Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: Some Philosophical questions on the limits of Computer chess

Author: Vincent Diepeveen

Date: 17:32:35 01/25/02

Go up one level in this thread


On January 25, 2002 at 16:46:38, ALI MIRAFZALI wrote:

>I would like to know the opinion of the readers of this forum on the following
>questions.
>1.What would be the Elo rating of the perfect chessplayer?

a very low one.

because within a few games they figure out the perfect game for white
and for black by seeing how the 'black perfect box' plays.

then everyone draws the perfect player forever as it doesn't
dare toplay other moves fearing to lose from low rated.

So then if it joins a tournament it always will get the perfect 50%
score or if white starts the game and wins then it will usually have
the bad luck to lose a bit more than 50% because if you play 5 times
black and 4 times white you have then 44.44% score.

So in short the thing would be pretty boring. apart from that. it
would be a nightmare for the military industry a thing that can decode
and crack near to any encryption not to mention passwords. how are
we going to have passwords online if such an enormeous calculation
power is available?

But suppose the perfect player is also perfect in swindling so it
scores 100% against everyone (this is unlikely but well let's assume
it as this discussion is nonsense anyway)

then it would have at most 750 rating points more than the highest
rated player. So right now that would mean around 3500.

>2.Are there natural limits to the strenght that can be achieved in chess
>for a computerchess player? (Not the present centuary but in any future;that
>is what I mean by natural limits).

Yes, even 10^30 calculation power is going to not play perfect for sure
and lose games. Note that 10^30 is enough to factorize composites up
to like 4000 bits easily.

>3.If the rating of perfect player is say x ;what would be the rating of
>the stongest computer player ever(that is the best chessprogram that can be
>ever contructed useing computer technology) .It would be x-?.Or would it be x?

No it will be never X, because computers are not intelligent.

X-300 is a good guess.

It also depends upon the rating system you use.

i assume everywhere FIDE rules. meaning K=10 and the table as they
use it now. this table is not perfect. also things are not added
to the calculation.

For example chance that i lose from a person older than 20 who is
1800 rated (me 2312) is like a few %. Well since i am over 2200
i *never* even drew a game against a 1800-.

I lost exactly ONE game against a youth talent who is 2100+ now
(was then 1950 rated or so and underrated for sure) and i drew
one game against a female of 2070.

So last 8 years that makes practical 0 losses against 2000 rated
(unless you want to count someone without rating and now 2070
and a youth talent who was underrated).

How accurate is then the theoretical list?

Never losing, not even drawing against < 2000. And i'm just 300 points
above it.

And now understand well that i am a smart guy, not a computer which has
no intelligence.

Learning a computer swindle is as difficult as learning a computer
drive a car and i have not seen any that can do this.

I am no physist of course, but already we see last years all kind of
troubles that hardware manufacturers had to make faster CPU's.

There are real natural limits here. electricity doesn't go faster than
1/3 of the lightspeed. Light computers can be only 3^2 = 9 times faster
or something than electricity computers.

In 1997 the fastest 64 bits machine ran 767Mhz at world champs.

Fastest Alpha is like 1 Ghz now, 5 years later?

Where is that 'factor 2' each 2 years lemma?

They got fast to the speed of electricity, but how do they continue
now? All kind of tricks they got stuff faster with. p5 did 2 instructions
a clock. P6 did 3, but now the disadvantage is clear too. HUGE
branch mispredictions. SMALL l1 data caches compared to what i would expect
P4 should have.

I am not hardware expert here but i really see the speed of the things
go slow down. The tricks used to get the programs faster are not
pure speed tricks.

The new tricks are more processors at one chip and such. But to get
from the 100k nps i get now to 10^30 NPs, that's more than a few tricks.

getting hardware 100 times faster than it is now, it is going to be hard,
but i believe they will manage, but then, 100 times faster is only 10^2?

10^5 x 10^2 = 10^7 that is not near 10^30 yet.

I'm missing lightyears of speed there.

like a factor 10^23 = 100,000,000,000,000,000,000,000 times

Also the prices for the machines to produce the new processors, they
go up each time if i am not mistaken, way faster than inflation of
money goes.

Another thing, How is an engineer going to make chips with a trillion
transistors?

We still didn't talk about the software problems, because sofware always
goes slower than hardware.






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.