Author: Dann Corbit
Date: 15:34:02 01/05/99
Go up one level in this thread
On January 05, 1999 at 18:13:57, Peter Kappler wrote: >On January 05, 1999 at 16:57:40, Dann Corbit wrote: >>On January 05, 1999 at 15:19:09, Peter Kappler wrote: >>[snip] >>>"Much stronger" is pretty vague. This could mean 50 ELO points or 200 ELO >>>points, depending on your point of view. Can you be more specific in your >>>answer? > >>Perhaps orders of magnitude stronger. > >Argh! I had just finished complaining that Eugene's "much stronger" comment was >too vague, and I'm afraid this isn't much better. I'm looking for quantitative >estimates, specifically in terms of ELO ratings. Deep blue is probably 2700-2800. So about 200 points better than the best programs would be fair. Of course, this is orders of magnitude stronger than the current programs. Only under very rare circumstances will a 2700 program be beaten by a 2500 program. >Please if anybody else has an opinion, and wants to respond, at least give a >rating range as part of your answer. Thanks. > >> I worked at Microsoft for about 10 years >>and I can tell you that they have a large number of very talented people who >>work there. You may imagine some giant pool of chowderheads, > >No, I don't imagine that at all. I know some Microsoft people too, and they are >all very bright. But I am of the opinion that computer chess is now in the >realm of diminishing returns. The current commercial programs already search so >deep that I think even a 50-100 point improvement in playing strength would be a >tremendous accomplishment. > >I suspect that Microsoft could build something that would reach that "50-100 >point stronger" point if they devoted enough resources. Again, keep in mind >that I'm talking about a pure software implementation, and not a mixed >hardware/software project like IBM's Deep Blue. If they attacked the problem, it would probably be dramatic, like IBM, to get attention. I suspect they would use hardware. Software only improvements? I think it is silly to try to predict how far you can get doing this. Someone may invent a smart program that can reason like a GM, by creation of a very elaborate evaluation function. I have a friend who is working on a substance which will store a terabyte of data per square centimeter. Suppose a square foot of this were used to store chess information? They have 50,000+ computers at Microsoft. Whenever someone says, "You could not possibly do better than 'X'." I always imagine that "X+Y" is possible, or even "X*Y". For example, one time someone showed me a proof that you can't calculate x, given y, such that x to the x power is equal to y. In five minutes I wrote a program to solve it. I do agree very strongly with your point about +100 elo points. It is a giant leap, especially at the top of the scale. It is also a non-linear leap as we near the top end. A 3000ELO program would basically never lose (except to another program of that strength or a super-GM, should one arrive of that 'size'). Despite the fact that this is only 200-300 points better than what the world has to offer right now. (Fiddle with a bit of math to find the probability that a 2500 GM will beat it, for instance).
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.