Large handicap games actually favor MoGo quite a bit since, from my experience playing the freely downloadable client, its specialty is fighting and making use of influence to attack.
This sounds very odd. One of the characteristics of MoGo is that it gives away territory nonchalantly when it feels confident of winning (because it only cares about win probability, not margin of victory). So I'd expect it to make poorer use of handicap stones than a human, who can think things like "he'll probably gain a couple of points on me in the endgame in ways I hadn't thought of".
But it does have a base of 3x3 patterns used to prime the pump for the MC search. These attempt to make the program prefer good shape. Probably those are the cause of the influence and tactical skill you see.
>Let us keep in mind that the program one by 1.5 points!
It gave away points freely in the endgame, as Monte-Carlo programs do: They care only about winning, not margin of victory. If all moves lead to a win, it's just as likely to play inside its own territory as take dame points.
At pro levels, one difference in rank is traditionally 1/3 of a stone, so no, he meant 2-3 amateur dan. The reason he was impressed is that this is extremely far above any other program.
That is American Checkers, or British Draughts. International Draughts is played on a 10x10 board, and has long kings. It's far from solved, although it is in much the same position as chess.
Actually, you're wrong too. There are many other complete information games humans play better than computers. I beat the current world champion Hex program recently, for instance. It took about a year of practice.
True, programmers haven't spent quite as much time on Hex as they have on Go (and not even remotely as much as they spent on Chess), but then again, humans have only a 60-year tradition for Hex, rather than the 1000-year one we have for Go. There's no reason to believe humans scale worse than computers.
I see a widespread misconception here, that 9 stone handicap is much against a 8 dan Korean pro. It's not. The "nerd" Sylvain Gelly has only been playing Go for a couple of years. I don't know his rank, but I would be quite surprised if he could beat his program without a handicap (it's ranked 2k at KGS)
Don Dailey, who runs the CGOS server, did a study to test MoGo's scalability. He put different versions of the program to fight in a tournament, each with a fixed number of playouts, and used Bayeselo to rank them. It turns out it scales very well: Every doubling of processing power led to a constant gain in ELO points, at least for the range of processing power he had access to.
You can't expect to get much better scalability than that. I doubt chess programs scale that well, for instance (of course, at the level they play at, it hardly matters).
And, judging by the computing resources required, it sounds like all they did here was brute-force it, so they really haven't "taught" the computer anything.
The program has little Go-specific knowledge as we know it, true. No openings, no joseki. Only a rudimentary idea of shape, as it can be encoded in 3x3 patterns, and some techniques for ladders and semeai. Every time it begins a new game, it figures out the same basic ideas that took humans no short time to develop. And likewise, it plays expert joseki as an emergent property.
You think just any old algorithm would do that, even with a supercomputer? Don't belittle the programmers' achievement, and don't underestimate the program (that's what Myungwan Kim did when he chose handicap to be 9. But to be fair, MC programs appear to be more stupid than they are to someone who doesn't know how they work.) Your "philosophy" may well fall in another ten years.
Why so defensive on behalf of humans? It's not as if the program wasn't made by humans. French ones, true, but humans nonetheless.
And it is a spectacular achievement. The supercomputer was the glazing on the cake; what this really shows is that the new approach they spearheaded (Monte-Carlo tree search with patterns) is scalable.
There are even simpler games that have the same stunning depth. Hex, by John Nash and Piet Hein, is my favourite. Dots and Boxes is also horribly intractable even on relatively small boards.
Perhaps it's for nerds, but it's hardly news in itself. But there are such things as MoGo's games against Catalin Taranu... which did not get a slashdot story. Hm, maybe you're right.
I play a little Go, but MoGo, Crazystone, and AyaMC are not at all like the old GNUgo of yore. These top Go programs are already good enough that it would take me years of serious practice to have a chance (they are hovering around 1k/1d on KGS) So I'm cheering for the computers, obviously:-)
You're right, game rules can not be copyrighted, only patented. I'm fairly sure Hasbro, Mattel etc. want to keep it like that. Thousands of people invent games every day, and accidental reinvention happens all the time. Apparently Hasbro has had trouble with people suing them for "stealing their idea" already.
If you ship them a game prototype they apparently send it back unopened with a note saying they can't risk looking at it for legal reasons.
Even if such crank lawsuits will fail, I bet they are a nuisance - and they would be much worse if a game actually could be copyrighted.
