“the machinery that was built up for computer chess is pretty useless"

May 26, 2014 5:31 PM

The Mystery of Go, the Ancient Game That Computers Still Can’t Win

The challenge is daunting. In 1994, machines took the checkers crown, when a program called Chinook beat the top human. Then, three years later, they topped the chess world, IBM’s Deep Blue supercomputer besting world champion Garry Kasparov. Now, computers match or surpass top humans in a wide variety of games: Othello, Scrabble, backgammon, poker, even Jeopardy. But not Go. It’s the one classic game where wetware still dominates hardware.

If you're actually interested in "the Eastern version of chess", you might try chaturanga, shogi, or xiangqi.

The WIRED piece was lined by The American Go Association, which also links the other four things your brain does better than a computer, and a piece in the New Yorker, “New Yorker” Reports on Computer Go:
The Electronic Holy War

Last March, sixteen years later, a computer program named Crazy Stone defeated Yoshio Ishida, a professional Go player and a five-time Japanese champion. The match took place during the first annual Densei-sen, or “electronic holy war,” tournament, in Tokyo, where the best Go programs in the world play against one of the best humans. Ishida, who earned the nickname “the Computer” in the nineteen-seventies because of his exact and calculated playing style, described Crazy Stone as “genius.”

Both the WIRED and New Yorker pieces cover the Densei-sen Competition (Japanese), the first of which was in March of 2013, in which 'Crazy Stone' won with a handicap of four stones, and the second was March of 2014.

All Systems Go, The Sciences, Volume 38, Number 1, 1998, David A. Mechner.

This past July, the annual conference of the American Association for Artificial Intelligence was marked by an oddly heroic event. Throughout the conference, held in Providence, Rhode Island, world-class players of chess, checkers and other games pitted their skills against state-of-the-art computer adversaries. For the most part the humans either lost or barely eked out victories. Then a twenty-seven-year-old native of New Mexico named Janice Kim challenged a computer to the oldest, most complex game of all: go.
When a video image of the match flickered onto one of the oversize monitors in the "Hall of Champions," audience members laughed and shook their heads in disbelief: the computer was being granted a twenty-five-stone handicap. Kim is an exceptional player--the only female go professional the West has ever produced--but this seemed ridiculous: the first stone she placed looked like a lone paratrooper dropped into an enemy garrison.

The Challenge of Go as a Domain for AI Research: A Comparison Between Go and Chess (PDF)

There are more links at AITopics.

posted by the man of twists and turns (72 comments total)

As somebody who has been trying off and on to learn Go for a decade, I can guarantee you without reservation that a computer can, indeed, do Go better than my particular brain.

A TRS-80 would be better at Go than my brain.

Most pocket calculators would be better at Go than my brain.

I have never been more frustrated, and more repeatedly entranced, by any other game in the world. I've had SmartGo Kifu installed on my iPad since Day One, and I still can't beat the computer on the nine-square board. I feel like my brain is wired entirely wrong for this game, but every six months or so I get the fever again and take another run at it for a week or two.
posted by Shepherd at 5:39 PM on May 26, 2014 [12 favorites]

I've had SmartGo Kifu installed on my iPad since Day One, and I still can't beat the computer on the nine-square board.

I've been trying this lately. My current strategy is to - somehow - win a game and go up a (half-)level, and then, on the next level.. don't let the computer win. When it looks like I'm going to lose, I just hit "undo" a couple dozen times until we're back at the start. I know that playing 9x9 is supposed to be good learning, but I just want to play on the full dang board. -.- (and I still have to go through 13x13 after this)

I've been stuck at 8+ for.. a month now?
posted by curious nu at 5:41 PM on May 26, 2014 [1 favorite]

It was interesting following the final 10 or so years when chess games between grandmasters and computers were interesting. As late as 2000, the GMs could still get in a crushing win like this, basically by building up an attack slowly, so that the program does not see threats until it's too late. Those days are long gone.
posted by thelonius at 5:45 PM on May 26, 2014 [1 favorite]

I would like to point out that all computers are equally good at Go, it's just a question of how long it will take to make a move, or if it will run out of memory.
posted by jeffamaphone at 5:56 PM on May 26, 2014 [11 favorites]

It was interesting following the final 10 or so years when chess games between grandmasters and computers were interesting. As late as 2000, the GMs could still get in a crushing win like this, basically by building up an attack slowly, so that the program does not see threats until it's too late. Those days are long gone.

