 2008-08-11 at 18:02 +0200, Erik van der Werf wrote: On Mon, Aug 11, 2008 at 4:54 PM, Gian-Carlo Pascutto [EMAIL PROTECTED] wrote: She was also a bit unlucky in the sense that Leela did not understand it was dead lost. I use quotes because had it understood better it was losing, it would have put up more of a fight :-) If Basti is correct that Leela resigned that would suggest that 'she' actually did understand. For the final position in the game record any strong human player will tell you that the game is clearly over. No points are left to be gained and the result is obvious. If Leela had persisted in attempting to push the opponent through the clock, then I guess any EGC referee would have considered that 'unsportsmanlike' behavior (but it would of course be nice to know for sure). But is it really? Now instead of clearly defined rules, you enter the domain of judgment calls and these should be minimized. How clear does it have to be there is a win? Who decides where the gray area is? In chess it's been an important part of the game. You can get great positions if you spend a lot of time thinking and it's clear that is true in GO too.The longer I think, the better on average my position will be. But if I am less honest than my opponent about managing my time, why should I be given a free pass? I think the best thing is to use a Fischer clock with 1 or 2 seconds added per move and be religiously strict about honoring the rules. The rules I'm talking about, by the way, are the rules that you agreed to play by, before starting the game. The Fischer clock will protect you from unexpectedly long end games. Maybe it's just me, but I don't want my games judged. No sane tournament director wants to have to use his judgement (though it may be necessary). I think Fischer time would be an excellent solution. Nick I don't want anybody saying that you lose even though my opponent used too much time. If you want to grant wins to the time loser, then instead of requiring someone to judge the result spell out the kinds of positions where the game should be stopped. If you cannot spell it out, then you have to judge it. - Don As time was running out and the robot played obstinate moves, I told the operator to kill it. However, it looked to me like he never touched the keyboard, so when a dialog appeared, stating that LeelaBot had resigned, I asked him if he had killed the robot, and he replied he did not. The KGS server should have recorded the resignation instantly, but there is no sign of it in the game record. Some time ago I observed that kgsgtp does not tell my program that the opponent has resigned (which is a bit annoying because it then keeps pondering when the game is already over). It's a long shot but maybe this behavior somehow also goes the other way around? Erik ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/ ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/ -- Nick Wedd[EMAIL PROTECTED] ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] Re: What's happening at the European Go Congress?
On 11-aug-08, at 15:23, Don Dailey wrote: But is it really? Now instead of clearly defined rules, you enter the domain of judgment calls and these should be minimized. I don't agree with such an unforgiving attitude at all. It works for tournaments but not for demonstration games. You don't want to give fuel to those who argue yeah, but the computer can respond in a millisecond where the human has a physical response time of at least half a second. In demonstration games what is important is the spirit. And it doesn't do the computer-Go community any good for a program to persist in an absolutely lost position, play one for a hundred more moves that the human is physically unable to play in time. It will have to give one way or another. I also don't like the fixed time-limit very much because Go has such an unpredictable game- length. So Fisher time could be a solution. On the other hand, once the level of the programs becomes well established, programmers could also make it resign a lot sooner. In a 1,000 ELO game a 99% win-rate might occasionally still turn around. But they'll probably find out that as the level gets higher, say 3,000 ELO, you end up never turning around a 90% win-rate. Or maybe even 80%. If programmers want humans to play their software then they have to be also a little accomodating in that respect, even if that means giving up a game where you might still win once in a thousand times. Mark ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
[computer-go] Re: mogo beats pro!
You don't need to know the whole tree, you only need to know some of the tree and it's a very small fraction of the whole. That's what alpha/beta pruning is all about. Certainly we are seeing gains by looking at smaller portions of the tree. Perfect play and the question of God however seem akin to generating the entire tree. Since we are using statistics to decide which parts of the tree are interesting or good... we are still missing other potential playouts. Judging by the number of samples vs. the number of possible games, we are missing many possible moves, some good, some bad. I guess it all depends on your definition of God... but my mental image would be that he knows all paths of the game.. and thus would not be tricked when we are only looking at a subset of the tree. Same goes for perfect move. We can never judge accurately if we found the perfect move unless we know what all possible moves are. We can only guess with a certain amount of confidence. However... humans are not anything like God and are missing huge portions of the tree as well. Generating a better move than a human is in no way related to the need to generate the entire tree. I do agree that it is very much possible that this technique alone could scale to the point of crushing a person. I just don't know yet if we can accurately guess that. People are rather flexible in the way they play and learn and it is still very trickly to evaluate a goban. I am not completely familiar with Mogo's inner workings.. but it does have a base heuristic that helps it judge where to start looking. If it doesn't find a certain group of moves.. it goes to random spots. It seems that adding this heuristic stage helped CrazyStone and Mogo greatly. I am totally on your side with beating human players. It is obvious that it will happen one day. I am just very curious of the number crunch vs. heuristics vs. a whole new paradigm debate. MC is kind of a new paradigm for number crunching (at least in go).. and it is great. I am just curious if new barriers will be hit. And I am happy either way... if MC does not turn out to be the panacea of computer go... then new methods will be combined with it or whole new methods will come out. A number of barriers have already been hit.. computer go seemed stuck around 12-8k for a while.. and it was a bit worse before right? I am not trying to crush anyone's dreams.. I am just curious if this breakthrough has really removed all of the barriers to number crunching or if it has just helped us push up to the next level, one where we will once again have to work on heuristics or go back to the drawing board. I wish we could rely on hardware scaling.. but I am not so sure that it is the answer yet. That is a hunch.. but so are many claims I have seen. I can't turn my hunch into a theory until I have more data. I haven't seen a source of data that convinces me we are home free yet. When we start putting these machines against seasoned pros... and watch what happens when you scale... I think we will have a much better answer. And to Don, I think we had plenty of misunderstandings and miscommunications... I know you aren't saying we are home free and you have put a lot more work in it than I have. When you said proven to be scalable to perfect play you probably meant it in a different sense than I read it and many people on here might have understood what you were trying to point out. I'm not trying to spray your parade with yellow rain... I just feel that there are still many unknowns. On Mon, Aug 11, 2008 at 12:23 PM, Robert Waite [EMAIL PROTECTED]wrote: Yes, but exhausitve search does not improve your player by 63% (eg.) for a doubling in CPU time. This part was done in an empirical scalability study. Please check the archives of the list. In the (inifinite) limit minimax+evaluation-function would find the perfect move too, but UCT/MC already find good moves before the limit. Yes... I agree... UCT/MC seems to find the good moves before the limit and from statistics.. seems that the good moves come out long before we have exhaustively searched the tree. I was questioning the rate at which we approach perfect play. This term seems silly to me... as it would imply actually solving the game. The whole idea of playing vs. god and drawing or winning only means one thing to me... and that would be actually knowing every possible path to determine the best path. The results of the MC statistics simply say that this move appears to be better given the sample size. To me.. I don't think anyone could say that you could beat god without actually knowing the whole tree. That would be conjecture at least at this point. And having God in the equation already moves us to mysticism (or some sort of statement that the game has a solution). As far as the 63% gain... I feel that there are certain additional descriptors needed there. We did not see a statistical increase in ability vs. human
[computer-go] Cultural differences: players vs programmers
Don Dailey wrote: On Mon, 2008-08-11 at 17:26 +0200, Rémi Coulom wrote: Basti Weidemyr wrote: What would you have done in a case like this? :) You could not declare that game a win for the computer and survive. Yes, and I really hate this. You have a situation where the actual winner has to resign the game in order to not be ridiculed as being petty. And is the human player supposed to feel good about his victory? The statements earlier point to the indication that the human player might not even have been really aware that there was a time limit. That shouldn't have happened. But I think this entire discussion is part of a larger problem, a cultural problem: To the pro's, the games against computers are probably halfway a joke. The computers obviously so weak that having to think seriously to win a game would be an insult. You can beat them with silly handicaps even if they run on insanely big computers. Can't consider such a game serious. To the program authors, the computers are their pride and what they spend all their available time on. A good performance fills them with joy and pride and a bad performance makes them look like this: http://www.morbo.org/pics/Mainz2008/DSC_7533.jpg (picture taken by my girlfriend after I scored very badly during the first half of a computerchess tournament) To a programmer, such a game against a stronger player is _always_ DEAD SERIOUS and he or she will do anything reasonable to win. If this means flagging a professional player who didn't manage his or her time well, then be sure that is what we'll do to claim victory. And good luck explaining afterwards why the program didn't won a game that was won by the rules. To hell with what the crowd thinks, they were on the side of the human to start with anyway :) I put programmers in quotes because this isn't actually about programmers only. Imagine you are a weak player that gets the right to play in a simul (or in go terms, a handicap game) against Kasparov (or let's say Cho Chikun in go terms?). For Kasparov/Cho the game is a joke, an aside they do as a part of their living as professional players. To the weak player, such a game is a very rare opportunity. Imagine winning the simul/handicap game! For sure, for the professional this is the result of a slight lapse in concentration, nothing to worry about. But good luck explaining the weaker player that the game was not serious - most likely, it's the only game he'll talk about for the rest of his life! A game between players of very different strengths is never not serious to the weaker player. PARTICULARLY not if he won (by any stretch of the regulations). However, this programmer at least is very happy that Ms. Xiao Ai Lin gave his program enough attention to pound it to pieces. Now, if I read in the tournament report that the second human-computer game didn't happen because The pro showed a lack of enthusiasm, and did not turn up in the room at the time it was meant to happen., it might as well have read: The pro drove a stick through the heart of the programmer while telling him he is an insignificant being not worthy of any attention and certainly not half an hour of his time. Not serious, eh? Ever seen a crazy programmer with a pitchfork? Arrr! -- GCP PS. I might have exaggerated ever so slightly in this post to get my point across, and I apologize in advance to all the (go) programmers, go players and go tournament directors I offended and who think I unjustly spoke in their name. ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] Re: mogo beats pro!
