I think if someone is willing to offer US$10 million for the first computer that can serious take on a 4-5dan professional player, the race will be on.
Actually, someone in Taiwan (a certain Mr. Ing, I beilieve), has been sponsoring tournaments for Go-playing programs, and offering a prize (something like a million dollars) for any program that can beat a strong human player (something like a 1-Dan pro). This has been going on for at least 15 years, and no one is even close to claiming the prize money.
Go-playing programs can be very good at LOCALIZED problems, e.g. a life-and-death problem in a small part of a corner. But they are horrible at making GLOBAL decisions.
The difficulty in programming Go is not just the width of the move tree (in the middle of a Go game, there are a couple of hundred possible moves; in the middle-game of chess, a couple of dozen). The other problem is evaluating a given position. In chess you can devise a simple, fast, fairly reliable static evaluation function that tells you for whom the situation is favorable, and by how much (by assigning values to the types of pieces, and by looking at which pieces are attacking which other pieces). You can even do this in hardware.
But for Go, no one has been able to devise such simple, fast, global evaluation function.
Seems that the version of NFS on the Red Hat 6.0 distribution is badly broken. My friendly local sysadmin spent some unpleasant hours figuring out why remote filesystems kept disappearing...
Slashdot Mirror
"I" is possesive
:)))
My, oh my, I didn't know that "I" was possesive. I guess "you" are, too, then
... Robot as president ... Doesn't everyone know that robots have been president for some time now?
I think if someone is willing to offer US$10 million for the first computer that can serious
take on a 4-5dan professional player, the race will be on.
Actually, someone in Taiwan (a certain Mr. Ing, I beilieve), has been sponsoring tournaments for Go-playing programs, and offering a prize (something like a million dollars) for any program that can beat a strong human player (something like a 1-Dan pro). This has been going on for at least 15 years, and no one is even close to claiming the prize money.
Go-playing programs can be very good at LOCALIZED problems, e.g. a life-and-death problem in a small part of a corner. But they are horrible at making GLOBAL decisions.
The difficulty in programming Go is not just the width of the move tree (in the middle of a Go game, there are a couple of hundred possible moves; in the middle-game of chess, a couple of dozen). The other problem is evaluating a given position. In chess you can devise a simple, fast, fairly reliable static evaluation function that tells you for whom the situation is favorable, and by how much (by assigning values to the types of pieces, and by looking at which pieces are attacking which other pieces). You can even do this in hardware.
But for Go, no one has been able to devise such simple, fast, global evaluation function.
Seems that the version of NFS on the Red Hat 6.0 distribution is badly broken. My friendly local sysadmin spent some unpleasant hours figuring out why remote filesystems kept disappearing ...