Slight derail, but ... I don't know what I'm looking at, here. As of end of game, Black is not in check and seems to have several options?
posted by kafziel at 5:59 PM on May 26, 2014

The Junior team would have resigned at that point, because White's threats are too numerous. Black has to move the dark-squared Bishop, and he has to defend against the White rook moving over to attack both the Black Knight and the other Bishop, and White is also threatening to take the Black pawn on e6. In addition, Black is down material - he has a Bishop for a Rook.

All that means that they are totally lost against a good player. Someone like Kramnik needs much, much less to win a game.
posted by thelonius at 6:03 PM on May 26, 2014 [2 favorites]

Advice for beginners: lose your first 100 games quickly.
posted by rustcrumb at 6:25 PM on May 26, 2014 [5 favorites]

I know that playing 9x9 is supposed to be good learning, but I just want to play on the full dang board. -.- (and I still have to go through 13x13 after this)

I think the 9x9 is best for learning local tactics, aka 'fighting'. It's not like you have to master 9x9 before you move on to the larger boards. You may actually do better on the larger boards; with more open space I think it's harder for the computer to calculate all the possibilities.
posted by isthmus at 6:25 PM on May 26, 2014 [2 favorites]

I've read a bit about this and I understand that a computer can predict and process more moves. Watson clicking first on Jeopardy seemed cheating and Deep Blue sending Kasparov into a fit seemed to make sense - and was kind of fun.

Humans consistently beating computers at go seems to indicate something. I have no idea what that is and I hope the computers don't either.
posted by vapidave at 6:30 PM on May 26, 2014

Don't worry. Computers may be good at faking it, but they don't have any ideas yet.
posted by saulgoodman at 6:36 PM on May 26, 2014 [2 favorites]

luv this post (and chess, and go). the computer that beat the japanese champion had a four stone advantage, which is substantial.

i learned go in a chess club when i was ~10. i have never lost a face-to-face game against a non-asian in my life. i've had my ass kicked on the internet a few times, i presume they were asians.
posted by bruce at 6:41 PM on May 26, 2014

It's not like you have to master 9x9 before you move on to the larger boards.

You do with SmartGo Kifu. =/ At least for playing against the AI.. which, having just checked, actually only has options for 9x9, 11x11, and 13x13. Maybe I should just start playing online again.
posted by curious nu at 6:46 PM on May 26, 2014

i have never lost a face-to-face game against a non-asian in my life. i've had my ass kicked on the internet a few times, i presume they were asians.

I am terrible at go but I've been playing it since I was a kid. My grandfather, if you believe the family lore (and I'm not sure that I do) was the first Caucasian go master in the US but I'm not sure if it counts since one of his parents was Uzbek. Thanks for the links.
posted by jessamyn at 6:50 PM on May 26, 2014 [6 favorites]

Many articles about Go, and computer Go in particular, get a lot of things wrong, but when I read the Wired article I was impressed by its accuracy. It's a good introduction to the current state of computer Go.
posted by dfan at 7:03 PM on May 26, 2014 [1 favorite]

jessamyn, a go master is "dan", an amateur is "kyu". i am definitely kyu, but i have a relatively lower number (higher rank) in the kyu. i would be happy to place at risk my unbeaten status against white people in a go game against you. if one of your great-grandparents was uzbek, that just might be the extra genetic oomph you need to silence (for a few hours) the perennially annoying "bruce".
posted by bruce at 7:05 PM on May 26, 2014

I dread the day when a computer becomes proficient at "Cards Against Humanity". THAT will be when Skynet becomes real.
posted by oneswellfoop at 7:28 PM on May 26, 2014 [19 favorites]

I've been on a traditional games kick lately - Bobby Fischer's chess book was eye opening, amazingly fun, and I've always enjoyed reading Poker strategy books. I probably enjoy reading about games as much as playing them, and I'm having a hoot going through some books on chess tactics - they set up a chess problem, and you try to figure out the right sequence of moves.