But I think they are perfectly happy with the situation as it is: Board games are basically lemons to most people. They have no way to judge their quality. Therefore, they buy based on whether they played a game with that name when they were young, or whether it's got Winnie the Pooh on the cover (if they're buying for their kids). The situation is perfect for them: they don't need to waste much money finding out what a truly good game looks like, they can just rebrand and resell for very good profit and very little risk.
Eventually, we may hope that people become more quality-conscious about boardgames... but first BoardGameGeek would have to change it's name, I think. And even then, in gamer forums there are always people who are either
1. Lamenting that they bought an awful game because they forgot to check the user comments at boardgamegeek first 2. Desperately looking for some OOP game, lousy by today's standards, but which they have a sentimental relationship to since they played it as children...
So Hasbro isn't going out of business anytime soon.
It's not property. Games can essentially not be copyrighted, only patented - there are rulings on this which are actually very generous to would-be game cloners. Explain the rules in your own way, draw your own board (even if it is functionally identical to the original board) and it's a different game.
What EA/Hasbro/Mattel/what-have-you have paid for is use of the brand. Trademark dilution is the only thing they have a chance with.
I made some great friends in elementary school, at a time I basically had no friends, and had just moved to a new school to get away from bullies.
I owe a lot to that man.
His game was the best, too. None of that emo self-loathing vampire bullshit, or pretentious "dark fantasy" that pretended to be something more than an entertaining game.
Eg er ikkje redd nokon mann på jorda, og stødt går eg trygg for mitt liv. Eg byggar mi makt på ein kniv i frå Mora, eit vådeleg beist av ein kniv. Skinande blank bak på buksa mi heng han, nyslipt og lett å nå, ferdig til bruk når i naud eg treng han, og det er nok ofte-tidt på.
Eg minnast eingong eg kom ut for ein tater, ein halvfull og brautande lurk, som oste av øl som eit rykande krater, og bante og svor som ein turk. Han treiv meg i strupen og skalla og klora, til stiv under greipet eg seig. Men straks eg slapp på han med kniven frå Mora, vart tateren mjuk som ein deig.
Og slik gjekk det presten vår og her um dagen, han tok meg i kristeleg tukt. Med augo som sylar han treiv meg i kraven, og vilde meg synda få bukt. Og orda han kom med var steinande sterke, men straks han vart kniven min var, vart presten so smørande blid som ein lerke, og sa eg var spøkje til kar.
Gud hjelpe den mann som med meinferd i sinnet, kjem kniven frå Mora for nær, for han skal bli rispa so suligt i skinnet, at merke for livstid han fær. Eg står her eg stend og tarv ikkje fira, um så eg er mindre til kar, for kniven er kvass og sitt laust i slira, og handa som treiv han er snar.
In a society were there is little divorce, there will still be need for divorce law, but it would not have to be so immensely comprehensive. A lot more can be settled on a case-by-case basis. It becomes safer (and more efficient) to leave things to arbiters such as judges, because they will have less influence in the big picture.
Religions don't die out entirely as long as there are good written records of them. Sometimes even that isn't needed... there are a bunch of people who try to revive "Åsatru" (the old Norse beliefs), despite most of our sources of them being their fairy-tale remnants, written down by inquisitive catholic priests like Snorre Sturlason. To would-be triumphalists out there: In the worst/best of cases, Christianity, Islam, Judaism, all will remain for thousands of years at the very least. Assuming historical continuity (of written records, that is).
Oh, German sentence structure is simple compared to many languages. It is a SVO language in general, it's only a certain class of sentences that have the verb-at-end oddity. It's no more odd than us Norwegians occasionally putting the verb first. ("Du sier det. Sier du det?")
> When adding 1/3 + 1/3 + 1/3, the computer needs to do 4 integer adds.
And a GCD + two divisions, although that can often be delayed. But you're right, spreadsheets should use exact arithmetic whenever possible. If my TI-89 could do it, why not them?
Slashdot Mirror
Large handicap games actually favor MoGo quite a bit since, from my experience playing the freely downloadable client, its specialty is fighting and making use of influence to attack.
This sounds very odd. One of the characteristics of MoGo is that it gives away territory nonchalantly when it feels confident of winning (because it only cares about win probability, not margin of victory). So I'd expect it to make poorer use of handicap stones than a human, who can think things like "he'll probably gain a couple of points on me in the endgame in ways I hadn't thought of".
