Not logged inRybka Chess Community Forum
Up Topic Rybka Support & Discussion / Rybka Discussion / empirical estimate of God's elo rating
1 2 3 Previous Next  
Parent - By BB (****) Date 2008-01-03 19:18

> Not from Uri! :-) :-)


I'm surprised that there hasn't been a claim that God knows less than nothing - "In the quantum world, we can know too much ... and it is in these situations where one finds negative knowledge."
Parent - - By 8lrr8 (***) Date 2007-12-30 13:59
well, to be God is akin to having 32-men TB, *and* knowing which line is the sharpest in any given position.  this is something we may never be able to emulate.

for the record, i am making the claim:

under true classical time controls, there will never exist a chess program (mated to any type of hardware) that can achieve an EV/game of 0.82 against: rybka 3.0 (64-bit), w/ a top-notch opening book, partial (i.e. most commonly encountered) 7-men EGTB, on a 5ghz octal (assuming 80% MP scaling efficiency) system.
Parent - - By Vasik Rajlich (Silver) Date 2007-12-30 14:05
Ok, you're on record. While we wait, you can browse a bit here :)

http://en.wikiquote.org/wiki/Incorrect_predictions

Vas
Parent - By Felix Kling (Gold) Date 2007-12-30 14:23 Edited 2007-12-30 14:25
Funny how much of this nonsense is out in the internet :)

My favourite is the bad prediction:
"They couldn't hit an elephant at this distance"
Last words of Gen. John Sedgwick, spoken as he looked out over the parapet at enemy lines during the Battle of Spotsylvania in 1864.
Parent - By richbell (**) Date 2007-12-30 14:26

> Ok, you're on record. While we wait, you can browse a bit here :-)
>
> http://en.wikiquote.org/wiki/Incorrect_predictions
>
> Vas


That was really funny Vas :-)
Parent - By 8lrr8 (***) Date 2007-12-30 14:28
i'd love to be wrong, fwiw.  but this is chess, not Go.  it's far more drawy.  :-)
Parent - By vroger007 (**) Date 2007-12-30 16:00
incorrect predictions: I think this interesting wiki page should be at least ten times larger in size.

Anyway, on the other hand I am even more (on the negative side!) impressed by preditions that HAVEN'T been realized,
mostly on account of an ill-justified "lack of money".

For example, back in the late sixties, when I was about 11 years old, a science chapter in a children's magazine claimed that by 1982, most - if not all - cities would have plexiglass covers and artificial climate!  But today, almost on the last day of 2007, when I take the bike to drive to the office, I have to take a second pair of trousers with me because of the numerous rain-filled pits on the poorly constructed "bicycle sidewalk".

Not even 8 years ago, on the edge of the 21st century,  local government decided to re-start making use of tramways to connect suburbs and villages with the centre of Antwerp, that "'big metropole" (which is nearly empty after 21 o' clock btw).  They claimed it would only take 4 months.  It took them about 3 or 4 years, and they had to do the entire job twice because no one knew how to lay the clinkers such that they didn't shake loose each time a tram passed.

That's why I laugh at predictions which are made today, because in spite of so called increased technological knowledge, a lot of basic knowledge has been lost by the exaggerated outsourcing to LCCs and globalisation which costs much more money than the claimed economical profit  (profit for who?).
I dare to claim that by 2082, cities will still not have plexiglass covers, and that chances are higher that people of 2082 will be living according to 1928 standards!

But I do hope that Rybka 7.0 will have 4000+ Elo :-)

Roger
Parent - By Uri Blass (*****) Date 2007-12-30 12:47
No need for 80% at some point white is going to never lose games so you are going to get 100 elo simply by winning part of white games
and by doing it again and again you may get 4000 if you choose the right opponents.

It is clear that rating is dependent on the opponents in chess.