There apparently is a whole series of these books for go - books and books full of go problems. I'm in for an enjoyable summer reading list. There are also a few for backgammon, which I like because of the element of chance overlaying the strategy.

I kind of wonder if International Chess was "solved" by computers because the queen was made into an overly powerful piece at the beginning of the renaissance. Asian and ancient chess varieties usually have more equally matched pieces. The games develop more slowly. Also, I kind of wonder how an AI would fare at Sittuyin, where the players choose where to place the pieces, or Maruk, where you can play for a draw in an end-phase where you're materially disadvantaged.
posted by Slap*Happy at 7:29 PM on May 26, 2014

I have to wonder whether the failure to write a program that can beat a human master is ultimately going to be solved by processing power.
posted by MeanwhileBackAtTheRanch at 7:35 PM on May 26, 2014

Go has always seemed more organic and more elegant to me than chess. Most of the time I can do pretty well by playing intuitively, just putting the pieces where they feel like they want to go, to make the right kinds of shapes, which I've never been able to do on a chessboard.

I've never thought about playing go on my phone, and now maybe I will. I used to play obsessively on a free little windows 9x9 program which taught me the basics of how to push through someone's lines, etc. When it comes to humans most people I've met are still just learning the basics of go so I beat them easily or they're already at some esoteric level of go-ness where I can just barely understand how they're beating me. I might download SmartGo Kifu tonight. So long productivity!
posted by no mind at 7:41 PM on May 26, 2014

I dread the day when a computer becomes proficient at "Cards Against Humanity". THAT will be when Skynet becomes real.

I bet you could have some really rudimentary natural language processing, along with a ranking of how crude and offensive a given card is, and do pretty well, because some people are terrible judges who don't appreciate any kind of effort or creativity.
posted by vogon_poet at 7:43 PM on May 26, 2014 [4 favorites]

I have to wonder whether the failure to write a program that can beat a human master is ultimately going to be solved by processing power.

More processing power certainly doesn't hurt, but the game is so qualitatively huge that you can't beat it just by throwing a thousand times more power at it (whereas having a thousand times more processing power while playing chess helps a lot). The Wired article discusses this a bit.
posted by dfan at 7:46 PM on May 26, 2014

I like playing Go but I routinely get my ass handed to me by toddlers and feral cats. A beautiful game, though.
posted by sandettie light vessel automatic at 7:47 PM on May 26, 2014 [5 favorites]

The difference between computer success in Go vs. Chess is, to me, a good demonstration of how limited the current state of AI is.

Way back in the '80s I was taking an undergraduate course in AI programming and a quote from the prof stuck in my head. "The current state of AI is a bunch of cute hacks," and his predictions for the future were not optimistic. Fast forward 30 years and a quote from an article on Douglas Hosfstader the Atlantic hits it pretty squarely (previously on MeFi):

Perhaps, as Russell and Norvig politely acknowledge in the last chapter of their textbook, in taking its practical turn, AI has become too much like the man who tries to get to the moon by climbing a tree: “One can report steady progress, all the way to the top of the tree.”