On Mon, 2008-08-11 at 15:21 -0400, Robert Waite wrote: You don't need to know the whole tree, you only need to know some of the tree and it's a very small fraction of the whole. That's what alpha/beta pruning is all about. Certainly we are seeing gains by looking at smaller portions of the tree. Perfect play and the question of God however seem akin to generating the entire tree. You clearly don't understand the principles of alpha/beta pruning. It is an admissible technique which means it guarantee's the same result as searching the entire tree, but only requires a very tiny subset of the entire tree. Let's say you have 381 possible first moves and you find that the first one you look at wins the game by 6.5 points.So you know that no matter what happens with the next 380 moves you can always fall back on the one that wins by 6.5. You of course hope that you will find a move that does BETTER so you are forced to search all 381 moves. So you start looking at the 2nd move. Just for fun let's say you look at A1 and your opponent, after looking at only 1 move to the end of the game sees that he can beat you. You already KNOW that you can do better and that you should not play A1. It does not matter if the opponent finds an even better refutation or not, it's enough to know that even if he doesn't try, he is going to beat you with his first choice - and you have already discovered that you can win. It is explained pretty well here: http://en.wikipedia.org/wiki/Alpha-beta_pruning Here is a quote: Alpha-beta pruning is a sound optimization in that it does not change the result of the algorithm it optimizes. - Don Since we are using statistics to decide which parts of the tree are interesting or good... we are still missing other potential playouts. Judging by the number of samples vs. the number of possible games, we are missing many possible moves, some good, some bad. I guess it all depends on your definition of God... but my mental image would be that he knows all paths of the game.. and thus would not be tricked when we are only looking at a subset of the tree. Same goes for perfect move. We can never judge accurately if we found the perfect move unless we know what all possible moves are. We can only guess with a certain amount of confidence. However... humans are not anything like God and are missing huge portions of the tree as well. Generating a better move than a human is in no way related to the need to generate the entire tree. I do agree that it is very much possible that this technique alone could scale to the point of crushing a person. I just don't know yet if we can accurately guess that. People are rather flexible in the way they play and learn and it is still very trickly to evaluate a goban. I am not completely familiar with Mogo's inner workings.. but it does have a base heuristic that helps it judge where to start looking. If it doesn't find a certain group of moves.. it goes to random spots. It seems that adding this heuristic stage helped CrazyStone and Mogo greatly. I am totally on your side with beating human players. It is obvious that it will happen one day. I am just very curious of the number crunch vs. heuristics vs. a whole new paradigm debate. MC is kind of a new paradigm for number crunching (at least in go).. and it is great. I am just curious if new barriers will be hit. And I am happy either way... if MC does not turn out to be the panacea of computer go... then new methods will be combined with it or whole new methods will come out. A number of barriers have already been hit.. computer go seemed stuck around 12-8k for a while.. and it was a bit worse before right? I am not trying to crush anyone's dreams.. I am just curious if this breakthrough has really removed all of the barriers to number crunching or if it has just helped us push up to the next level, one where we will once again have to work on heuristics or go back to the drawing board. I wish we could rely on hardware scaling.. but I am not so sure that it is the answer yet. That is a hunch.. but so are many claims I have seen. I can't turn my hunch into a theory until I have more data. I haven't seen a source of data that convinces me we are home free yet. When we start putting these machines against seasoned pros... and watch what happens when you scale... I think we will have a much better answer. And to Don, I think we had plenty of misunderstandings and miscommunications... I know you aren't saying we are home free and you have put a lot more work in it than I have. When you said proven to be scalable to perfect play you probably meant it in a different sense than I read it and many people on here might have understood what you were trying to point out. I'm not trying to spray your parade with yellow rain... I just feel that there are still many unknowns. On Mon, Aug 11, 2008 at
Re: [computer-go] Cultural differences: players vs programmers
In message [EMAIL PROTECTED], Gian-Carlo Pascutto [EMAIL PROTECTED] writes Don Dailey wrote: On Mon, 2008-08-11 at 17:26 +0200, Rémi Coulom wrote: Basti Weidemyr wrote: What would you have done in a case like this? :) You could not declare that game a win for the computer and survive. Yes, and I really hate this. You have a situation where the actual winner has to resign the game in order to not be ridiculed as being petty. And is the human player supposed to feel good about his victory? The statements earlier point to the indication that the human player might not even have been really aware that there was a time limit. That shouldn't have happened. But I think this entire discussion is part of a larger problem, a cultural problem: To the pro's, the games against computers are probably halfway a joke. The computers obviously so weak that having to think seriously to win a game would be an insult. You can beat them with silly handicaps even if they run on insanely big computers. Can't consider such a game serious. To the program authors, the computers are their pride and what they spend all their available time on. A good performance fills them with joy and pride and a bad performance makes them look like this: http://www.morbo.org/pics/Mainz2008/DSC_7533.jpg (picture taken by my girlfriend after I scored very badly during the first half of a computerchess tournament) To a programmer, such a game against a stronger player is _always_ DEAD SERIOUS and he or she will do anything reasonable to win. If this means flagging a professional player who didn't manage his or her time well, then be sure that is what we'll do to claim victory. And good luck explaining afterwards why the program didn't won a game that was won by the rules. To hell with what the crowd thinks, they were on the side of the human to start with anyway :) I put programmers in quotes because this isn't actually about programmers only. Imagine you are a weak player that gets the right to play in a simul (or in go terms, a handicap game) against Kasparov (or let's say Cho Chikun in go terms?). For Kasparov/Cho the game is a joke, an aside they do as a part of their living as professional players. To the weak player, such a game is a very rare opportunity. Imagine winning the simul/handicap game! For sure, for the professional this is the result of a slight lapse in concentration, nothing to worry about. But good luck explaining the weaker player that the game was not serious - most likely, it's the only game he'll talk about for the rest of his life! A game between players of very different strengths is never not serious to the weaker player. PARTICULARLY not if he won (by any stretch of the regulations). However, this programmer at least is very happy that Ms. Xiao Ai Lin gave his program enough attention to pound it to pieces. Now, if I read in the tournament report that the second human-computer game didn't happen because The pro showed a lack of enthusiasm, and did not turn up in the room at the time it was meant to happen., it might as well have read: The pro drove a stick through the heart of the programmer while telling him he is an insignificant being not worthy of any attention and certainly not half an hour of his time. Not serious, eh? Ever seen a crazy programmer with a pitchfork? Arrr! -- GCP PS. I might have exaggerated ever so slightly in this post to get my point across, and I apologize in advance to all the (go) programmers, go players and go tournament directors I offended and who think I unjustly spoke in their name. I think you expressed things remarkably well. (Is English really not your native language?) I don't have an image of you for http://www.computer-go.info/db/operson.php?a=Pascutto%2C+Gian-Carlo Will you mind if I use that one? :-) Nick ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/ -- Nick Wedd[EMAIL PROTECTED] ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] Cultural differences: players vs programmers
I guess we're all different. Last week, I actually did win a 9-stone handicap game in a simul match against a pro, but I'm not about to claim that this gives me bragging rights or anything, lol. My guess is that a) I did have a formidable handicap; b) he was distracted by playing half a dozen other people, and c) this was probably a teaching game; spending 40 minutes fighting to encircle and kill his group taught me a lot about the many snares which a pro can set, and how to circumvent them. This might not help me defeat Lee Chang Ho any time soon, but it will help my next club match. :) If my program won on time in an obviously lost position, I'd be turning every rock to find a way to improve the actual play; that matters much more to me than the win-loss record. Terry McIntyre [EMAIL PROTECTED] “Wherever is found what is called a paternal government, there is found state education. It has been discovered that the best way to insure implicit obedience is to commence tyranny in the nursery.” Benjamin Disraeli, Speech in the House of Commons [June 15, 1874] ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] Re: What's happening at the European Go Congress?
On Mon, 2008-08-11 at 16:16 -0300, Mark Boon wrote: On 11-aug-08, at 15:23, Don Dailey wrote: But is it really? Now instead of clearly defined rules, you enter the domain of judgment calls and these should be minimized. I don't agree with such an unforgiving attitude at all. It works for tournaments but not for demonstration games. You don't want to give fuel to those who argue yeah, but the computer can respond in a millisecond where the human has a physical response time of at least half a second. This is not an unforgiving attitude as you cast it. It is just the opposite. How is it you view taking a game away from the rightful winner as being forgiving? It shows no respect for the human being behind the program. It's real easy when you don't see anything but an unfeeling robot, but if it had been another person sitting behind that chair he would likely feel that someone had been heavy handed. It's easy to be gracious when you are not the victim. I would be angry if I worked hard to control my time usage, only for my opponent to be forgiven at my expense, despite the rules. And if this really is just a fun little demonstration game, then do not use clocks, that was certainly not in the spirit of things. - Don In demonstration games what is important is the spirit. And it doesn't do the computer-Go community any good for a program to persist in an absolutely lost position, play one for a hundred more moves that the human is physically unable to play in time. It will have to give one way or another. I also don't like the fixed time-limit very much because Go has such an unpredictable game-length. So Fisher time could be a solution. On the other hand, once the level of the programs becomes well established, programmers could also make it resign a lot sooner. In a 1,000 ELO game a 99% win-rate might occasionally still turn around. But they'll probably find out that as the level gets higher, say 3,000 ELO, you end up never turning around a 90% win-rate. Or maybe even 80%. If programmers want humans to play their software then they have to be also a little accomodating in that respect, even if that means giving up a game where you might still win once in a thousand times. Mark ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
[computer-go] Re: What's happening at the European Go Congress?
I think the result computer in hopelessly lost position resigns. is much more satisfactory than computer in hopelessly lost position wins by playing 100 additional pointless moves I think a human who used this tactic in a tournament situation might win the trophy, but would be unable to show his face again. ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
[computer-go] Re: What's happening at the European Go Congress?
I think the result computer in hopelessly lost position resigns. is much more satisfactory than computer in hopelessly lost position wins by playing 100 additional pointless moves I think a human who used this tactic in a tournament situation might win the trophy, but would be unable to show his face again. ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] Re: What's happening at the European Go Congress?
