Archived material from Doug's go blog.
Spalding Gray is gone, and the David Kelly affair still resonates. Perhaps there's something in the air.
A long time ago, when I was in high school, I believed in the legalization of assisted suicide. After all, if it doesn't harm anyone else, why restrict the choices that people make about themselves? It went along with a simplistic libertarian, An it harm none, do what you will view of the universe.
It's not a valid argument, obviously, because there are people harmed other than the individual. I was once told that suicide was the most vicious, vile, and (in particular) selfish thing one could possibly do. It kind of shocked me at the time, because it came pretty much out of the blue, totally disconnected from the conversation. (In retrospect, it ought to have been a clue about the speaker's state of mind.) Regardless, the statement is valid.
A physiologically normal response to stress, whether due to a bad haircut, loss of a spouse or child, a terminal illness, or a particularly nasty grilling by a parliamentary subcommittee, is to grieve for some period of time, encase the emotional scar somewhere inside the psyche, and somehow recover the unrealistically optimistic worldview required to function in society. Offing yourself is defined as abnormal. Abnormal not because it's unheard of: suicide is clearly documented back into classical times and earlier, but by societal convention. You could argue that this because the people doing the defining -- i.e., the physicians -- have a hard time arguing that they did their best to help their patient when their patient is dead. The no-suicide view is essentially that even in extremum, efforts are better put into addressing the patients' quality of life concerns than in hastening the end life altogether. People who are normal and well routinely and seriously underestimate the quality of life of those who are sick or abnormal. Depression, even in the terminally ill, can and should be considered a treatable disease.
The urge to kill oneself is biochemical in origin, and is becoming better understood. Ironically, physicians seem more prone than the rest of the population, particularly female physicians and psychiatrists. Suicide is the leading cause of death for physcians under forty. Antidepressant drugs are a mainstay of treatment, but paradoxically may increase the risk of death. This may be due to providing a convenient means through overdose, or perhaps by lifting the deep clouds of depression which prevented any action, including attempts or completion.
In go, suicide is really a misnomer. The stones are not conscious intentional actors, deciding where to land of their own free will, but rather pieces placed by some godlike player sitting above the fray. It's much more akin to intentional murder, or sending out soldiers on a mission from which they likely (or certainly!) will not return. Still, even sacrificing stones is not controversial, and seems to carry no moral overtones. The only activity prohibited is self-capture, where stones are taken off the board by one of their own. Friendly fire, if you will.
In terms of tactics, the principal change allowing suicide makes is the creation of extra ko threats. Eg., in oshitsubushi, legalizing suicide creates a ko threat. There are other ko threats made available as well.
There are also obscure situations where suicide does more. One example occurs in a capturing race, and relies on the fact that big eyes can be small in the corner. Other rare consequences may exist. Of the adopted rulesets, only New Zealand Rules allows self-capture. In the rest, for whatever reason, suicide is prohibited. But we allow resigning. Weird.
I finished the game I plotted on 15 Jan. It just got worse and worse, and I resigned at move 176. Next time I play, I will be more humble, even against the computer. Next time I play, I will not look at the computer's estimate of the score: not only is it cheating, it's dispiriting. Next time I play, I will take black.
This might actually be some kind of milestone for computer go. I have an author credit on GnuGo, due to some trivial contribution I made years ago. And the current version of GnuGo can not only give me a good game, but might well be better than me. Are there any other go programs which can defeat even one of their authors?
The discussion of the demerits of superko seems to have wandered into a general discussion of rules. I think Robert's focus on easing the resolution of disputes that can or do occur in tournament play is a good one. I also think it'd be easy to achieve, if that were the sole goal. Consider no pass with prisoner return, either simple or super ko. There are no resolution phases or agreements to mediate. Games end in very limited ways: resignation, illegal move, or failure to move (time exhaustion). (If simple ko, then add draw by repetition of position.) All of these are easy to verify, and difficult to dispute. Most games will end by resignation, when the loser accepts that they are behind (perhaps after counting after the end of positive play). Once you've resigned, it's hard to dispute the result. If they stick it out, they will eventually have to either make an illegal move or fail to move at all and have their flag fall. Disputes wouldn't be impossible, but they would be more difficult.
I think I can guess the first dispute, though, were this implemented: "I didn't know I'd need to leave enough time on the clock to fill territory!"
It's easy to get GnuGo to estimate the score for you, and plot a graph of how that changes during the game. The oscillation (the size of which is related to the temperature) can obscure the trend, so you have to pay attention to whose move it is. This plot also shows odd (black to move, in black) and even (white to move, in red) numbered movesby themselves. GnuGo ought to be able to tell you the temperature by switching the side to play. This is shown by the error bars, but most of the time, --score estimate ignores whose move it is.