Sure the computers win at chess, but what is interesting is that the way they play is nothing like how a grand master does. They play in a way we are able to train them to play, that's all good fun, but it just serves to show how little we know about how humans play chess. What is inspiration? Love? Creativity?
posted by Walleye at 7:50 PM on May 26, 2014 [16 favorites]

I still remember the lesson about life that was taught to me at the NY Go center: Build from where you're strong.
posted by Obscure Reference at 7:59 PM on May 26, 2014 [4 favorites]

I would not recommend learning Go by playing against a computer. Find a human to teach you, play against people online (some are willing to offer teaching games with commentary), or read through the Go Teaching Ladder. But computers bring no real sense or artistry to the game, just mini-tactics.
posted by argybarg at 8:28 PM on May 26, 2014 [3 favorites]

Walleye: "Perhaps, as Russell and Norvig politely acknowledge in the last chapter of their textbook, in taking its practical turn, AI has become too much like the man who tries to get to the moon by climbing a tree: “One can report steady progress, all the way to the top of the tree.”"

I like to say Artificial Intelligence is the study of things we don't know how to make computers do yet. Once we figure it out a particular domain, it's no longer AI, just a set of algorithms with names like Depth First Search, or Min-Max with Beta Pruning, or Bayesian Networks.
posted by pwnguin at 8:37 PM on May 26, 2014 [19 favorites]

Human chess skills are pretty clearly based on pattern-matching, combined with training to let you do pattern-matching some number of levels forward. Computers aren't great at generalised pattern-matching, although they're awesome at specific instances of it.

Has anyone compared live chess strategy versus play-by-mail? I think good chess players aren't just reading the board, but also the other player: intuitively picking up on their opponent's strategy by observing their timing and body language. That would be a hard thing to teach a computer!
posted by Joe in Australia at 9:22 PM on May 26, 2014

I know that playing 9x9 is supposed to be good learning, but I just want to play on the full dang board.

Well, you would probably actually have a chance at beating smartgo kifu at 19x19, it is really really weak at the full board game (not that this is helpful, because at this board size it really doesn't play like a human either). In fact, it is basically strongest at 9x9. You can play any board size you want at any handicap you want by unchecking the "automatic handicap" button in the new game screen. What you should do though is try full board games against players of roughly equivalent skill level (of which there are plenty). When people say 9x9 is good for learning they mean (at least at the beginning; later it is good for other things) good for teaching the rules, and good for not overwhelming a new player's attention span.
posted by advil at 9:31 PM on May 26, 2014

What you should do though is try full board games against players of roughly equivalent skill level (of which there are plenty).

(by "players" I meant "human players".)
posted by advil at 9:37 PM on May 26, 2014

no joe, it isn't pattern matching in chess, it's rote memorization of openings, followed by analysis after that.

computers are actually better at pattern-matching (as opposed to pattern recognition, where humans excel). in the old days before computers took over chess, you could beat them by "looking over their move horizon". that meant you could foresee a situation happening in advance more moves out than the computer could foresee by looking at every single permutation. nowadays, the computer's move horizon is farther away than yours.

live chess strategy, you have to do it yourself, chess by mail, my opponents go to books, computers and colleagues.

the opponent's timing and body language doesn't mean diddly-squat. this isn't poker. as a single-digit age child, i had adults try to go past midnight to shut me down. as a double-digit age adolescent, i had rivals from other schools summon their girlfriends to lean over the chessboard from their side in low-cut tops to shut me down. doesn't work. the moves are the thing, and the clock, nothing more.
posted by bruce at 9:42 PM on May 26, 2014 [4 favorites]

Wow. My stepfather taught me how to play Go circa 1988, and I probably haven't played since 1998.

I suddenly have an itch to scratch.
posted by feckless fecal fear mongering at 10:10 PM on May 26, 2014

It’s the one classic game where wetware still dominates hardware.

For now.
posted by Mental Wimp at 10:15 PM on May 26, 2014

...intuitively picking up on their opponent's strategy by observing their timing and body language. That would be a hard thing to teach a computer!