Sent from my iPhone On Aug 11, 2008, at 4:00 PM, Don Dailey [EMAIL PROTECTED] wrote: I would be angry if I worked hard to control my time usage, only for my opponent to be forgiven at my expense, despite the rules. Hmmm... This sounds very familiar... ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] Re: Strength of Monte-Carlo w/ UCT...
erm. you guys seem to be arguing for the sake of arguing, without a clear or precise definition of what you're even arguing about. there is a mathematical proof that go, for any fixed sized board, can be completely solved. there is a mathematical proof that given a fixed komi and fixed number of handicap stones, every game is either a forcible win or loss or draw for a particular one of the two players. we don't know this function yet, so we don't know if there's advantage for white or black or not, but it's guaranteed to exist. is proven to exist. there is a mathematical proof that current algorithms can solve go. it makes no sense to ask if there is a mathematical proof of anything related to humans. the two are simply incommensurate. the mathematical proofs are simply about whether a computer with a lot of memory and a lot of free time can win or force draws in every game of go against any player. and it turns out that this is true. whether or not computers can beat humans at go on a 19x19 board in a reasonable amount of time is unrelated to mathematics. * computers are getting better and better at go. most people on this mailing list are mainly interested in helping (*) to happen. s. On Sun, Aug 10, 2008 at 11:46 PM, Denis fidaali [EMAIL PROTECTED] wrote: Hi there. I do agree with your point Robert Waite. I have yet seen no such paper as one that would prove that there is such thing as scalability based on any mathematical proofs. So all your points at criticizing the mathematical certainty of the scalability, is probably 100% right. There is no such things as mathematical certainty there. It can be modelized easily, as you already did : what if the the evaluation function is giving on purpose wrong data. How would one mathematically prove that it doesn't ? You would at a minimum have to know WHAT the evaluation function ACTUALLY exactly is ... In fact all the evidences that we have gathered about the scalability may rather been surprising to some persons : why in hell does all that works so well ? But, it's a proven fact that it does indeed works well so far. So that it seems perfectly natural to speak such phrases as there are evidences that given the hardware we got in twenty years, human will be beaten by current algorithms. I don't see how those evidences can be qualified with the term mathematical, but they are here (hiding among us !). Now if someone has the feeling that maybe there is a roadblock, it has to be considered for what it is : a personal intuition. What is this intuitions precisely based on ? Why are you trying to share it with us in the first place. For myself, i believe that what you are trying to do, is to begin to analyses all the data the community has gathered so far, trying to understand why indeed it worked so well that it even beaten out a pro with a 9 stones handicap and with as few as 1.7 million evaluations/second (running on some 800 hundreds cores). To the point that the pro felt he had no chances of wining at all with that much of a handicap. Your are trying to understand this, and are probably right on track for that goal. The term mathematical is very valuable to you, and you'll find it that it has a much wider use (on this list) than what you would like it to. But now, mathematics as proven to be of little use in the context of go programming lately. It's more of a physician world. You make up a (mathematical) model. You test it again reality via experimentations. You then get empirical certitudes that the model is indeed correct. There is no way of mathematically proving that light speed would still be constant if i chose to dance naked on the champs-Elysée some day. You'll definitely find no paper on that. Yet to speak of it as mathematically certain, is probably not as wrong as it sound. But as it is, i'm playing the devil advocates here. I'm totally agreeing with you. I found your way to fight irrationnality very interesting indeed. It's been very refreshing. - Robert Waite has wrote : I would really like to see what paper you are referring to. Do you mean Bandit based Monte-Carlo Planning? Please post the name of the paper which you are referring to. I do not think that the empirical evidence is overwhelming that it is scalable in a practical way for the problem of beating a human. Now the topic has moved to scalable to beat a human and I disagree with the interpretation of the data. We are both interpreting data. Your data doesn't count as a theory.. where you reduced my theory to one that has no data. We are both interpreting the same data. Diminishing returns was just an example of something that could be a roadblock. I was questioning how this necessarily scales to humans. It seems more data is needed from MC-programs vs. humans to make a rigorous theory of scalability. So far.. the only scalability
[computer-go] Re: What's happening at the European Go Congress?
But let's not exaggerate. This was not just a simple matter of filling empty points. It was obviously unclear enough to some of us that it required some analysis. Even the strong Leela did not see this as merely filling in the empty points. At the very least the game should not be stopped until both players understand the position. - Don On Mon, 2008-08-11 at 13:16 -0700, Dave Dyer wrote: I think the result computer in hopelessly lost position resigns. is much more satisfactory than computer in hopelessly lost position wins by playing 100 additional pointless moves I think a human who used this tactic in a tournament situation might win the trophy, but would be unable to show his face again. ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
[computer-go] cgos bayes rated lists
The main page now has links to a bayes rated chart for each board size. This will be updated periodically, a period to be determined later but at least a couple of times per day. http://cgos.boardspace.net/ I am going to only show recently playing bots, but for not I'm showing ALL bots. So the 9x9 page shows more than even the all-time list. - Don ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
[computer-go] Re: mogo beats pro!
You clearly don't understand the principles of alpha/beta pruning. It is an admissible technique which means it guarantee's the same result as searching the entire tree, but only requires a very tiny subset of the entire tree. Okay... congratulations... you are right... if you are able to generate a completely pruned tree using alpha/beta pruning... you don't have to generate the whole game tree. But exactly how are you going to do this? In chess... you can look 9 moves in and quickly evaluate if a branch is looking good or is looking bad (wow... you just lost a rook three moves in). It seems you can't really do this until pretty deep in the tree for go. Plenty of moves would look bad 20 moves deep.. but would turn out to be good 80 moves later. How can you know if a move is good or not until you move towards the end of a branch? Isn't this just a little computationally expensive? You need some sort of early evaluation function... and we don't seem to have that yet. It seems that alpha/beta pruning hit a roadblock a long time ago in go. Now we have MC... which as you increase the number of samples.. you start to get closer to an equivalent alpha/beta. But... there are still huge groups of samples that are not checked... and if you want to somehow prove you have the best move... how will you do it? Will you make the sample size equivalent to the number of possible samples? How will you do this practically? You can only state with a certain confidence that you did make the best move and this would be bound by our computational resources. It is pretty obvious that go has many computationally difficult sub problems. MC is a shortcut... and we don't know yet how it will perform in the wild. Let's say you have 381 possible first moves and you find that the first one you look at wins the game by 6.5 points. That is a huge if. Man.. I would love the algorithm you did that with (or the hardware). You really looked at one opening move.. and realized that it won by 6.5 points, unconditionally? Was this somehow considered practical? Or are we on a machine that has as much time and space as you want? I don't know... most of your emails are pretty condescending and you tend to only respond to a point that you see a weakness in. In fact, I am not sure if you have made any kind of concession towards my viewpoint. So far... I got a nice illustration about how your data is a duck and a nice explanation that seemed tailored to a three year old. I have yet to be given your sources of information. Plenty of times.. I hear it is proven scalable to perfect play... and the only paper I can imagine you are talking about is the bandit problem paper. After I read it... it seems that it is not used as an argument of practicality.. which is the way you used it. It is used as an argument of correctness. I also heard of overwhelming empirical evidence that this will scale to beat humans. Someone else sent me your personal experiments on 9x9 and 13x13. I do not really see how this obviously means it will scale in the same way vs. a human (it doesn't, after all, include any humans). I don't think you could make that kind of statement in an academic paper based on the data. But quite clearly.. to hear you tell it.. it is the most obvious thing imaginable and only a fool wouldn't see how clear it is. Your duck drawing was really clever.. but I still don't see your overwhelming evidence. Maybe I am stupid. There is no doubt it is successful, I never denied that. I just questioned whether it is the panacea of go programming and all we need is hardware. I thinking questioning a potential problem is a bit different from declaring that it has a problem. Your burden of proof argument could lie with you. I would wager that a large portion of the go/computer go world think computers ain't gonna do it anytime soon. If you look at the game records... computer go hasn't been doing so well. So now you present new evidence but it is in a very early stage. Perhaps it is you that would need to prove your belief.. not me. Of course... we are on a mailing list so you don't have to prove anything to me. But earlier you made it quite clear that you don't need to prove anything... that I have the complete burden of proof. If you published a paper on this topic.. does that mean that the burden of proof against your paper lies with the whole academic community? I would hope that you would have to defend your paper... instead of everyone else having to prove you wrong. (You very easily could be right but I questioned your certainty). Unless you want to more calmly discuss this... and are willing to accept some sort of possibility that you are not 100% correct... I don't think we need to continue this. It is increasing both of our blood pressures and I am sure many people on the list find the whole discussion annoying. And feel free to send one more rant my way... I won't respond unless it is civil.. but at least you can feel I didn't have the
[computer-go] Re: mogo beats pro!
