Author: Dann Corbit
Date: 16:31:40 02/13/04
Go up one level in this thread
On February 13, 2004 at 19:03:36, Russell Reagan wrote: >On February 13, 2004 at 17:26:49, Dann Corbit wrote: > >>That would be roughly 1,125,899,906,842,624 >>or one quintillion times faster. > >Let's make some assumptions. Assume that today we have a go program that can get >a branching factor of 9, and the program can get 10Mnps. That means it could >complete a 14 ply search in about a month. With the kind of speed up you talk >about, it could complete that same search in 0.02 pico seconds! This also means >we could do a 30 ply search in about 40-50 days. But I have to wonder, what is >30 plies in go? Games last several hundred moves regularly. We still have to >find some method to evaluate positions with some degree of accuracy. 30 plies >sounds like a lot for a human to cope with in any game, but if no method of >accurately evaluating positions is discovered, it doesn't really matter how deep >we can search (unless we can search to the end of the game). It should be >interesting. In 1970, the chess programs were pitiful. Amateurs could easily beat them (In 1976 they bought a chess program to my school and every one of my friends that tried to play against it won -- none of us were anywhere near to a GM). Now, here we are 30 years later and the GMs are fighting for their lives against these things. 30 years might not be enough in Go. But I am pretty sure that 100 years will be. You don't have to be able to see 100 moves ahead. Just farther than your opponent by a couple plies and also not make any stupid positional errors. In 2104, the computational strength of a computer will dwarf that of a human. They could (by then) encode neural net algorithms and teach the machine to play. In very short order, it will be pounding the stuffings out of humans.
This page took 0 seconds to execute
Last modified: Thu, 15 Apr 21 08:11:13 -0700
Current Computer Chess Club Forums at Talkchess. This site by Sean Mintz.