But apparently they don't need it.
posted by Mental Wimp at 10:17 PM on May 26, 2014

Caught between atheism and a crippling fear of death, Ray Kurzweil and other futurists feed this mischaracterization by trumpeting the impending technological apotheosis of humanity, their breathless idiocy echoing through popular media.

lol, quote of the whole article right there. Perfectly captures the tone of Singularitarianism.
posted by cthuljew at 10:56 PM on May 26, 2014 [13 favorites]

Slap*Happy, there are lots of books like that (I enjoy the Elementary Go Series), but www.goproblems.com is easily a multi-hour diversion.
posted by klausman at 11:08 PM on May 26, 2014

Computers still can't win at poker, especially no-limit THE. They're too consistent to be unpredictable.
posted by Potomac Avenue at 11:12 PM on May 26, 2014

Rubick's Cube tip: Peel the coloured stickers off the cubes and place them on the appropriate sides.
posted by islander at 11:28 PM on May 26, 2014 [2 favorites]

One approach: Monte Carlo AIXI
posted by 0rison at 11:37 PM on May 26, 2014 [1 favorite]

So I'm still not sure what I'm reading here - has Go been finally cracked by the computer, beating the human? or as the title and the first links imply, its still a challenge for the computer programs to crack it?
posted by infini at 11:52 PM on May 26, 2014

has Go been finally cracked by the computer, beating the human?

Yes, if you give the computer the ability to place four stones first.
posted by the man of twists and turns at 12:02 AM on May 27, 2014 [3 favorites]

Another: the game of go as a complex network.

Go-playing programs have come a long way, but their programmers still have a slog ahead of them. I for one am confident they will learn to compete at human levels of play, and beyond.

Aside from the intrinsic differences between chess and go, it's also worth considering that the game of chess is a traditional computer problem. Claude Shannon was interested, and so was John von Neumann. Mikhail Botvinnik and Max Euwe spent some years at it, and a "chess-playing machine" astounded audiences well before actual computers were in any position to do so.
posted by 0rison at 12:06 AM on May 27, 2014

has Go been finally cracked by the computer, beating the human?

Yes, if you give the computer the ability to place four stones first.

Ah, the computer needs a handicap to beat the human.
posted by infini at 12:41 AM on May 27, 2014

What about Monopoly? Until they programme it to spot my sister hiding 500s under the board the computer has no chance.
posted by biffa at 1:27 AM on May 27, 2014 [4 favorites]

the opponent's timing and body language doesn't mean diddly-squat. this isn't poker. as a single-digit age child, i had adults try to go past midnight to shut me down. as a double-digit age adolescent, i had rivals from other schools summon their girlfriends to lean over the chessboard from their side in low-cut tops to shut me down. doesn't work. the moves are the thing, and the clock, nothing more.

bull.

psychology is massively important in chess.

take the Match of the Century, between Fischer and Spassky. Fischer forfeited the second game, and was going to forfeit the championship entirely when he was persuaded not to by his coach and second, and by Spassky's offer to play the third game in a small room out of sight of spectators.

only without the pressure of hundreds of eyes on him could Fischer have pulled off the win he did in the third game, his first ever win against Spassky, giving him a massive confidence boost. and no doubt when Spassky realised that his sportsmanship and sympathy had cost him a match his equanimity would've been disturbed.

in fact, he himself said after the match "my acceding to Fischer's groundless demand to play in a closed room was a big psychological mistake."

not convinced?

try Alekhine: "psychology is the most important factor in chess."
or Kotov: "you must not let your opponent know how you feel."
Kasparov: "you can't overestimate the importance of psychology in chess, and as much as some players try to downplay it, I believe that winning requires a constant and strong psychology not just at the board but in every aspect of your life."

chess is psychologically brutal.
posted by president of the solipsist society at 2:09 AM on May 27, 2014 [6 favorites]

I would like to see a computer consistently win at Dixit. That's the true test of an AI
posted by Just this guy, y'know at 3:16 AM on May 27, 2014 [3 favorites]

The Wired article is seriously one of the best popular articles I've read on artificial intelligence. It's clear that the author understands AI and how it works, and more importantly, what the state of the art is – that is, programmed single-task agents that are totally unrelated to the concept of human thought or intelligence.