You clearly don't understand the principles of alpha/beta pruning. It is an admissible technique which means it guarantee's the same result as searching the entire tree, but only requires a very tiny subset of the entire tree. Okay... congratulations... you are right... if you are able to generate a completely pruned tree using alpha/beta pruning... you don't have to generate the whole game tree. But exactly how are you going to do this? In chess... you can look 9 moves in and quickly evaluate if a branch is looking good or is looking bad (wow... you just lost a rook three moves in). It seems you can't really do this until pretty deep in the tree for go. Plenty of moves would look bad 20 moves deep.. but would turn out to be good 80 moves later. How can you know if a move is good or not until you move towards the end of a branch? Isn't this just a little computationally expensive? You need some sort of early evaluation function... and we don't seem to have that yet. It seems that alpha/beta pruning hit a roadblock a long time ago in go. Now we have MC... which as you increase the number of samples.. you start to get closer to an equivalent alpha/beta. But... there are still huge groups of samples that are not checked... and if you want to somehow prove you have the best move... how will you do it? Will you make the sample size equivalent to the number of possible samples? How will you do this practically? You can only state with a certain confidence that you did make the best move and this would be bound by our computational resources. It is pretty obvious that go has many computationally difficult sub problems. MC is a shortcut... and we don't know yet how it will perform in the wild. Let's say you have 381 possible first moves and you find that the first one you look at wins the game by 6.5 points. That is a huge if. Man.. I would love the algorithm you did that with (or the hardware). You really looked at one opening move.. and realized that it won by 6.5 points, unconditionally? Was this somehow considered practical? Or are we on a machine that has as much time and space as you want? I don't know... most of your emails are pretty condescending and you tend to only respond to a point that you see a weakness in. In fact, I am not sure if you have made any kind of concession towards my viewpoint. So far... I got a nice illustration about how your data is a duck and a nice explanation that seemed tailored to a three year old. I have yet to be given your sources of information. Plenty of times.. I hear it is proven scalable to perfect play... and the only paper I can imagine you are talking about is the bandit problem paper. After I read it... it seems that it is not used as an argument of practicality.. which is the way you used it. It is used as an argument of correctness. I also heard of overwhelming empirical evidence that this will scale to beat humans. Someone else sent me your personal experiments on 9x9 and 13x13. I do not really see how this obviously means it will scale in the same way vs. a human (it doesn't, after all, include any humans). I don't think you could make that kind of statement in an academic paper based on the data. But quite clearly.. to hear you tell it.. it is the most obvious thing imaginable and only a fool wouldn't see how clear it is. Your duck drawing was really clever.. but I still don't see your overwhelming evidence. Maybe I am stupid. There is no doubt it is successful, I never denied that. I just questioned whether it is the panacea of go programming and all we need is hardware. I thinking questioning a potential problem is a bit different from declaring that it has a problem. Your burden of proof argument could lie with you. I would wager that a large portion of the go/computer go world think computers ain't gonna do it anytime soon. If you look at the game records... computer go hasn't been doing so well. So now you present new evidence but it is in a very early stage. Perhaps it is you that would need to prove your belief.. not me. Of course... we are on a mailing list so you don't have to prove anything to me. But earlier you made it quite clear that you don't need to prove anything... that I have the complete burden of proof. If you published a paper on this topic.. does that mean that the burden of proof against your paper lies with the whole academic community? I would hope that you would have to defend your paper... instead of everyone else having to prove you wrong. (You very easily could be right but I questioned your certainty). Unless you want to more calmly discuss this... and are willing to accept some sort of possibility that you are not 100% correct... I don't think we need to continue this. It is increasing both of our blood pressures and I am sure many people on the list find the whole discussion annoying. And feel free to send one more rant my way... I won't respond unless it is civil.. but at least you can feel I didn't have the
[computer-go] Re: Strength of Monte-Carlo w/ UCT...
Steve, You mentioned three proofs relating to go... could you post the links to the papers? it makes no sense to ask if there is a mathematical proof of anything related to humans. I didn't ask for a mathematical proof saying if a computer can beat a human. I asked in a roundabout way if this algorithm (or any known algorithm) has a proven complexity that is somehow tractable or useful to beat humans. Just by throwing human in does not mean you are out of the realms of math. What about statistics? The object of many statistical models is a group of people. So please don't say it makes no sense to ask about mathematical proofs of anything related to humans. A mathematical proof can have a result that affects humans. If it was proven tomorrow that there is a set of algorithms that can solve the game in poly time.. we could draw relevant conclusions with regards to beating a human being. Relating humans to math does not destroy the accuracy of the relation. whether or not computers can beat humans at go on a 19x19 board in a reasonable amount of time is unrelated to mathematics. Why? Let's say you can prove that the game is solvable so that black wins. Let's say that you can prove that it is solvable in linear time. You can then infer that we could build a machine to play the solved game and beat a human unconditionally. Why can't you use the math here to make a statement about beating humans? ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] Re: What's happening at the European Go Congress?
On Mon, 11 Aug 2008, Don Dailey wrote: But let's not exaggerate. This was not just a simple matter of filling empty points. It was. It was obviously unclear enough to some of us that it required some analysis. Even the strong Leela did not see this as merely filling in the empty points. That's because it involves a Seki that Leela does not handle properly, but any 10 kyu should recognize. Christoph ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] Re: Strength of Monte-Carlo w/ UCT...
Robert Waite wrote: whether or not computers can beat humans at go on a 19x19 board in a reasonable amount of time is unrelated to mathematics. Why? Let's say you can prove that the game is solvable so that black wins. Let's say that you can prove that it is solvable in linear time. You can then infer that we could build a machine to play the solved game and beat a human unconditionally. Why can't you use the math here to make a statement about beating humans? Because solving the game is not a prerequesite for beating the humans. There are very obvious examples(chess) -- GCP ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] Re: mogo beats pro!
On Mon, 2008-08-11 at 17:29 -0400, Robert Waite wrote: Okay... congratulations... you are right... if you are able to generate a completely pruned tree using alpha/beta pruning... you don't have to generate the whole game tree. But exactly how are you going to do this? In chess... you can look 9 moves in and quickly evaluate if a branch is looking good or is looking bad (wow... you just lost a rook three moves in). It seems you can't really do this until pretty deep in the tree for go. Plenty of moves would look bad 20 moves deep.. but would turn out to be good 80 moves later. How can you know if a move is good or not until you move towards the end of a branch? Isn't this just a little computationally expensive? You need some sort of early evaluation function... and we don't seem to have that yet. I'm only responding to your comment that you have to look at the entire tree. You DON'T have to look at the entire tree in order to get a perfect answer and find the very best move with 100% certainty. That's all I'm saying. It seems like you have a gift for obfuscation. And yes, of course it's computationally expensive even with alpha beta pruning which saves many orders of magnitude times more work. It's trivial to write either a mini-max or alpha beta searcher program that searches to the end of the game. It doesn't require much memory either because you only store 1 line at a time. Of course such a program will not finish executing in the lifetime of this universe. But it is easy to write. - Don ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] Cultural differences: players vs programmers
terry mcintyre wrote: I guess we're all different. Last week, I actually did win a 9-stone handicap game in a simul match against a pro, but I'm not about to claim that this gives me bragging rights or anything, lol. [explanation of how this game made you a better player deleted] I see. If my program won on time in an obviously lost position, I'd be turning every rock to find a way to improve the actual play; that matters much more to me than the win-loss record. I would recommend to the human player to improve her time management. I am sure good time management makes you a better player :) As an aside, isn't the level of the actual play *defined* by the win-loss record? -- GCP ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] Re: What's happening at the European Go Congress?
Jason House wrote: On Aug 11, 2008, at 4:00 PM, Don Dailey [EMAIL PROTECTED] wrote: I would be angry if I worked hard to control my time usage, only for my opponent to be forgiven at my expense, despite the rules. Hmmm... This sounds very familiar... Yes. Notice how there is a clear discrimination on this list in favor of 19-year old females. This is an unexplicable attitude for a group of 16 to 70 year old male computer freaks. Leela now plays faster in some situations as a result of the loss on time against HouseBot. It might have been a factor in this game. -- GCP ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] Re: mogo beats pro!
We are in agreement on the general nature of things, but seeing it in person was just so amazing. I did see comments about the quality of the pro, but it may have been in the game chat rather than here. I slept very little over the 10 days in Portland, so things are all mixed up in my head. Cheers, David On 10, Aug 2008, at 1:06 PM, Mark Boon wrote: On 10-aug-08, at 13:11, David Doshay wrote: As an aside, the pro in question won the US Open, so comments about him being a weak pro seem inappropriate. I don't see where anybody questioned the level of the pro. As far as I'm concerned I consider a Korean (is that correct?) 8-dan pro to be close enough to the ultimate top as to be indistinguishable for the sake of this discussion. All I tried to do was put this achievement in perspective to other achievements in the past. I don't think anybody disputes the great progress that has been made either, no matter the hardware requirements. I don't think a computer will beat a pro on even in ten years just using the faster hardware that will be avaliable by then. I believe considerable improvements will have to be made in the software as well. Is it impossible? No, it's not impossible. But it's impossible to make predictions about it, IMO. If I had to put money on it I'd rather go for 20 years than 10 years. But even 20 years isn't going to be a lay-up. But if in ten years we have a million-CPU computer to our disposal and there has been progress in the software in the order of 4-5 stones as well we might be getting close. I say 'might', as I'd like to see more games. Considering the low availability of such a powerful computer, that data needed to make stronger claims is a bit hard to come by. Mark ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/ ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
[computer-go] Re: Strength of Monte-Carlo w/ UCT...
* whether or not computers can beat humans at go on a * 19x19 board in a reasonable amount of time is unrelated * to mathematics.* Because solving the game is not a prerequesite for beating the humans. There are very obvious examples(chess) I never questioned that. The way I read Steve's statement... mathematics has no bearing on the ability of a human to beat a computer. Maybe he doesn't quite mean that statement but I was responding to it. I offered a way that math could lead to an important finding for computer go. Most of the papers I have read on computer go did include math. It seems that complexity of algorithms is something that affects all go programmers alike. I don't see why Steve is so against me pondering the complexity of various solutions for making a computer that plays go well. It also seems that his proofs are contested and not necessarily accepted as fact. ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] Games vs professionals
On 11, Aug 2008, at 4:56 AM, Basti Weidemyr wrote: - The review of Xiao Ai Lin vs Leela: http://www.weidemyr.com/egc/cg/XiaoAiLin_Leela-review.sgf - Several people at the congress expressed worries to me about what would happen to the sport Go, if computer programs became stronger and threatened to defeat the strongest men. Go would lose its advantage over chess, they said, and people would feel redundant as computers could do it better. Nobody in Portland asked me this question, either before or after the Mogo/Kim match. One man asked me repeatedly to quit running challenges between professionals and computers. The professionals themselves became very nervous when we asked them to play against a computer. It is not hard to imagine the bold headlines after losing, but it is hard to imagine them after winning. I had an interchange from the other side. I wanted to know the sensible thing to try next, and asked a pro who will remain un-named. He thought the whole question absurd. He felt that the match meant nothing and that the distance between the program and any pro was still so large that to discuss next was equally meaningless. And this is a pro that I have spoken to many times, who has always been very polite. I have only seen him more emphatic and less composed once when he yelled at me If you can't read well then don't try so hard to read it out! The game between MoGo and Kim Myung Wan was unique, since MoGo run on a large cluster and interesting to watch. (Congratulations MoGo team!) It was also a great way of showing people the progress that has been made in computer-go recently. However, maybe we do not need to use these kinds of challenges as a means of getting media attention. If the reporters understood what we were doing and got it right it would be useful ... I think we do need these matches, but we need to be careful in how they are structured and how often we have them. I also understand the difficult position the pro is in: If they win it means nothing but if they loose they are for all history the goat. But as Don keeps saying, we are not treading on new ground, the exact same thing happened with chess. We would like to find a way to cooperate with the traditional go- community with little friction. What do you think? I agree that we need to insert ourselves politely, but I think still with a little force. I will not go into all of the details of the meeting with the AGA regarding getting honest and accurate ratings for programs, but there were many obvious prejudices being expressed and more than a few Catch-22 situations to carefully be danced around. Fortunately, the incoming AGA President saw the situation clearly and crafted a comfortable compromise. We will be accepted, even if only Incrementally. Best regards Basti Weidemyr kgs: sestir ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/ ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] Re: mogo beats pro!
