I’ve been playing the ancient game of Go for years now– badly; I am ranked around 6 kyu at my best– and there are some fascinating lessons to be learned from Go play and attempts to program computers to play Go.
Go, like chess, is a strategy game where each player moves in turn, each sees everything the other player sees, each move opens certain possibilities and closes others (it’s deterministic), and there can generally be only one winner (draws happen only very occasionally). Go is big business and has a long history: the top players earn seven-figure salaries; there are Go-only TV channels in China, Korea and Japan; there is a rich history of game recording and analysis stretching back to the sixteenth century; Go championship matches draw hundreds of millions of viewers; Go was considered one of the essential brain-training tools for samurai, emperors, etc.
Now, in terms of complexity, Go is literally billions of times more complex than chess. A Go game has 361 possible opening moves; a chess game 20. In Go, the typical game length is 200-250 moves; in chess it is 40. There are more potential Go games than there are atoms in the Universe. Computer games reflect this: in chess, Deep Blue beat Grand Master Kary Kasparov in 1997; off-the-shelf software now always beats top humans; chess software’s greatest challenger is other chess software. Go, however, has proven a harder nut for the egg-heads to crack. If you imagine players ranked from zero (your kid learning to play) to 20 (the best two or three players in the world), Go programs now operate at about a five; chess programs at 25 or so.
In this article, the challenges facing Go programmers are detailed. And there are a lot of interesting things we languages teachers can learn from research into programming Go and into what makes top players tick.
First , when analysing top players, an interesting pattern emerges. It turns out that there is a fairly predictable order of acquisition of strategies and a fairly consistent (as a function of time spent playing and analysing) speed of getting better. Indeed, top players can often tell their lesser opponents’ ranks within ten moves. But…when good players become great players, they don’t just get more skills…they get entirely different skills. Move (and response) unpredictability goes way up in a kind of quantum leap. What look like wild, crazy moves start happening, and these moves end up being the keys to winning. Go programs don’t do that. With them, it is a “more of same” and slow-and-steady approach…that doesn’t work.
So it is with languages. As Chomsky and others have noted, while there are definite orders of acquisition (of sounds, grammar etc) with language, people start being able to do unusual things. They can generate sentences they havn’t heard, they use grammar rules they havn’t had explained, there are mistakes they could make (but don’t) and they can understand things they have not had explained to them.
A friend of mine did her PhD in linguistics on ambiguous noun classes. She knew that kids somehow figured out that, for example, the word “school” meant different things in different contexts, and what those meanings were. E.g. you can say “I like school,” and “my school is close,” and “school is hard.” In each, “school” has a slightly different meaning. Kids exposed to nouns that are morphologically identical figure out the differences in meaning, and they do it amazingly quickly, and they do it without help. The upshot of her experiments with noun acquisition was that the kids could not have figured out (from context) what the different meanings/uses were. So she concluded, as Chomsky predicted and then showed, that the brain’s “language organ” has powers that cannot be explained by the quality/type/context of input alone.
Second, since people massively beat even the best computers at Go, we know that– somehow– they are making better decisions than the computer. In other words, Lee Sedol is somehow calculating– and comparing– quadrillions of decision trees per second in a championship match. However, we know that the conscious brain processes only 2,000 bits of info per second. So most of the real processing is not happening consciously. When these top guys (and they are all guys) are asked “what are you thinking about?” during games, they usually say things like “well I just kinda look” or “that move felt right.” Indeed, they cannot often explain– especially at the very top level– why they do certain things. On the way to becoming masters, there is conscious study, reflection, etc– oral analysis of games, moves etc are part of the tradition– but when you’re in the moment, you just…let go and play.
Third, “skill” in Go (as in chess, poker, bridge, etc) comes– I think– as much if not more from observation as it does from play. Good players spend an immense amount of time replaying old matches from the masters, dating right back to the 16th century. They also ruthlessly review their own games, and watch other players playing, and now, with computers, can watch other people’s saved games. (On Go servers, sometimes hundreds of people will tune in to watch the 8-dan players duke it out.) This is input. It’s also something like reading: you can “examine” at your own speed, go back, pause, etc.
While with games, playing obviously matters (and is the point), and observation is, as with languages, central.
(a) Most language acquisition cannot happen consciously. If you want to have maximum acquisition, you are going to have to let the unconscious do its work. Indeed, you are going to have to get focus on rules etc out of the way as fast as possible. Minimising grammar explanations, maximising interest, and making people happy and comfortable will get people “immersed” in the story. Indeed, if you focus on the conscious brain, you are majorly limiting yourself and your students: 2,000 bits per second of processing, or billions?
(b) If we provide quality input, we will eventually get “quantum leaps” in skills. Kids will pick up and say things that you havn’t consciously “taught” them. Today in Spanish we were playing with a story where Farakh doesn’t serve his chair-stacking detention with Mr Stolz, because, on his way to Mr Stolz’s class, he meets a talking cat, and he sooo wants a talking cat, and when he asks the cat “do you want to be my cat?” and the cat answers “no, I don’t want to be your cat, you have to stack chairs!” he goes home and eats 3 pizzas in disappointment. One kid, Wasim, blurted out Sólo Farakh quiere hablar con una gata. (“Only Farakh wants to talk to a cat”). Perfect, unexpected, unprompted Spanish. I have never used that sentence before. These guys– level 2s with only a month or so left– are now starting to blurt things out.
The bottom line seems to be, while practice (via input) is necessary, most of what is happening in the brain– in Go as well as in learning a language– is beyond/below the conscious.