> http://www.youtube.com/watch?v=zG9vEgITUOg

Look like future chess player  :)

Why not?
I asume: they will reach 4000 ELO.
But there will be the Question: what is such a '4000 ELO'-value.
It is a calculated value, based on a fiew thousand Engine games. OK.
And no one will be able to say, what this has in detail to do with the strength of strongest humans.
"Much stronger, anyhow!" OK!

"We had a strong storm, much stronger than normal Beaufort 12. It must have been wind-strongness 15!"
15? :-o

Quap

Until now, most chess programmers improved their engines, version by version. That is true, but I doubt that an Elo rating of 4000 will ever be assigned. That was recently discussed here, and we too a look and the official FIDE rating performance tables. There you see which performance (in percent) is required for a particular Elo performance difference.

http://www.fide.com/component/handbook/?id=75&view=article

%   diff.
.99  677
.98  589
.97  538
.96  501
(etc.)

So, for a 4000 Elo perf., a player would need to score 96% against opposition with ~3500(!!) average, or 99% against 3323 :-)

Against good opponents, even for a vastly superior engine it is unlikely that such percentages can be achieved, due the draws. For example, 96% requires total results like +23 =2 -0. For comparison, In 1105 CCRL games, Quad-Rybka 3 has drawn 403 games. And that is already a small share of draws, relative to the next competitors.

http://computerchess.org.uk/ccrl/4040/

Engine rating lists have many opponents with very high ratings (far beyond the top human ratings, already). At CCRL, Quad-Rybka 3 scored 76.9% against an opponent's average which was 190.1 points lower rated =3035.9, and got 3226. (The rating system there apparantly assigns slightly different preformance differences, but very similar.) If the opponents became clearly stronger, like 3100, then 85% would give ~3400. So, both with even stronger opponents AND an even better percentage which a new version might achieve, 3500 seems out of reach, let alone 4000...

> Even if a 32 EGTB chess entity was available it would not score 4000 ELO

Would there be a draw between a 32 EGTB chess entity and, say, a 3200 Elo engine? :-)

Ah, but if you had a 32 EGTB chess entity that ALSO was able to emulate the opposing engine and foresee with great exactitude likely losing moves....  :)  Then you would have true omniscience!

Hi

Wihle I agree there must be a wall and that many possibilities may occur to draw a game the number of 3400 seems low to me.

Already today on a simple Q6600 Rybka 3 should have an ELO of around 3200.

If we take next Nehalem 8 core machines with 2 sockets, that is 16 cores.

I assume such a machine OC is around 5x faster than the Q6600. If doubling the speed is around 60 ELO => around 150 ELO gain => 3350

Do you think now if Vas improves Rybka 4 by 50 ELO he can stop programming and retire because the wall is there ?

I dont think so and dont think we should underestimate chess.

my figure would be around 4000 ELO because we today have no clue what does it mean...

regards

We are basically talking about "Perfect" chess at the 4000 ELO level.A 32 EGBT chess entity would never lose and would be playing in effect "Perfect Chess".And due to the draw ratio two entities of this magnitude would not likely reach 4000ELO unless they are "paired" down constantly that is if they are playing lower competition say around 3000-3200 ELO opponent.This brings the question to mind what is the ELO of "Perfection" is it 4000,6000 or will perfection cancel out such a huge ELO due to draws?

This theory will be disproven very soon :)

Vas