It is of no consequence what words WE use to describe this. Journalists will ALWAYS print it that way. If you use too many big words or ideas that are accurate but convoluted, you will either not get the publicity or the journalist will make up something even more absurd. Sorry if I am a bit over sensitive ... getting misquoted, my work ignored, and getting credit for the work of others in this past week has me very aware of how these people work. They are on a deadline and meeting the deadline with a headline that captures a lay reader's attention is the only priority. I know how my attempts to get a correction were greeted ... Cheers, David On 11, Aug 2008, at 8:37 AM, Hideki Kato wrote: Hi all, I'd like to say first Congratulations! to MoGo team. I have a question. Why do you all call the game as human vs. computer? It's obviously a match between Kim 8p and MoGo, a program developped by MoGo team, running on a supercomputer. As both MoGo and the supercomputer were developped by human, the game is clearly (a special type of) human vs. human. I'm afraid it may raise unnecessary emotional thoughts of against computers among people. It might be better to call such a game something of a style a professinal Goplayer vs. a program with its developper(s) to emphasize the program was created by human. -Hideki ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] Re: What's happening at the European Go Congress?
On Mon, Aug 11, 2008 at 7:58 PM, Gunnar Farnebäck [EMAIL PROTECTED] wrote: Erik van der Werf wrote: For the final position in the game record any strong human player will tell you that the game is clearly over. No points are left to be gained and the result is obvious. Actually there's one point left to gain in the seki, since the game is played with Chinese rules. ;-) You're right, my reply was sloppy (it seems I'm too much used to Japanese rules). Also I should have read GCP's email more carefully; I did not realize that his program, even with a large tree, would not be able to recognize the seki. I knew of course that the original Mogo playouts had this problem, but I thought all strong programs had solved it by now... Erik ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] Re: Strength of Monte-Carlo w/ UCT...
You mentioned three proofs relating to go... could you post the links to the papers? the first two statements are consequences of the following: all two-person, finite, zero-sum games have solutions. * for a more precise statement, see john von neumann's 1928 paper: Von Neumann, J: Zur theorie der gesellschaftsspiele Math. Annalen. 100 (1928) 295-320 and the definitions of the terms used in the statement (*). or perhaps more helpfully, a modern treatment on the subject of game theory. the third statement is true simply because the minimax algorithm exists. i am not claiming that any of this has anything to do with the actual problem of beating a human. i am not making this claim because i also make the claim that beating humans at go is pretty much unrelated to the mathematics in these proofs. I didn't ask for a mathematical proof saying if a computer can beat a human. I asked in a roundabout way if this algorithm (or any known algorithm) has a proven complexity that is somehow tractable or useful to beat humans. Just by throwing human in does not mean you are out of the realms of math. What about statistics? The object of many statistical models is a group of people. So please don't say it makes no sense to ask about mathematical proofs of anything related to humans. A mathematical proof can have a result that affects humans. If it was proven tomorrow that there is a set of algorithms that can solve the game in poly time.. we could draw relevant conclusions with regards to beating a human being. Relating humans to math does not destroy the accuracy of the relation. algorithms do not have complexities, problems do. algorithms may have asymptotic runtimes, but even this isn't always true. poly time doesn't mean tractable. just like exptime doesn't mean intractable. there's a coefficient in front of the polynomial (or exponential function) that can radically affect the real-world tractability of the problem. another thing is that complexity classes are used to describe problems like, find me an algorithm that can solve the game of go for *any* sized board. in this sense, go is quite difficult. however, nobody on this list is seriously hoping to write a program to solve go for any sized board and hoping that it will succeed on a 19x19 board. what they are doing is trying to engineer very good programs to beat humans on a 19x19 board. whether or not computers can beat humans at go on a 19x19 board in a reasonable amount of time is unrelated to mathematics. Why? Let's say you can prove that the game is solvable so that black wins. Let's say that you can prove that it is solvable in linear time. You can then infer that we could build a machine to play the solved game and beat a human unconditionally. Why can't you use the math here to make a statement about beating humans? what if the linear function is this one: time = 10^10^10^10^10^10^10 * (size of board) that doesn't imply tractability, but it is still linear. the problem that i mentioned earlier: how much effort is required to completely solve the game of go for _any_ given boardsize is known not to be linear. it's known not to be polynomial. the problem of solve the game of go for a 19x19 board is known to be a *constant*. since there are only a finite number of legal board situations, there are only a finite number of legal games. enumerating all of these takes a constant amount of time. storing these takes a constant amount of space. this is not useful to make good programs for playing go, however. the thing is, all of the talk of asymptotics that you seem to be referring to are perfectly useful to prove things about arbitrary games on arbitrary sized boards, but when you have a fixed-size board, what matters is much more an issue of engineering a fixed problem. s. ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/ ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] Re: What's happening at the European Go Congress?
- Original Message From: Christoph Birk [EMAIL PROTECTED] On Mon, 11 Aug 2008, Don Dailey wrote: It was obviously unclear enough to some of us that it required some analysis. Even the strong Leela did not see this as merely filling in the empty points. That's because it involves a Seki that Leela does not handle properly, but any 10 kyu should recognize. Thank you! At present, computer go programs may be strong relative to each other, and they may actually beat some humans of moderate ability, especially at timescales too quick for amateur humans, but most programs also have high-kyu-sized gaps in their knowledge, including seki and nakade concepts. We won't see programs regularly beating pros until those gaps are filled. Substituting billions of playouts for a life-or-death or seki analysis which 10 kyu players can manage in seconds is inefficient; computers could be doing something more effective with that time. ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
[computer-go] beating mogo with time (funny post)
I am about as strong as the mogos running on kgs. I get a kick out of trying to learn how to beat the mogos there. It's certainly not as easy as beating gnu go with a few stones (just surround it) or beating aya giving it 6 stones (just don't make tactical mistakes, but take advantage of its tactical mistakes). With mogo there are other techniques, but this isn't a post about the strategical considerations for beating mogo at go thinking. Often, mogo can beat me because the game is short, thus I can't think quickly enough in some of the complexity it presents. I get back, though, occasionally. When I stay properly ahead of mogo on time, sometimes in the endgame it has the standard few points ahead, but sometimes mogo doesn't estimate time properly... and gets down to 5 ... 4 ... 3 ... 2 ... Then mogo gets scared. It doesn't play any more stones, it just passes. (pass doesn't remove time on kgs). I calmy destroy one of mogos groups, quickly making 5 or 6 moves in a row ... I still have 10 seconds left ... mogo passes again, I pass, I win ... Of course I was the loser of the real go game, but because the mogos are running with bizarre time settings, and they beat me sometimes because of them, I get my revenge. Pedro Costa Rica ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] Re: mogo beats pro!
I have a question. Why do you all call the game as human vs. computer? It's obviously a match between Kim 8p and MoGo, a program developped by MoGo team, running on a supercomputer. Quick answer: it is the established term. (human-machine is perhaps even more common?) Longer answer: Mogo is on its own choosing moves; the programmers cannot help it while it is playing. Similarly the human player is on his own and not allowed to discuss positions with his teachers, students, go books, etc. (BTW, just about everybody here has congratulated the Mogo team not Mogo. But the human side is the same: if the human player won an important game and his parents were in the room people would go up and shake their hand and say Congratulations, you must be very proud.) I'm afraid it may raise unnecessary emotional thoughts of against computers among people. People like that will get emotional whichever words you use. Darren -- Darren Cook, Software Researcher/Developer http://dcook.org/mlsn/ (English-Japanese-German-Chinese-Arabic open source dictionary/semantic network) http://dcook.org/work/ (About me and my work) http://darrendev.blogspot.com/ (blog on php, flash, i18n, linux, ...) ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] Re: What's happening at the European Go Congress?
