Computer Go |
Another View of Computer GoPeople often ask "Why are computers so bad at playing Go, relative to their success at other games?" A standard answer refers to Go's branching factor. A more subtle answer refers to the difficulty of evaluating a Go position.I would like to reword the question as "Why are computers so much worse than humans at playing Go?", or better, "Why can humans play Go so much better than computers?" Put like this, the question is easier to answer. Humans are good at Go because they are good at reading out lines of play. They are good at reading out lines of play because the pieces don't move about. In chess (and most other games) the pieces move about. This makes lines of play hard for humans to visualise, but is no problem for computers. In "sowing" games, such as awari and mancala, the pieces move about a lot: computers can be very good at these games. In Othello, the pieces don't move about, but they change state repeatedly. This makes things difficult for humans to visualise, but is no problem for computers. Computers are good at Othello. What happens when Go pieces do move about?
Rarely, Go involves a position which humans are bad at,
relative to computers. Here is one. Black is to play and live.
Before you read any further, try to solve this problem. Thomas Wolf's program GoTools solves it in a fraction of a second. The obvious answer, Black a4, does not work. White answers at a6.
Here is the solution: This problem is harder for humans than for computers because the pieces do move about. Twice, in the correct line of play given above, a white stone appears at a4, and twice it disappears again. Problems like this are called "ishi no shita" problems. They can be very hard for humans, but present no particular problem to programs such as GoTools. |
HomePlease send any comments to the IGF webmaster |