And the takedown of Kurzweil is poetic in its magnificence.
posted by graymouser at 4:28 AM on May 27, 2014 [1 favorite]

Obligatory xkcd
posted by zinon at 5:09 AM on May 27, 2014

Joe in Australia is correct, chess is all about pattern-matching. Of course you have to calculate to make a good move, but effective calculation requires cutting down the branching factor of the game tree, and you do this by pattern-matching, saying (if only implicitly) "I know that Bxh7 is something I have to check out in this sort of position, let me see if it works", or "I know that knights are really strong on f5 in this sort of position, let me see if that looks like a good idea."
posted by dfan at 5:28 AM on May 27, 2014

Are these games available to be stepped through online? I love clicking through and watching games unfold, hopefully with a simple web interface...
posted by Theta States at 6:39 AM on May 27, 2014

When I worked at Bellcore in 1987 or so, I remember talking to Peter Langston a bit about programs to play Go. Apparently, no too much earlier, the state of the art was abysmal. Peter organized a tournament in 1984 and won in 1985.

My recollection was that he had come up with a decent way to prune the search tree, but for computer Go, that was a huge move forward.
posted by plinth at 6:39 AM on May 27, 2014

because some people are terrible judges who don't appreciate any kind of effort or creativity.

posted by vogon_poet


'nuff said
posted by ChurchHatesTucker at 6:47 AM on May 27, 2014 [2 favorites]

The saving grace is that real life is a game of poker, and humans are still better than AI at that game.
posted by KokuRyu at 7:06 AM on May 27, 2014

I also wonder if the current enthusiasm for boutique card and boardgames (it's been going on since the '90s, no longer a fad) is based on the realization that chess is no longer a measure of player-vs-player intelligence ever since computers started beating grandmasters regularly.

If you're going to match wits with another player, may as well add some whimsy and color to the contest, and build in enough unpredictable variables where computer scientists will only isolate and define winning strategies into algorithms long after the game is no longer being published.

(Of course, that means that some genius will build a system to automatically analyze any given game being played and determine rules and idealized gameplay, the bastard.)
posted by Slap*Happy at 7:15 AM on May 27, 2014 [1 favorite]

cthuljew: Caught between atheism and a crippling fear of death, Ray Kurzweil and other futurists feed this mischaracterization by trumpeting the impending technological apotheosis of humanity, their breathless idiocy echoing through popular media.

lol, quote of the whole article right there. Perfectly captures the tone of Singularitarianism.

It might be fun to dismiss nerd rapture, but that line is just sneering posturing. Obviously we're no where near uploading our minds to computers, but death is a pretty healthy thing to be afraid of, and I don't see anything "crippling" about that fear to these "futurists".

Of course these are "one use" machines, but you've got to crawl before you can walk. The "futurists" may or may not be be doing a diservice in their pop-sci writings, but if we do evetually create a "mind", it's hard to say how our current machines aren't a step on that road. It took evolution billons of years to create animal brains that can do what these machines can do. We only started trying a few decades ago. It's an impressive achievement, even if they aren't good conversationalists.

Death is the Final Boss of the Universe and musing about maybe one day these machines can help us beat it seems at least as worthwhile as making fun of people on the internet. For the first time in history, we understand some of the concepts around it, and I wouldn't be suprised at all if in a century or so we can slow down aging so much that individuals could last long enough to actually get to some later "singularity"-looking breakthrough. Too late for Kurzweil, sure, but hey, why wouldn't you want to try to make it?

I more or less agree with one part of futurist thought; we're all rightly terrified of "premature" death, but we're all supposed to find something spiritual, glorious and right about death in the abstract, and the assurance that death is what ultimately gives life meaning sounds a lot like someone trying to convince themselves. I'm not deluded to think we are anywhere near it, but if some unexpected, crazy advance happens in the 2050s? Sign me up.
posted by spaltavian at 7:49 AM on May 27, 2014 [2 favorites]

The trick in chess is that the same pattern can be present in two positions where none of the pieces are in the same places, either on the board or relative to each other. Spotting a pattern in that event is a bit beyond what most people think of as pattern matching.
posted by Flexagon at 7:56 AM on May 27, 2014 [1 favorite]

spaltavian: I pretty much totally agree with everything you said, except that I think Kurzweil and his ilk are hacks and charlatans, and his being hired at google is going to set the search for an artificial mind back by decades.
posted by cthuljew at 8:13 AM on May 27, 2014

I know a prominent simgularitarian with a definitely crippling fear of death. In 2009 he signed up for one of those services were they will chop of your head and freeze it when you die.