[The pro] was also a bit unlucky in the sense that Leela did not understand it was dead lost. I use quotes because had it understood better it was losing, it would have put up more of a fight :-) My first impression of watching the game was that Leela was handicapped by having a handicap. By that I mean it would have seen itself so far ahead for the first few moves that is was playing arbitrarily. This is a direct consequence of the UCT algorithm playing for the win, instead of trying to maximize the score. I'm fine with that (please see the archives for numerous passionate discussion on the subject), but surely you need an opening book to allow for the fact that evaluations are going to have a high error margin in the early game? (I'll go out on a limb and say black 3 was a mistake; I'm sure it is a win at 0.5pt komi, but I strongly suspect it is dubious at 5.5pt komi or higher.) From another angle: if a UCT computer program is being given a handicap against a stronger player it should lie to itself about the komi at the start. It could then gradually adjust komi so it is at the correct value by the early middle game (e.g. move 6 in 9x9 go, move 30 in 19x19 go). Or it could keep adjusting komi (until it reaches the actual komi) so that it thinks it is only just winning. Darren -- Darren Cook, Software Researcher/Developer http://dcook.org/mlsn/ (English-Japanese-German-Chinese-Arabic open source dictionary/semantic network) http://dcook.org/work/ (About me and my work) http://darrendev.blogspot.com/ (blog on php, flash, i18n, linux, ...) ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] beating mogo with time (funny post)
If we do concentrate for just a moment on how to beat mogo, I can report that in the 3 blitz games the pro figured out that multistep kos were the easy way. But in the longer game he presented the same pattern to mogo to start it, but mogo played differently. I thought that was a huge difference. Cheers, David On 11, Aug 2008, at 5:10 PM, Peter Christopher wrote: I am about as strong as the mogos running on kgs. I get a kick out of trying to learn how to beat the mogos there. It's certainly not as easy as beating gnu go with a few stones (just surround it) or beating aya giving it 6 stones (just don't make tactical mistakes, but take advantage of its tactical mistakes). With mogo there are other techniques, but this isn't a post about the strategical considerations for beating mogo at go thinking. Often, mogo can beat me because the game is short, thus I can't think quickly enough in some of the complexity it presents. I get back, though, occasionally. When I stay properly ahead of mogo on time, sometimes in the endgame it has the standard few points ahead, but sometimes mogo doesn't estimate time properly... and gets down to 5 ... 4 ... 3 ... 2 ... Then mogo gets scared. It doesn't play any more stones, it just passes. (pass doesn't remove time on kgs). I calmy destroy one of mogos groups, quickly making 5 or 6 moves in a row ... I still have 10 seconds left ... mogo passes again, I pass, I win ... Of course I was the loser of the real go game, but because the mogos are running with bizarre time settings, and they beat me sometimes because of them, I get my revenge. Pedro Costa Rica ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/ ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] Re: What's happening at the European Go Congress?
For the final position in the game record any strong human player will tell you that the game is clearly over. No points are left to be gained and the result is obvious. Actually there's one point left to gain in the seki, since the game is played with Chinese rules. ;-) You're right, my reply was sloppy (it seems I'm too much used to Japanese rules). Also I should have read GCP's email more carefully; I did not realize that his program, even with a large tree, would not be able to recognize the seki. I knew of course that the original Mogo playouts had this problem, but I thought all strong programs had solved it by now... Can someone confirm this one way or the other? Has Mogo started explicitly recognizing seki, and if so which release version did that start at? More generally, has anyone seen increases/decreases in overall strength from explicitly checking for seki at leaf nodes? I remember reading that when nakade support was adding to Mogo it made it slightly stronger at 9x9, but weaker at 19x19. Was this version released, and can nakade support be switched on and off at the commandline? Darren -- Darren Cook, Software Researcher/Developer http://dcook.org/mlsn/ (English-Japanese-German-Chinese-Arabic open source dictionary/semantic network) http://dcook.org/work/ (About me and my work) http://darrendev.blogspot.com/ (blog on php, flash, i18n, linux, ...) ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] Re: What's happening at the European Go Congress?
On Tue, 2008-08-12 at 09:55 +0900, Darren Cook wrote: [The pro] was also a bit unlucky in the sense that Leela did not understand it was dead lost. I use quotes because had it understood better it was losing, it would have put up more of a fight :-) My first impression of watching the game was that Leela was handicapped by having a handicap. By that I mean it would have seen itself so far ahead for the first few moves that is was playing arbitrarily. I was blasted for making that observation many months ago concerning the possibility of handicap matches on CGOS. I thought it not a good idea for Monte Carlo players because each player starting with a dead won or dead lost game. The response was that it didn't matter, the programs would still fight. So I yielded to the opinion of others since I am not a go player. I now think they were probably right. MCTS still tries to maximize the chances of winning. If you are up 8 or 9 stones, that is STILL the right strategy isn't it? Also, if you are down 8 or 9 stones, maximizing your winning chances is still the right strategy, right? I'm trying to come up with some kind of analogy to real life. How about investing your money? Let's say you play a game where the goal is to turn 500 thousand into 1 million dollars in 10 years. Double your money in 10 years is not particularly difficult so if the only thing that matters is winning this game then you would use very conservative investments. This is like being up 9 stones because in theory you have a relatively simple task to perform, just double your money. The temptation is to be foolish by thinking if you are a lot more aggressive, you can get ahead of the game and get there faster. Surely, if you have a good year or two, you can coast the rest of the way! Have you ever been with someone who is about to run out of gas? They want to drive FASTER thinking that if they get there faster, they will use less fuel. Or maybe they just get anxious which causes you to drive a little faster. This is a direct consequence of the UCT algorithm playing for the win, instead of trying to maximize the score. I'm fine with that (please see the archives for numerous passionate discussion on the subject), but surely you need an opening book to allow for the fact that evaluations are going to have a high error margin in the early game? (I'll go out on a limb and say black 3 was a mistake; I'm sure it is a win at 0.5pt komi, but I strongly suspect it is dubious at 5.5pt komi or higher.) From another angle: if a UCT computer program is being given a handicap against a stronger player it should lie to itself about the komi at the start. It could then gradually adjust komi so it is at the correct value by the early middle game (e.g. move 6 in 9x9 go, move 30 in 19x19 go). Or it could keep adjusting komi (until it reaches the actual komi) so that it thinks it is only just winning. It could turn out that the best strategy is simply to let the opponent play desperately and not over-react, because to have any chance when giving 9 stones you must in some sense over-play it. - Don Darren ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] Re: mogo beats pro!
Hi Darren, Darren Cook: [EMAIL PROTECTED]: I have a question. Why do you all call the game as human vs. computer? It's obviously a match between Kim 8p and MoGo, a program developped by MoGo team, running on a supercomputer. Quick answer: it is the established term. (human-machine is perhaps even more common?) Sure, it's the same here in Japan. Longer answer: Mogo is on its own choosing moves; the programmers cannot help it while it is playing. Similarly the human player is on his own and not allowed to discuss positions with his teachers, students, go books, etc. It does not change the fact MoGo was developped by the programmers. And the fact the programmers spent many resources, like the people fighting at Beijing right now, to develop MoGo. (BTW, just about everybody here has congratulated the Mogo team not Mogo. But the human side is the same: if the human player won an important game and his parents were in the room people would go up and shake their hand and say Congratulations, you must be very proud.) (Really the same?) I'm afraid it may raise unnecessary emotional thoughts of against computers among people. People like that will get emotional whichever words you use. Don't you think it cannot be changed or, at least, improved? #Assuming we agree it's unnecessary. Hideki -- [EMAIL PROTECTED] (Kato) ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] Re: What's happening at the European Go Congress?
Don Dailey: [EMAIL PROTECTED]: On Tue, 2008-08-12 at 09:55 +0900, Darren Cook wrote: [The pro] was also a bit unlucky in the sense that Leela did not understand it was dead lost. I use quotes because had it understood better it was losing, it would have put up more of a fight :-) My first impression of watching the game was that Leela was handicapped by having a handicap. By that I mean it would have seen itself so far ahead for the first few moves that is was playing arbitrarily. I was blasted for making that observation many months ago concerning the possibility of handicap matches on CGOS. I thought it not a good idea for Monte Carlo players because each player starting with a dead won or dead lost game. The response was that it didn't matter, the programs would still fight. So I yielded to the opinion of others since I am not a go player. I now think they were probably right. MCTS still tries to maximize the chances of winning. If you are up 8 or 9 stones, that is STILL the right strategy isn't it? Also, if you are down 8 or 9 stones, maximizing your winning chances is still the right strategy, right? I think NO because of the model of the opponent. MCTS uses itself for the model but it's obvously not correct in hadicapped games. Hideki I'm trying to come up with some kind of analogy to real life. How about investing your money? Let's say you play a game where the goal is to turn 500 thousand into 1 million dollars in 10 years. Double your money in 10 years is not particularly difficult so if the only thing that matters is winning this game then you would use very conservative investments. This is like being up 9 stones because in theory you have a relatively simple task to perform, just double your money. The temptation is to be foolish by thinking if you are a lot more aggressive, you can get ahead of the game and get there faster. Surely, if you have a good year or two, you can coast the rest of the way! Have you ever been with someone who is about to run out of gas? They want to drive FASTER thinking that if they get there faster, they will use less fuel. Or maybe they just get anxious which causes you to drive a little faster. This is a direct consequence of the UCT algorithm playing for the win, instead of trying to maximize the score. I'm fine with that (please see the archives for numerous passionate discussion on the subject), but surely you need an opening book to allow for the fact that evaluations are going to have a high error margin in the early game? (I'll go out on a limb and say black 3 was a mistake; I'm sure it is a win at 0.5pt komi, but I strongly suspect it is dubious at 5.5pt komi or higher.) From another angle: if a UCT computer program is being given a handicap against a stronger player it should lie to itself about the komi at the start. It could then gradually adjust komi so it is at the correct value by the early middle game (e.g. move 6 in 9x9 go, move 30 in 19x19 go). Or it could keep adjusting komi (until it reaches the actual komi) so that it thinks it is only just winning. It could turn out that the best strategy is simply to let the opponent play desperately and not over-react, because to have any chance when giving 9 stones you must in some sense over-play it. - Don Darren ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/ -- [EMAIL PROTECTED] (Kato) ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] Re: mogo beats pro!
David, I didn't intend to offend any person in this list, sorry for short of my words. I'm just trying to prevent people misunderstand the truth. Hideki David Doshay: [EMAIL PROTECTED]: It is of no consequence what words WE use to describe this. Journalists will ALWAYS print it that way. If you use too many big words or ideas that are accurate but convoluted, you will either not get the publicity or the journalist will make up something even more absurd. Sorry if I am a bit over sensitive ... getting misquoted, my work ignored, and getting credit for the work of others in this past week has me very aware of how these people work. They are on a deadline and meeting the deadline with a headline that captures a lay reader's attention is the only priority. I know how my attempts to get a correction were greeted ... Cheers, David On 11, Aug 2008, at 8:37 AM, Hideki Kato wrote: Hi all, I'd like to say first Congratulations! to MoGo team. I have a question. Why do you all call the game as human vs. computer? It's obviously a match between Kim 8p and MoGo, a program developped by MoGo team, running on a supercomputer. As both MoGo and the supercomputer were developped by human, the game is clearly (a special type of) human vs. human. I'm afraid it may raise unnecessary emotional thoughts of against computers among people. It might be better to call such a game something of a style a professinal Goplayer vs. a program with its developper(s) to emphasize the program was created by human. -Hideki ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/ -- [EMAIL PROTECTED] (Kato) ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] Re: What's happening at the European Go Congress?