But it does have a base of 3x3 patterns used to prime the pump for the MC search. These attempt to make the program prefer good shape. Probably those are the cause of the influence and tactical skill you see.
>Let us keep in mind that the program one by 1.5 points!
It gave away points freely in the endgame, as Monte-Carlo programs do: They care only about winning, not margin of victory. If all moves lead to a win, it's just as likely to play inside its own territory as take dame points.
At pro levels, one difference in rank is traditionally 1/3 of a stone, so no, he meant 2-3 amateur dan. The reason he was impressed is that this is extremely far above any other program.
Rocks, papers and scissors are all recognised ways of dealing with cockroaches, although scissors may be slightly nastier.
That is American Checkers, or British Draughts. International Draughts is played on a 10x10 board, and has long kings. It's far from solved, although it is in much the same position as chess.
Actually, you're wrong too. There are many other complete information games humans play better than computers. I beat the current world champion Hex program recently, for instance. It took about a year of practice.
True, programmers haven't spent quite as much time on Hex as they have on Go (and not even remotely as much as they spent on Chess), but then again, humans have only a 60-year tradition for Hex, rather than the 1000-year one we have for Go. There's no reason to believe humans scale worse than computers.
I see a widespread misconception here, that 9 stone handicap is much against a 8 dan Korean pro. It's not. The "nerd" Sylvain Gelly has only been playing Go for a couple of years. I don't know his rank, but I would be quite surprised if he could beat his program without a handicap (it's ranked 2k at KGS)
Don Dailey, who runs the CGOS server, did a study to test MoGo's scalability. He put different versions of the program to fight in a tournament, each with a fixed number of playouts, and used Bayeselo to rank them. It turns out it scales very well: Every doubling of processing power led to a constant gain in ELO points, at least for the range of processing power he had access to.
The study is here.
You can't expect to get much better scalability than that. I doubt chess programs scale that well, for instance (of course, at the level they play at, it hardly matters).
You have the prescription ready, eh? Why don't you, ah, go for it?
And, judging by the computing resources required, it sounds like all they did here was brute-force it, so they really haven't "taught" the computer anything.
The program has little Go-specific knowledge as we know it, true. No openings, no joseki. Only a rudimentary idea of shape, as it can be encoded in 3x3 patterns, and some techniques for ladders and semeai. Every time it begins a new game, it figures out the same basic ideas that took humans no short time to develop. And likewise, it plays expert joseki as an emergent property.
You think just any old algorithm would do that, even with a supercomputer? Don't belittle the programmers' achievement, and don't underestimate the program (that's what Myungwan Kim did when he chose handicap to be 9. But to be fair, MC programs appear to be more stupid than they are to someone who doesn't know how they work.) Your "philosophy" may well fall in another ten years.
Why so defensive on behalf of humans? It's not as if the program wasn't made by humans. French ones, true, but humans nonetheless.
And it is a spectacular achievement. The supercomputer was the glazing on the cake; what this really shows is that the new approach they spearheaded (Monte-Carlo tree search with patterns) is scalable.
There are even simpler games that have the same stunning depth. Hex, by John Nash and Piet Hein, is my favourite. Dots and Boxes is also horribly intractable even on relatively small boards.
Perhaps it's for nerds, but it's hardly news in itself. But there are such things as MoGo's games against Catalin Taranu ... which did not get a slashdot story. Hm, maybe you're right.
I play a little Go, but MoGo, Crazystone, and AyaMC are not at all like the old GNUgo of yore. These top Go programs are already good enough that it would take me years of serious practice to have a chance (they are hovering around 1k/1d on KGS) So I'm cheering for the computers, obviously :-)
Chess and barehand fighting are individual sports pared down to the minimally interesting essentials.
I dare say many Go players would disagree. And you should try Hex sometime.
I'd just say that "I have an appointment soon. Am I free to go?"
Ah, but what if you didn't have an appointment soon, and they found out that you lied? Ha!
You're right, game rules can not be copyrighted, only patented. I'm fairly sure Hasbro, Mattel etc. want to keep it like that. Thousands of people invent games every day, and accidental reinvention happens all the time. Apparently Hasbro has had trouble with people suing them for "stealing their idea" already.
If you ship them a game prototype they apparently send it back unopened with a note saying they can't risk looking at it for legal reasons.
Even if such crank lawsuits will fail, I bet they are a nuisance - and they would be much worse if a game actually could be copyrighted.