Since then he has restricted all his travel to a narrow strip between San Francisco and San Jose. He will not even go to a picnic at the park in case he dies and the ambulance can't find him in time to save his head.

He'd love to visit his family, but that means a transatlantic flight. Unacceptable risk.

On the other hand, using his highly honed Bayesian mind he made a lot of money on Intrade some years ago, and spent most of it on bitcoin. Who am I to argue against a millionaire?

I sent him this article, he says it is very good for a popular magazine, but singularitarians will win in the long term. I think he is right if long term is measured in geological eras.
posted by Doroteo Arango II at 8:18 AM on May 27, 2014 [1 favorite]

Are these games available to be stepped through online?

Yes. Here's Yoda vs. Crazy Stone.
posted by sfenders at 8:33 AM on May 27, 2014 [2 favorites]

For chess, the best programs use a form of minimax, which searches all reasonable moves to a limited depth using a complex evaluation function. For go, the current best approach is Monte Carlo tree search, which simulates lots of complete games assuming naive or mediocre players, prioritizing the most promising lines of play.

The go approach doesn't work with chess, because chess has many "traps", so missing a single move can be fatal. Also it requires careful planning to set up a mate, so you need two somewhat-competent players to even finish a game.

Conversely the chess approach doesn't work with go, because we haven't been able to write an evaluation function that encapsulates the entire board. It's better to make a good move that might win the game than a great move that improves your position in the short term.

It's interesting that people often characterize go as "flowing" and having more holistic play, because your computer sort of thinks the same way.
posted by RobotVoodooPower at 8:39 AM on May 27, 2014 [1 favorite]

Chess playing programs also have massive opening and closing dictionaries. Which means they aren't really doing any search for the first moves, or the last moves. At a certain point enough pieces have been eliminated that the computer can just do a lookup to find the next move toward mate, or know to resign. The oldest tactic against computers was to try to take them "off-book" as quickly as possible in the opening. That doesn't work anymore because opening dictionaries have gotten bigger and search has gotten deeper.

Of course with Go, you can't use closing dictionaries. But the real difference, despite all this talk about "flow" and "nature" and other things that make you feel good about being a human and not a computer, is simply Go's search space is too big to be searched effectively by computers at the moment. Massively parallel machines will make a dent in that. Eventually computers will own Go. I don't care to make a prediction on when they will, but there is no question as to if they will.

As far as reading your opponent, that is certainly a factor in chess. At least in amateur chess. I know I have a tendency to sigh in a certain way when I make a move and then immediately realize that it was blunder. About half the time my opponent won't notice... unless I sigh. Furthermore, watching your opponents eyes does give you insight into the moves they are considering. However, I think claiming that this would be hard for computers to analyze is wrong. Eye tracking software exists. Extending that to reading body language, visually and audibly, is very much within the range of things we can program computers to do. Though I don't think there's much point in doing it since the advantage at the professional level would be minuscule compared to the effort involved. And a false positive would be quite damaging... but it's an interesting problem. And one that I believe is tractable.
posted by jeffamaphone at 8:52 AM on May 27, 2014

As an aside, I'd like to plug the Dragon Go Server for anyone looking to play online at a nice pace - it's at a "play-by-email" pace, so you can take a game a turn or two a day, if desired.

I found it when I decided I wanted to learn Go. Looking forward to my first win - maybe I can move up from 30 kyu afterwards...
posted by evilangela at 9:28 AM on May 27, 2014

Of course with Go, you can't use closing dictionaries.

