Go : The future of computing

August 2, 2002 8:35 AM

Go : The future of computing "In recent years, computer experts, particularly those specializing in artificial intelligence, have felt fascination - and frustration - with an ancient Asian game called Go. To date, no computer has been able to achieve a skill level beyond that of the casual player."
posted by jragon (29 comments total)

Go is an amazing game. The article is exactly right -- it's really easy to learn, really hard to play well. Check out this online community, or download a software version and try it yourself.
posted by me3dia at 9:00 AM on August 2, 2002

Yeah, but they kick our butts at Stay Alive.

Alternate joke: Sure, but can they predict the stock market?

*9:08am: press Return*
posted by Kafkaesque at 9:07 AM on August 2, 2002

You know, of all the people I know who play Go, none are math-types. Of all the people I know who are math-types, none play Go. Yet, in the movies, all math-types play Go, and no one else understands it.
posted by Fabulon7 at 9:12 AM on August 2, 2002

I still don't understand what's so hard about Go?
posted by starvingartist at 9:19 AM on August 2, 2002

Slashdot discussed this story a few days ago, if you're interested in seeing hundreds more opinions.

The NYT story is here.
posted by wackybrit at 9:32 AM on August 2, 2002

... 'Course, I always liked Othello when I was a kid, so I guess I'm predisposed to enjoying Go.
Found another site for online Go, including a Java version. Unfortunately, you either have to play both sides yourself or have a friend sit with you at your desk -- no computer play.

Kafkaesque: have you been staring at the sun again?
posted by me3dia at 9:34 AM on August 2, 2002

It's all about the branching factor. I wonder if the article places too much emphasis on the notion of building an intuitive computer player. As technology advances, would it be possible to build a strong player just by throwing compute power at the problem (see Deep Blue)?

I'm an AI person, but not a Go player. Someone, please enlighten me.
posted by kelperoni at 9:39 AM on August 2, 2002

As I understand it from my (admitedly fairly limited layman's) reading on cognitive whatsits, what we call 'intuition' is mostly just sort of informational 'macros' which execute without our being aware of it. so, idea 'a' might be related to 'b' which leads to 'c' and on to 'd'; but 'd' might be related to 'z', which 'a' is also related to... and your brain seeing that connection and skipping right to 'd' is what we call 'intuition.' the point is that anything you can do with 'intuition' you can do slower with brute force., it's just gonna be slower.

So, you can get from the beginning of a chess game to the end of the game by following the path of formal logic - a -> b-> c etc. Really good chess players, however, keep maps of hundreds of board layouts stashed in their memory, so that they don't have to follow the formal logic for every move, they can just know that board layout 'a' goes with board layout 'd' and move accordingly... so it seems impossibly fast. A computer can still get to the best move as well by following the logical steps, and get there just as fast because computers just go faster at chunking their way mindlessly through large blocks of data.

Go requires you to be able to do that memory trick in order to be a reasonably good player, because the number of positions available to you for any move in the game are so high that thinking your way through them is impossibly slow. So in order to be able to defeat a human at Go, the computer has to have that same set of 'maps' generated by experience playing the game. Intuition.

At least, that's how I understand it.
posted by hob at 10:11 AM on August 2, 2002

Current rankings of go-playing computer programs with lots of links

So in order to be able to defeat a human at Go, the computer has to have that same set of 'maps' generated by experience playing the game. Intuition.

I'm sure Wintermute is close to solving this one.
posted by mediareport at 11:33 AM on August 2, 2002

For varying definitions of "close", surely.

Give me... five years, I'll see what I can do.
posted by beth at 11:45 AM on August 2, 2002

read The Master of Go by Kawabata
posted by scarabic at 11:56 AM on August 2, 2002

To paraphrase Trevanian, Go is to chess what philosophy is to accounting.

All the Go software I've used is still primitive, easily defeated even after handicapping oneself many stones. The branching factor in Go is enormous compared to chess, and at anything other than the lowest tactical level, search strategies are futile.

As noted above by hob and others, strategies that may more closely mimic human "intelligence" (which is really nothing more than the mechanical manipulation of symbols) will produce better Go programs.

Take a few minutes and learn the few rules of Go. You'll be hooked for life.
posted by fold_and_mutilate at 12:15 PM on August 2, 2002

For varying definitions of "close", surely.

Do Not Doubt Wintermute.
posted by mediareport at 12:18 PM on August 2, 2002

fold_and_mutilate : Take a few minutes and learn the few rules of Go. You'll be hooked for life.

My problem is that everyone I know who knows how to play Go is too good to want to play Go against the likes of newbie me. Sigh. I guess I'm stuck with chess for the time being.
posted by hob at 12:52 PM on August 2, 2002

human "intelligence" (which is really nothing more than the mechanical manipulation of symbols)

Oh, so that's what it is.
posted by bingo at 1:33 PM on August 2, 2002

human "intelligence" (which is really nothing more than the mechanical manipulation of symbols)

Oh, so that's what it is.