In this game, White does okay for awhile, then things go south. Here's the first significant uptick for Black, starting with move 16. It's not obvious to me what's so wrong with , except for maybe being on the wrong side, but maybe that's why I'm a DDK. Still, as far as GnuGo is concerned, White might have well passed as played - . They aren't that bad, are they?
I googled GnuGo, and the #2 result looks vaguely familiar. It's quite odd (in a good way) to see one's own work reappear, transmogrified into another language, complete with silly chatter. One of these days I'll fix it for the latest version.
In other news, 3.4 had its revenge on me, even on 9x9.
One lens for looking at the three group capturing race is as two simultaneous two group races which happen to share a common group. It's not right, of course, because it conflates outside liberties and shared liberties, but it may be a useful approximation, because two group races are reasonably understood.
It seems like it should be obvious, somehow, that if White (the one in the middle) can win either of the two races, examined in isolation, then she wins everything. Once one of the side groups is captured, she's alive, and can pick up the other one whenever. And going about capturing a side group can't hurt her, if she wins that race in isolation. So the only interesting cases are the ones where White loses both races considered on their own: everything else is a complete White victory.
Played GnuGo, gave it four stones, and even though I was certain that I was being crushed the entire game, it ended up B+3. I really must learn to count someday. For revenge I played W on 9x9 and stomped, despite being surprised and outread multiple times.
I still can't figure out if these guys are for real. "Chess boxing" also titles a series of obscure kung fu movies.
In the three group capturing race --, can the group have any outside liberties? The diagram shows an (admittedly unnatural looking) example. Outside stones are alive, so the group is neighbored by two black groups who are participating in the capturing race, and one who is not. The shared liberty with the live black group, marked with a circle, is an outside liberty. Black can move there for seki, or White will fill a shared liberty and win everything.
In constructing such positions, one way is to imagine the entire board as filled with stones and sketch the lines dividing Black from White. These lines can terminate only at the edges of the board, and can cross only in a four-way junction -- a crosscut. It's kind of like a two color map in the map coloring problem. A couple moments of sketching should convince you that you need at least two crosscuts to give White an outside liberty if you use the edge of the board, or a couple more to build such a position in the middle of the board.
Happy Christmas, Merry Solstice, and all the rest. Holidays mean travel, which for me meant a plane flight and a bit of time to work on three group capturing races. Despite my best wishes and many delays, no one has bothered to scoop me on a three-group formula, which would really have saved me some work. (Or, if they have, they haven't bothered to tell me, either.) So I guess if I want it to get done, I have to do it myself. In pursuit of that end, I'll transcribe my notes from the airline napkins they were taken on.
So here's the setup: --, S1,2 = Black's shared liberties, S = White's, y1,2 = Black eye sizes, y = White's, and O, O1,2 are outside liberties. Eye liberties we'll call Y = L(y)-ybar, where ybar is the number of opponent stones in the eye. Possible outcomes are (from best to worst for Black) White dies, seki or exchange, Black dies. The state is given by (O1 y1/y1bar S1 O y/ybar S2 O2 y2/y2bar), an 11-tuplet. Black moves are: decrement O, decrement Si, increment ybar, and decrement yi and set yibar=0, with various conditions attached. White moves are similar: decrement Oi, decrement Si, increment yibar, and decrement y and set ybar=0. As can be seen, there's some complexity possible. One way to go forward is to just start collecting special cases, and look for a general answer later.
In this notation, the exchange sequence worked out before starts from (4/3 1 1 1 4/3) (leaving out the O's, which are all zero). The Black groups are identical and have White beat on exclusive libs, 2 to 1. White attacks by removing a shared liberty, (4/3 1 1 0 4/3), 4/3 1 1 0 3/0, 4/3 1 1 0 3/1, 3/0 1 1 0 3/1. If White 3/1 1 1 0 3/1, then 3/1 0 1 0 3/1 and White is in atari and loses everything. Better is 3/0 1 1 0 3/2, letting Black live the group and capturing the other.
A similar situation is 5/4 1 4/3 1 5/4, where White can 5/4 1 4/3 0 5/4, 5/4 1 4/3 0 4/0, 5/4 1 4/3 0 4/1, 4/0 1 4/3 0 4/1; and if W 4/1 1 4/3 0 4/1, then 4/1 0 4/3 0 4/1, 4/1 0 3/0 0 4/1, 4/1 0 3/1 0 4/1, and White has the move but is behind two liberties to four for either Black group.