You kinda can, since the endgame stage (yose) is, pretty much by definition, the stage at which the game can be decomposed into discrete areas that don't affect each other, and those areas can be evaluated exactly. There's lots of information in Mathematical Go Endgames by Berlekamp and Wolfe. The very end of a Go game is the one phase that computers can play better than the best humans.
posted by dfan at 10:13 AM on May 27, 2014 [1 favorite]

The force of psychology really showed in the last few matches that were held between GMs and top programs. The computer has none, and it has no memory. It can get crushed one day and come back right as rain the next day. Humans have to recover their confidence, though, after a loss, and that is a weakness.
posted by thelonius at 10:32 AM on May 27, 2014 [1 favorite]

Rubick's Cube tip: Peel the coloured stickers off the cubes and place them on the appropriate sides.

Did people really do this? Sounds like a lot of tedious work—not to mention a great way to permanently ruin your cube. Our method was to simply pull apart the individual pieces, rearrange them into a solved configuration, and then pop them back in place. With practice, it could be done in under a minute.
posted by Atom Eyes at 1:56 PM on May 27, 2014 [2 favorites]

The very end of a Go game is the one phase that computers can play better than the best humans.

Ah okay. I'm not a Go player. :)
posted by jeffamaphone at 2:03 PM on May 27, 2014

Earlier I assumed that sgf records of the UEC Cup games would be trivially easy to find. In fact it may require guessing that links to the 1st and 2nd days' results show up only on the japanese version of the page, not the english translation.

One can report steady progress, all the way to the top of the tree.

It's interesting to think of go in relation to Hofstadter's approach to AI. The way I play (not that I'm any good; at best I was rated 8 kyu on KGS) subjectively seems to have a lot in common with the things I remember his algorithms doing in Fluid Concepts and Creative Analogies. The point on which I would disagree with him as quoted in that Atlantic article is the degree of emphasis on "understanding human intelligence". Efforts to do so can and do suggest interesting things to try, but to me the important thing is that any resulting intelligence be recognizable as intelligent, not as human. Any AI that can demonstrate general intelligence ought to be counted as a strong success even if its mind isn't anything like ours. I don't know exactly what the test should be, but if there's a computer program that could learn to play chess, learn to play go, learn to fly an airplane, appreciate Shakespeare, empathize with humans, and play with a dog, all without being programmed in advance for any of those specific tasks, then that would be good enough even if its mind didn't work anything like mine. I always wanted to take some of those little programs from Hofstadter's group, add some more structure and complexity to their thought processes, let them reproduce and mutate, and put them in a simple virtual world to see how they evolved. It's unlikely they would evolve to the point where they'd invent the game of Go, but maybe they'd do something interesting. I wonder how many people have tried it.

I'm pleased to learn that the best humans are still better at go than are the best computers. If "another breakthrough" is required, just possibly it could be one that requires something more recognizable as intelligence.
posted by sfenders at 3:26 PM on May 27, 2014

Our method was to simply pull apart the individual pieces

That's how I always did it too. A small flathead screwdriver made the whole thing ridiculously easy.
posted by Mr. Bad Example at 3:58 PM on May 27, 2014

What's the Go equivalent of chessgames.com ?
posted by jcruelty at 1:06 AM on May 28, 2014

What's the Go equivalent of chessgames.com ?

I'm not sure there's anything like that that is free, at least with an English interface. The closest I know of is the gogod database. The Smartgo Kifu app that some people mentioned above comes with this database as well.
posted by advil at 5:23 AM on May 28, 2014

Excellent post, the man of twists and turns. Thanks.
posted by homunculus at 10:48 PM on May 29, 2014

If an AI could master The Four Arts, I would bow before it.
posted by ovvl at 6:33 PM on June 8, 2014

Let's "Go!": Ancient Chinese Game Boards
posted by homunculus at 10:11 AM on June 14, 2014 [1 favorite]