Yeah, that one stopped me in my tracks, too. There are so many different kinds of intelligence. And what about emotional intelligence?

I'm sure when the machine intelligences decide to announce themselves to us monkeys (I'm sure there's a heated debate going on in the machine community as we speak about how best to do that), they'll point out many measures of intelligence that make sense to them but never occurred to us.

Won't that be a grand day?
posted by mediareport at 3:20 PM on August 2, 2002

what about emotional intelligence?

It's an oxymoron. Emotions are things that happen to you; intelligence is the you that they happen to.
posted by kindall at 3:58 PM on August 2, 2002

My problem is that everyone I know who knows how to play Go is too good to want to play Go against the likes of newbie me.

You might try playing online, at either the Internet Go Server or the (open source) No Name Go Server. There seems to be some good educational material at Sensei's Library, a wiki about Go. The Go Teaching Ladder also looks very interesting--more experienced players review games by less experienced players and offer advice and criticism. (I should probably note that I haven't played Go for a couple of years, so I can't say from personal experience how good these sites are.)
posted by moss at 4:29 PM on August 2, 2002

[Emotional intelligence] is an oxymoron.

Do you feel the same way about "animal intelligence?" Why or why not?

Emotions are things that happen to you; intelligence is the you that they happen to.

Well, that's a unique definition of intelligence, to say the least. It sure doesn't fit with what scientists who actually study intelligence are saying.
posted by mediareport at 5:23 PM on August 2, 2002

You can also play Go at Yahoo Games, which is a very easy process.
posted by wackybrit at 5:25 PM on August 2, 2002

For those unfamiliar with Go, I recommend Steven Den Beste's essay on Chess and Go.
posted by D.C. at 1:09 AM on August 3, 2002

For decades, the AI community viewed chess-playing ability as one of the highest examples of our intelligence, something that separates us from the machines. Turing even included a chess problem in his sample transcript of a Turing test.

Chess' status as the pinnacle of human logical reasoning explains why IBM was ready to invest so much in Deep Blue.

Years of hard work and Moore's law did the job, and suddenly everyone says that chess is just a simple, brute-force problem. "Nope, no real human intelligence there. To be truly intelligent, a computer needs intuition, and just look at how bad computer Go programs are."

Artifical Intelligence will always be defined as the thing that we can do but computers can't do yet.
posted by fuzz at 12:55 PM on August 3, 2002

fuzz: Good point. Then again, there are plenty of human beings without much intuition, who are terrible at both Go and chess. And yet while "intelligent" might not be the most apt way to describe such people, they are certainly sentient in the way that AI strives for. Isn't that the important thing? For a computer to be "intelligent," does it really have to be able to win a strategy game against every human it plays?
posted by bingo at 2:28 PM on August 3, 2002

Some good points have been raised here, and as the sort of person you're talking about, I have to jump in.

I am awful, awful, awful at logic games, like Chess, Othello, Checkers, Go.. and also bad at most card games where logic is the order of the day. I don't score particularly high on IQ tests (usually around 120) and I am awfully blind when it comes to strategy.

I am, however, very good at more abstract games like Poker, and skills such as running my own business, working out how things work, and doing things that other more intelligent people just can't work out how to do (indeed, that's my job).

Since even a computer on the easiest setting can beat me at almost every logic game, does this mean I'm not intelligent? Just because an average ten year old can beat me at chess, does that mean the same thing? I was always at the far bottom of the chess ladder at school, yet I was always top of the class. This made no sense at the time.

I'd like to think that perhaps there are different types of intelligence. Perhaps this can explain why I can do things that are traditionally difficult (writing compilers, device drivers, operating system code), yet I have a very rigid intelligence that cannot deal with the random and flexible things that most people deal with easily (like games, which are random).. Could be said I have no intuition? Maybe.

Just thoughts. I could be wrong. I am dumb after all ;-)
posted by wackybrit at 2:54 PM on August 3, 2002

My advice for learning go is to do as I did: I sat down with a good friend with a similar level of intelligence and a desire to learn the game. For six months, we played about a game every two or three days against each other, and by the time that was finished and we went off to our respective corners of the world for research, we could find no one of a caliber high enough to make competition fun.
posted by kaibutsu at 3:13 PM on August 3, 2002

Remember that scene with john Nash losing at Go in A Beautiful Mind? He invented a game that suited him better - Hex. Probably because he hated losing.
posted by Lame_Dave at 7:49 AM on August 4, 2002

I'd make my own game that I could get good at, but strip poker already exists.
posted by wackybrit at 9:39 AM on August 4, 2002

For a computer to be "intelligent," does it really have to be able to win a strategy game against every human it plays?

Well, it all depends on how you define intelligence. Does intelligence mean that you always make the "correct" decision? Then yes, the computer should always be able to beat its opponent. But you could also define intelligence as being "human-like", a la Turing...
posted by kelperoni at 11:13 PM on August 4, 2002

Does intelligence mean that you always make the "correct" decision?

No.
posted by bingo at 12:27 AM on August 5, 2002