A few observations at this point: eye size and not just liberties matters (also true of two group races), it's not always the same side filling shared liberties (not true of two group races, although in this case White has made a mistake earlier), and making a placement to kill a big eye is not a no-brainer.
Clearly the circled point is a liberty for White and not for Black. (It is also coincidentally where Black needs to play to win the semeai.) So you need a 12-tuplet, really.
(Hikaru79 interrupted me in the middle of trying to get my diagram just right. :) This one should be correct. White at the circled point makes a temporary seki; Black there wins the semeai.)
Bill: What to call the circled point is an interesting question. Clearly, it is a liberty for White. But since White's play there kills Black, it would seem to be a liberty for Black, as well.
DougRidgway: White can have outside liberties -- they are counted as O in the 11-tuplet above -- but the circled point in the diagram isn't an example. If you fill one of your own outside liberties, nothing changes except that your own liberties decrease. If White plays the circled point, White's liberties remain the same, and shared liberties with Black increase. The diagram actually doesn't fit into the model, although it does after White plays the circled point.
To get outside liberties for White, add a third Black neighbor group, known alive, and share liberties with it.
Long time, no updates. In a sense, this is natural. Perhaps the story arc of this blog is coming to an end. Still, there are some loose ends I'd like to tie up before I call it quits. I don't have the energy for that today, so instead I'll just give a link to yet another ruleset:
This one is apparently written by an equipment seller. Why people think it's a good idea to rewrite their own version of the rules, rather than borrow an existing, tested ruleset, perhaps coupled to a carefully thought through introduction to the game, I don't know. If you're aiming at existing go players, you don't need to include rules at all. And complete novices can be completely and permanently baffled by badly written rules. There seems to be no benefit, so why do it?
Still, perhaps they can be appreciated in a kind of folk art sense, as a spontaneous outpouring of creative energy. These rules certainly have some interesting features. I particularly like the compensation for first move: play two games, colors reversed, and add the scores from the games. Resigning is impossible (how many points is a resignation worth?), and it's important to keep fighting in the first game, no matter what the score: loss by thirty or loss by forty determines how much you'll have to make up in the second game, and could be decisive. No slacking just because you're ahead, no risky speculation just because you're behind, every point counts. It's kind of like bangneki.
Robert Jasiek posted a screed on r.g.g. about people editing his work on this wiki, which inspired me to read the history of Chinese Counting page. And hey, there in version 21, is a description of my current favorite theoretical ruleset: no pass with prisoner return, which is equivalent to stone scoring. I've written about this before, but I wasn't sure where I heard it first. Now I think I know.
There was a r.g.g. thread on teaching go to small children recently. My kids have some exposure to go, and a friend recently asked me about finding (or maybe even starting) a go club for kids, so it's a subject I'm interested in. I won't pretend to give advice, but perhaps I can identify certain developmental stages.
They don't necessarily occur in any order, and they can't necessarily be taught. Some are hard even for adults. The obstacles aren't necessarily where I would've expected them to be, either: saving stones seems like the flip side of capturing, but attacking might be direct and easy, while defending is indirect and difficult. Acceptance of the game as the child wants to play it has been the key to my enjoyment.
Capture go uses the same rules of play as go, but with a different end condition. Instead of scoring territory after passing, the winner is the first to capture anything. It's sometimes used to teach the capturing rule and give practice in some basic tactics.
I've played the game a bit against a Palm program, which starts with a crosscut in the middle of the board. A lot of tactics are, in fact, the same: cutting, connecting, surrounding, ladders and nets, competing for liberties and free space. Certainly, some tactics are different, obviously: no snapbacks, squeezes, or throw-ins. But until recently, I'd thought life and death remained the same: capturing is capturing, two eyes make life, right?
Life and death is different, because many of the standard tactics for capturing involve sacrifice. A group with only one single point eye can be captured, and false eyes are still false, because the ataried string must be connected. But big eyes cannot be reduced to small eyes, so a 2 point big eye, like in the diagram, in fact counts as two eyes and is alive. A three point big eye can't be killed either, but by playing on the vital point it can be turned into seki. Throw away the Davies book! This has a big impact on what invasions work, and rescuing disconnected stones: all you need is one big eye. Of course, you still have to save every single stone...
Grauniad: If the rules prohibit passing and declare a player without a legal move (suicide is illegal) the loser, then surrounding more territory than your opponent also becomes important, so late in the game you can play safely in your own territory rather than suicidally in your opponent's territory. This makes it a little more like real Go.
DougRidgway: Initial conditions matter: with a crosscut, I've never seen it get that far.