Author: Peter McKenzie
Date: 20:37:06 06/11/02
Go up one level in this thread
On June 10, 2002 at 11:39:18, Alberto Rezza wrote: >On June 09, 2002 at 19:29:33, Russell Reagan wrote: > >>You do a very nice job of manipulating the intention of everything I said. > >I have to say you are wrong here. > >Of course, with deep enough search (say 50- to 100-ply for chess, 300-ply or >more for go) strategy completely disappears and everything is tactics. Then you >would be right. > >But if we keep the distinction betwen strategy and tactics, it is true that by >reducing the size of the go board you lose a lot of strategic contents. You lose >all of it if you go all the way to 2x2. So he was right and you wrong: it's not >the same game. Of course this same reasoning would apply to any form of >mini-chess. > >To put the whole discussion in perspective, here are some facts (I can give no >sources now, but I could dig them up in variuos places). All the rankings I give >are comparable to Japanese amateur rankings (NOT to European, USA, IGS, etc. >rankings). > >1. 7x7 go is almost solved. 9x9 go is widely regarded as somewhat less complex >than chess, but more complex than checkers/draughts. 9x9 go is actually played, >even by masters, though less than 19x19 go; 13x13 is played very little. > >2. The best go programs play at about 5- or 6-kyu level; but they go down to >about 10-kyu in just a few games, after the human adapts to their artificial >style of play (and peculiar blunders). In my opinion, a very dedicated human >beginner (one who plays every day) could reach 6-kyu in one month; however, very >few people actually do it in less than one year, and some NEVER reach that >level. > >3. BTW, I used to be 3-dan (I'm weaker now for I don't play) and I could beat >the best programs giving a 9-stone handicap, or more. > >4. It's a fact that go is much more difficult TO PROGRAM than chess; but this >tells us nothing about which game is more difficult FOR HUMANS to play. I'm not so sure this is a proven fact. If you compare the amount of effort that has gone into computer chess, it is orders of magnitude greater than for computer go. There have been hundreds of computer chess programs, many hundreds of academic papers and many books and doctorates on the subject plus a rich history that goes right back to the 1940s. Certainly go is a different challenge that does not respond well to the same techniques used in computer chess, but I don't think we can say for sure that it is more difficult to program. It may well be, but only time will tell us this. Think of it this way: if someone produced a world champion go program tomorrow, would that mean that go was a more difficult to program than chess? I think you can measure the difficulty of a problem by looking at the amount of time and effort required to solve it. If it turns out that it takes less overall time (in terms of total effort put into the field) to write a 'grandmaster' go program than it did for chess then we can perhaps say that go was in fact easier to program than chess. Of course increasing computer power is a factor, but not the critical one in my opinion. Peter > >5. The difficulty in programming go lies in the evaluation function, not in the >branching factor. Chess programs don't even try to "think like a human" because >they have found a simple shortcut that works: alpha-beta with a simple eval >(material, mobility and little more) gives a reasonable level of play, even >without any other refinement. In go this does not work, and no other shortcut >has been found yet, so the programs have to do complex things like pattern >recognition to try to find the strategic features of a position, or deep >tactical searches INSIDE the eval just to see whether a group is alive or dead. >Unless some unexpected breakthrough is found, I guess no program will beat a go >professional (more or less the equivalent of a grandmaster) in this century. > >6. The relationships (if any) among the concepts of "most complex", "most >difficult" and "most interesting" game are best left to each one's opinion. > >Alberto
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.