But I think they are perfectly happy with the situation as it is: Board games are basically lemons to most people. They have no way to judge their quality. Therefore, they buy based on whether they played a game with that name when they were young, or whether it's got Winnie the Pooh on the cover (if they're buying for their kids). The situation is perfect for them: they don't need to waste much money finding out what a truly good game looks like, they can just rebrand and resell for very good profit and very little risk.
Eventually, we may hope that people become more quality-conscious about boardgames... but first BoardGameGeek would have to change it's name, I think. And even then, in gamer forums there are always people who are either
1. Lamenting that they bought an awful game because they forgot to check the user comments at boardgamegeek first
2. Desperately looking for some OOP game, lousy by today's standards, but which they have a sentimental relationship to since they played it as children...
So Hasbro isn't going out of business anytime soon.
It's not property. Games can essentially not be copyrighted, only patented - there are rulings on this which are actually very generous to would-be game cloners. Explain the rules in your own way, draw your own board (even if it is functionally identical to the original board) and it's a different game.
What EA/Hasbro/Mattel/what-have-you have paid for is use of the brand. Trademark dilution is the only thing they have a chance with.
I made some great friends in elementary school, at a time I basically had no friends, and had just moved to a new school to get away from bullies.
I owe a lot to that man.
His game was the best, too. None of that emo self-loathing vampire bullshit, or pretentious "dark fantasy" that pretended to be something more than an entertaining game.
TOLLEKNIVEN
Eg er ikkje redd nokon mann på jorda,
og stødt går eg trygg for mitt liv.
Eg byggar mi makt på ein kniv i frå Mora,
eit vådeleg beist av ein kniv.
Skinande blank bak på buksa mi heng han,
nyslipt og lett å nå,
ferdig til bruk når i naud eg treng han,
og det er nok ofte-tidt på.
Eg minnast eingong eg kom ut for ein tater,
ein halvfull og brautande lurk,
som oste av øl som eit rykande krater,
og bante og svor som ein turk.
Han treiv meg i strupen og skalla og klora,
til stiv under greipet eg seig.
Men straks eg slapp på han med kniven frå Mora,
vart tateren mjuk som ein deig.
Og slik gjekk det presten vår og her um dagen,
han tok meg i kristeleg tukt.
Med augo som sylar han treiv meg i kraven,
og vilde meg synda få bukt.
Og orda han kom med var steinande sterke,
men straks han vart kniven min var,
vart presten so smørande blid som ein lerke,
og sa eg var spøkje til kar.
Gud hjelpe den mann som med meinferd i sinnet,
kjem kniven frå Mora for nær,
for han skal bli rispa so suligt i skinnet,
at merke for livstid han fær.
Eg står her eg stend og tarv ikkje fira,
um så eg er mindre til kar,
for kniven er kvass og sitt laust i slira,
og handa som treiv han er snar.
It's a matter of degree.
In a society were there is little divorce, there will still be need for divorce law, but it would not have to be so immensely comprehensive. A lot more can be settled on a case-by-case basis. It becomes safer (and more efficient) to leave things to arbiters such as judges, because they will have less influence in the big picture.
Religions don't die out entirely as long as there are good written records of them. Sometimes even that isn't needed... there are a bunch of people who try to revive "Åsatru" (the old Norse beliefs), despite most of our sources of them being their fairy-tale remnants, written down by inquisitive catholic priests like Snorre Sturlason.
To would-be triumphalists out there: In the worst/best of cases, Christianity, Islam, Judaism, all will remain for thousands of years at the very least. Assuming historical continuity (of written records, that is).
Oh, German sentence structure is simple compared to many languages. It is a SVO language in general, it's only a certain class of sentences that have the verb-at-end oddity. It's no more odd than us Norwegians occasionally putting the verb first. ("Du sier det. Sier du det?")
currently $17.85 dollars, you mean. Wait, $128.20...
> So in other words Excel is using some form of floating or fixed point BCD?
It should. Your databases does, and your pocket calculator probably does as well.
> Maybe somebody can come up with a better algorithm but my first thought is to prime factorize both numbers.
:-)
There's an "Euclid" here who says he has come up with a better algorithm
> When adding 1/3 + 1/3 + 1/3, the computer needs to do 4 integer adds.
And a GCD + two divisions, although that can often be delayed. But you're right, spreadsheets should use exact arithmetic whenever possible. If my TI-89 could do it, why not them?