On 11, Aug 2008, at 7:23 PM, Don Dailey wrote: On Tue, 2008-08-12 at 09:55 +0900, Darren Cook wrote: My first impression of watching the game was that Leela was handicapped by having a handicap. By that I mean it would have seen itself so far ahead for the first few moves that is was playing arbitrarily. I was blasted for making that observation many months ago concerning the possibility of handicap matches on CGOS. I thought it not a good idea for Monte Carlo players because each player starting with a dead won or dead lost game. The response was that it didn't matter, the programs would still fight. I wonder if this was part of the beginning move selection of Mogo in the games against Mr Kim. Can anyone on that team check their logs and respond? It could turn out that the best strategy is simply to let the opponent play desperately and not over-react, because to have any chance when giving 9 stones you must in some sense over-play it. This exact point was made in the post game analysis by Mr Kim. He explained that he expected about 1 dan replies to his approach in the lower right, and thus played to live in the corner and have an extension across the bottom. An observer said Oh, so you made an overplay. Mr Kim replied I have to overplay (against 9 stones). He later showed how he would have played it had he expected mogo to find what he called 4 or 5 dan moves. He also said that he was impressed with Mogo's ability to avoid overreacting, that it could not be provoked like a human once it was ahead. Mr Kim also said that from his perspective his opponent in the last 2 games felt completely different than in the first 2 games. The difference, of course, was the additional search time. In the 2nd game mogo played the first half thinking it had 10 minutes, even though the KGS clock was set to 15, and mid-game the operator realized the mistake, took it offline and fixed the clock before reconnecting. But it was too late, so Mr Kim, other than being confused by the opponent abandoning and reappearing, did not get much chance to see the difference in play. Cheers, David ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] Re: mogo beats pro!
No offense at all taken by your words. I only meant to say that I have had personal experience with how reporters and journalists turn what they hear into what they write. It is my opinion that we could try very hard to fix our words and they will either change them back or make up something even more dramatic. Cheers, David On 11, Aug 2008, at 7:42 PM, Hideki Kato wrote: David, I didn't intend to offend any person in this list, sorry for short of my words. I'm just trying to prevent people misunderstand the truth. Hideki David Doshay: [EMAIL PROTECTED]: It is of no consequence what words WE use to describe this. Journalists will ALWAYS print it that way. If you use too many big words or ideas that are accurate but convoluted, you will either not get the publicity or the journalist will make up something even more absurd. Sorry if I am a bit over sensitive ... getting misquoted, my work ignored, and getting credit for the work of others in this past week has me very aware of how these people work. They are on a deadline and meeting the deadline with a headline that captures a lay reader's attention is the only priority. I know how my attempts to get a correction were greeted ... Cheers, David On 11, Aug 2008, at 8:37 AM, Hideki Kato wrote: Hi all, I'd like to say first Congratulations! to MoGo team. I have a question. Why do you all call the game as human vs. computer? It's obviously a match between Kim 8p and MoGo, a program developped by MoGo team, running on a supercomputer. As both MoGo and the supercomputer were developped by human, the game is clearly (a special type of) human vs. human. I'm afraid it may raise unnecessary emotional thoughts of against computers among people. It might be better to call such a game something of a style a professinal Goplayer vs. a program with its developper(s) to emphasize the program was created by human. -Hideki ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/ -- [EMAIL PROTECTED] (Kato) ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/ ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] Re: What's happening at the European Go Congress?
Also, if you are down 8 or 9 stones, maximizing your winning chances is still the right strategy, right? With MCTS algorithms the error margin is high at the start of the game, and low in the endgame. In a handicap game against a stronger opponent the assumption is that the weaker player will make more mistakes (i.e. has a higher error margin overall). But MCTS programs don't see it that way - their opponent model is the same strength as they are. So they choose a move that gives them 95% (+/- 20%) win (against themselves) instead of the better move that they only gives them a 90% (+/- 20%) win (against themselves). (I.e. I'm saying their error margin in the opening is much greater than the difference in their estimate of move values.) Darren -- Darren Cook, Software Researcher/Developer http://dcook.org/mlsn/ (English-Japanese-German-Chinese-Arabic open source dictionary/semantic network) http://dcook.org/work/ (About me and my work) http://darrendev.blogspot.com/ (blog on php, flash, i18n, linux, ...) ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] Re: mogo beats pro!
It does not change the fact MoGo was developped by the programmers. And the fact the programmers spent many resources, like the people fighting at Beijing right now, to develop MoGo. And Kim was developed by his parents, his go teachers, go books, and each opponent he has played against and learnt something from. But when the game starts each is on his own, and you are just left with a human with knowledge, and a machine with knowledge. I'm afraid it may raise unnecessary emotional thoughts of against computers among people. People like that will get emotional whichever words you use. Don't you think it cannot be changed or, at least, improved? Yes, you should have it play across a board using this android to play the moves: http://news.bbc.co.uk/1/hi/sci/tech/4714135.stm It should smile sweetly, and flutter its eyelids across the board every now and again. (In case anyone says I forgot the smiley, I'm being serious: people will mind losing less to a pretty machine than losing to a cube of metal; just as an elderly man would rather the above android helps them to the toilet than something that looks R2-D2.) Darren -- Darren Cook, Software Researcher/Developer http://dcook.org/mlsn/ (English-Japanese-German-Chinese-Arabic open source dictionary/semantic network) http://dcook.org/work/ (About me and my work) http://darrendev.blogspot.com/ (blog on php, flash, i18n, linux, ...) ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] Re: What's happening at the European Go Congress?
On Tue, 2008-08-12 at 11:50 +0900, Darren Cook wrote: Also, if you are down 8 or 9 stones, maximizing your winning chances is still the right strategy, right? With MCTS algorithms the error margin is high at the start of the game, and low in the endgame. In a handicap game against a stronger opponent the assumption is that the weaker player will make more mistakes (i.e. has a higher error margin overall). But MCTS programs don't see it that way - their opponent model is the same strength as they are. So they choose a move that gives them 95% (+/- 20%) win (against themselves) instead of the better move that they only gives them a 90% (+/- 20%) win (against themselves). (I.e. I'm saying their error margin in the opening is much greater than the difference in their estimate of move values.) There could be something to that. Do you believe that they will play the 90% move if they are told they are not really down 9 stones? I did a bunch of experiments and ALWAYS got a reduced wins when I faked the komi. But there are a million ways to do this and I may not have stumbled on the right way. - Don Darren ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
RE: [computer-go] Re: mogo beats pro!
Sorry, but I can't let this statement go past. The go programs in the 90s did local search, but not much global search. For example Many Faces did a one ply global search, with a variable depth quiescence search. I added an alpha-beta search to Many Faces last year, and it made a huge improvement in strength. So it is not true that alpha-beta pruning hit a roadblock. For me, the big advantage of UCT/MC is that it eliminates the huge, slow, buggy evaluation function. Simple playouts are much much easier to make bug free. Bugs in the evaluation function caused many losses, and are almost impossible to eliminate in traditional programs, since the evaluation functions are so complex. David It seems that alpha/beta pruning hit a roadblock a long time ago in go. Now we have MC... which as you increase the number of samples.. you start to get closer to an equivalent alpha/beta. But... there are still huge groups of samples that are not checked... and if you want to somehow prove you have the best move... how will you do it? Will you make the sample size equivalent to the number of possible samples? How will you do this practically? You can only state with a certain confidence that you did make the best move and this would be bound by our computational resources. ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] Re: What's happening at the European Go Congress?
-Original Message- From: Don Dailey [EMAIL PROTECTED] To: computer-go computer-go@computer-go.org Sent: Mon, 11 Aug 2008 11:09 pm Subject: Re: [computer-go] Re: What's happening at the European Go Congress? On Tue, 2008-08-12 at 11:50 +0900, Darren Cook wrote: Also, if you are down 8 or 9 stones, maximizing your winning chances is still the right strategy, right? With MCTS algorithms the error margin is high at the start of the game, and low in the endgame. In a handicap game against a stronger opponent the assumption is that the weaker player will make more mistakes (i.e. has a higher error margin overall). But MCTS programs don't see it that way - their opponent model is the same strength as they are. So they choose a move that gives them 95% (+/- 20%) win (against themselves) instead of the better move that they only gives them a 90% (+/- 20%) win (against themselves). (I.e. I'm saying their error margin in the opening is much greater than the difference in their estimate of move values.) There could be something to that. Do you believe that they will play the 90% move if they are told they are not really down 9 stones? I did a bunch of experiments and ALWAYS got a reduced wins when I faked the komi. But there are a million ways to do this and I may not have stumbled on the right way. If my engine plays in a high handicap game (and it has to be a pretty high handicap), for the first moves, it can't see a difference for any moves and plays randomly. I can fix this by making the playout asymetrical. I make the playout moves for black lighter (higher probability of being random). With this adjustment, it makes reasonable looking moves. I haven't tested this extensively because I don't have any need for an engine that plays better in high handicap games. - Dave Hillis ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
RE: [computer-go] Re: mogo beats pro!
On Mon, 2008-08-11 at 20:10 -0700, David Fotland wrote: Sorry, but I can’t let this statement go past. The go programs in the 90s did local search, but not much global search. For example Many Faces did a one ply global search, with a variable depth quiescence search. I added an alpha-beta search to Many Faces last year, and it made a huge improvement in strength. So it is not true that alpha-beta pruning hit a roadblock. I never doubted alpha-beta but when you say alpha-beta and GO in the same sentence, people automatically believe the program is going from 99% evaluation to 1% evaluation and 99% stupid. In fact you are still spending most of your time evaluating positions. I'm still not convinced that you can't make a strong alpha beta GO program if you have some imagination. It cannot just be a converted chess program, it has to be different, but still alpha beta at heart. It would have to be extremely selective. - Don For me, the big advantage of UCT/MC is that it eliminates the huge, slow, buggy evaluation function. Simple playouts are much much easier to make bug free. Bugs in the evaluation function caused many losses, and are almost impossible to eliminate in traditional programs, since the evaluation functions are so complex. David It seems that alpha/beta pruning hit a roadblock a long time ago in go. Now we have MC... which as you increase the number of samples.. you start to get closer to an equivalent alpha/beta. But... there are still huge groups of samples that are not checked... and if you want to somehow prove you have the best move... how will you do it? Will you make the sample size equivalent to the number of possible samples? How will you do this practically? You can only state with a certain confidence that you did make the best move and this would be bound by our computational resources. ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/ ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
RE: [computer-go] Re: mogo beats pro!
Yes, my alpha-beta searcher still has the big slow evaluation function (about 50 to 100 evaluations a second). When I get some free computer time I'll put it on the 19x19 server. I think it will be much closer to the 1 cpu uct many faces than to the older version 11 many faces. Uct also has the advantage that it is much easier to use with multiple CPUs. I know parallel alpha-beta exists, but my evaluation function is not designed to be thread safe. If I put a big lock around it, there will be almost no SMP scaling, since almost all the time is in the evaluation, not in the search. David -Original Message- From: [EMAIL PROTECTED] [mailto:computer-go- [EMAIL PROTECTED] On Behalf Of Don Dailey Sent: Monday, August 11, 2008 8:31 PM To: computer-go Subject: RE: [computer-go] Re: mogo beats pro! On Mon, 2008-08-11 at 20:10 -0700, David Fotland wrote: Sorry, but I can�t let this statement go past. The go programs in the 90s did local search, but not much global search. For example Many Faces did a one ply global search, with a variable depth quiescence search. I added an alpha-beta search to Many Faces last year, and it made a huge improvement in strength. So it is not true that alpha-beta pruning hit a roadblock. I never doubted alpha-beta but when you say alpha-beta and GO in the same sentence, people automatically believe the program is going from 99% evaluation to 1% evaluation and 99% stupid. In fact you are still spending most of your time evaluating positions. I'm still not convinced that you can't make a strong alpha beta GO program if you have some imagination. It cannot just be a converted chess program, it has to be different, but still alpha beta at heart. It would have to be extremely selective. - Don For me, the big advantage of UCT/MC is that it eliminates the huge, slow, buggy evaluation function. Simple playouts are much much easier to make bug free. Bugs in the evaluation function caused many losses, and are almost impossible to eliminate in traditional programs, since the evaluation functions are so complex. David It seems that alpha/beta pruning hit a roadblock a long time ago in go. Now we have MC... which as you increase the number of samples.. you start to get closer to an equivalent alpha/beta. But... there are still huge groups of samples that are not checked... and if you want to somehow prove you have the best move... how will you do it? Will you make the sample size equivalent to the number of possible samples? How will you do this practically? You can only state with a certain confidence that you did make the best move and this would be bound by our computational resources. ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/ ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/ ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] Re: What's happening at the European Go Congress?
Do you believe that they will play the 90% move if they are told they are not really down 9 stones? I just did a quick test of Mogo in that same position (black E5, white E3). (After switching off its opening book, which ironically instantly plays the same black 3 F4 move I just said was bad.) At komi 7.5 it starts off liking E4 at 100,000 playouts, then switches to F3 (the keima attach) at 260,000 playouts, and sticks with F3 until the end (1.3 million playouts) with 49% confidence at the end. (F3 is a good move.) At komi 3.5 it starts with F3 this time, then at 190,000 playouts switches to D3 (the symmetrical move!) and sticks with that (1.4 million playouts, 55% confidence). At komi 0.5 it choose C5 (the whole way, except for a period of preferring G5, the symmetrical move), 60% confidence. (C5 is also a strong move, but I'd personally prefer F3 or E7.) So, in that very small experiment, faking komi chooses different moves, but they are probably equally good. Darren -- Darren Cook, Software Researcher/Developer http://dcook.org/mlsn/ (English-Japanese-German-Chinese-Arabic open source dictionary/semantic network) http://dcook.org/work/ (About me and my work) http://darrendev.blogspot.com/ (blog on php, flash, i18n, linux, ...) ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
RE: [computer-go] Re: mogo beats pro!
On Mon, 2008-08-11 at 20:39 -0700, David Fotland wrote: Yes, my alpha-beta searcher still has the big slow evaluation function (about 50 to 100 evaluations a second). When I get some free computer time I'll put it on the 19x19 server. I think it will be much closer to the 1 cpu uct many faces than to the older version 11 many faces. Uct also has the advantage that it is much easier to use with multiple CPUs. I know parallel alpha-beta exists, but my evaluation function is not designed to be thread safe. If I put a big lock around it, there will be almost no SMP scaling, since almost all the time is in the evaluation, not in the search. This is not the case with alpha-beta. With additional processors, alpha-beta always does wasted work, and the more processors the more wasted work. It still always benefits from additional CPU's. - Don David -Original Message- From: [EMAIL PROTECTED] [mailto:computer-go- [EMAIL PROTECTED] On Behalf Of Don Dailey Sent: Monday, August 11, 2008 8:31 PM To: computer-go Subject: RE: [computer-go] Re: mogo beats pro! On Mon, 2008-08-11 at 20:10 -0700, David Fotland wrote: Sorry, but I can�t let this statement go past. The go programs in the 90s did local search, but not much global search. For example Many Faces did a one ply global search, with a variable depth quiescence search. I added an alpha-beta search to Many Faces last year, and it made a huge improvement in strength. So it is not true that alpha-beta pruning hit a roadblock. I never doubted alpha-beta but when you say alpha-beta and GO in the same sentence, people automatically believe the program is going from 99% evaluation to 1% evaluation and 99% stupid. In fact you are still spending most of your time evaluating positions. I'm still not convinced that you can't make a strong alpha beta GO program if you have some imagination. It cannot just be a converted chess program, it has to be different, but still alpha beta at heart. It would have to be extremely selective. - Don For me, the big advantage of UCT/MC is that it eliminates the huge, slow, buggy evaluation function. Simple playouts are much much easier to make bug free. Bugs in the evaluation function caused many losses, and are almost impossible to eliminate in traditional programs, since the evaluation functions are so complex. David It seems that alpha/beta pruning hit a roadblock a long time ago in go. Now we have MC... which as you increase the number of samples.. you start to get closer to an equivalent alpha/beta. But... there are still huge groups of samples that are not checked... and if you want to somehow prove you have the best move... how will you do it? Will you make the sample size equivalent to the number of possible samples? How will you do this practically? You can only state with a certain confidence that you did make the best move and this would be bound by our computational resources. ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/ ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/ ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] Re: What's happening at the European Go Congress?
Don Dailey: [EMAIL PROTECTED]: On Tue, 2008-08-12 at 11:50 +0900, Darren Cook wrote: Also, if you are down 8 or 9 stones, maximizing your winning chances is still the right strategy, right? With MCTS algorithms the error margin is high at the start of the game, and low in the endgame. In a handicap game against a stronger opponent the assumption is that the weaker player will make more mistakes (i.e. has a higher error margin overall). But MCTS programs don't see it that way - their opponent model is the same strength as they are. So they choose a move that gives them 95% (+/- 20%) win (against themselves) instead of the better move that they only gives them a 90% (+/- 20%) win (against themselves). (I.e. I'm saying their error margin in the opening is much greater than the difference in their estimate of move values.) There could be something to that. Do you believe that they will play the 90% move if they are told they are not really down 9 stones? I did a bunch of experiments and ALWAYS got a reduced wins when I faked the komi. But there are a million ways to do this and I may not have stumbled on the right way. Mr. Okasaki, a strong amatur, tested MoGo with a 9 stones handicap game at winning rate around 50% by adjusting komi on each move and reported it played clearly stronger than others, say, on KGS and the cluster version at Paris. Hideki -- [EMAIL PROTECTED] (Kato) ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] Re: mogo beats pro!
Darren Cook: [EMAIL PROTECTED]: It does not change the fact MoGo was developped by the programmers. And the fact the programmers spent many resources, like the people fighting at Beijing right now, to develop MoGo. And Kim was developed by his parents, his go teachers, go books, and each opponent he has played against and learnt something from. But when the game starts each is on his own, and you are just left with a human with knowledge, and a machine with knowledge. Mr. Kim was _created from scratch_ by his parents? :) I'm afraid it may raise unnecessary emotional thoughts of against computers among people. People like that will get emotional whichever words you use. Don't you think it cannot be changed or, at least, improved? Yes, you should have it play across a board using this android to play the moves: http://news.bbc.co.uk/1/hi/sci/tech/4714135.stm It should smile sweetly, and flutter its eyelids across the board every now and again. Yah, it's well known in Japan but I feel rather it's weird (some around me agree). (In case anyone says I forgot the smiley, I'm being serious: people will mind losing less to a pretty machine than losing to a cube of metal; just as an elderly man would rather the above android helps them to the toilet than something that looks R2-D2.) Interesting. Hideki -- [EMAIL PROTECTED] (Kato) ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] Re: What's happening at the European Go Congress?
Erik van der Werf wrote: You're right, my reply was sloppy (it seems I'm too much used to Japanese rules). Also I should have read GCP's email more carefully; I did not realize that his program, even with a large tree, would not be able to recognize the seki. I knew of course that the original Mogo playouts had this problem, but I thought all strong programs had solved it by now... No, far from it in fact. If anyone has found a clean solution that does not make the program worse in other LD situations, I'm all ears. -- GCP ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