Uri
Parent - By billyraybar (***) Date 2008-01-02 22:05
I doubt that Rybka 2.3.2a on a quad could draw more than a couple of games out of a hundred against God. He has among many other things a very good understanding of contempt  Vas

But Rybka 3 on the other hand .....
Parent - - By skulibj (*) Date 2007-12-30 12:44
Why do you think Rybka can get a draw against God?
Parent - - By Felix Kling (Gold) Date 2007-12-30 12:49
because we think chess is a theoretical draw and there's a probability to make allways good (i.e. not losing) moves, which is not that bad for Rybka :)
Parent - By skulibj (*) Date 2007-12-30 12:54
But God can think up quite deep traps :)
Parent - - By skulibj (*) Date 2007-12-30 13:16
If someone sees 30 plys and god can see much more, you think that it is not always possible to think up deep enough traps? Or if there are positions where it is not possible they can't be avoided? What I think is that God would be able to trick Rybka and everyone. So my estimate is infinite.

You guys have seen ridiculous wins from table bases. It would be interresting to see a chess program play against tablebases from a drawn but really tricky position. That could be a clue about the outcome of a match against God. Because God should always be able to get into really tricky positions isnt that true?

Im not saying that chess isnt theoritically a draw. I think it is.
Parent - - By Uri Blass (*****) Date 2007-12-30 13:29
Your example is a bad example.
When there is more than one right move tablebases do not give you the move with the best chances
so drawin against tablebases maybe easy because the program that use them may choose a move that force stalemate instead of correct move that give the opponent to work hard.

Another point is that I am not sure if god is able to play chess.
It is possible that god is dead and can do nothing.

Uri
Parent - By skulibj (*) Date 2007-12-30 13:36
That is true. Lets just put in a theoretical perfect chess player instead then :)

Very true what you say about my tablebase example.
Parent - - By ebutaljib (****) Date 2007-12-31 11:54

> It is possible that god is dead and can do nothing.
>


Oh no!!! Who is then watching over this messed up universe???  :)
Parent - - By Uly (Gold) Date 2007-12-31 20:46

> Oh no!!! Who is then watching over this messed up universe???


Maybe God created Karma and Dharma, and left.
Parent - - By turbojuice1122 (Gold) Date 2007-12-31 21:54
Ahh, very interesting--a Hindu Deist perspective! :-)
Parent - By Uly (Gold) Date 2008-01-01 00:31
Well, I try to be in some sort of "Openmindism", in where I think that everything is possible.
Parent - - By Uly (Gold) Date 2008-01-02 21:08

> Another point is that I am not sure if god is able to play chess.


I'm not sure if God'd want to play chess. Try this:

Set up a chess board, with a clock and everything ready. Make your first move as white and then pray to God to play black's first move.

Chances are that God will forfeit the game on time.
Parent - - By turbojuice1122 (Gold) Date 2008-01-02 21:59
Well, that's certainly not the proper way to play against God!  We must have a prophet to communicate the moves...

Actually, I think that I may have one played against God on the Internet Chess Club.  I played 1.e4, and my opponent immediately resigned!
Parent - - By Uly (Gold) Date 2008-01-02 22:13

> We must have a prophet to communicate the moves...


But, the devil may get in the way, and then we end playing a game against Satan. I'd wonder what's Satan's ELO rating, though.
Parent - By turbojuice1122 (Gold) Date 2008-01-02 22:15
Well, you can create a poll, if you'd like!
Parent - - By ebutaljib (****) Date 2008-01-02 22:23

> But, the devil may get in the way, and then we end playing a game against Satan. I'd wonder what's Satan's ELO rating, though.


Definatelly lower than god's ELO, because he lost against him

http://www.chessbase.com/newsdetail.asp?newsid=2833     :)
Parent - By Uly (Gold) Date 2008-01-02 22:48
Some things:

On my belief, you can't surprise God ;) and since with her/his perfect timeless knowledge, it can beat you in the least possible moves, it seems you can't beat the devil in less than 27 moves.

However, it's disappointing that the devil can't play perfect chess, that means it's beatable, and so it's weaker than the strongest player conceivable.
Parent - - By 8lrr8 (***) Date 2007-12-30 14:27
the draw probability increases as the TB man size increases (i.e. 3-men EGTB vs 6-men pieces).  furthermore, the super long forced mates are of combination of pieces that never appear in actual games.  so making conclusions about white's advantage from TB studies is ill advised.

white can almost certainly force a draw.  so the question that remain is: is white's opening advantage large enough such that "God" can win almost all the time against the latest version of rybka mated to monster hardware?  or to put it another way: is rybka searched to, say, 34-ply dramatically stronger than rybka searched to, say, 26-ply?  is the draw window in chess really that narrow?
Parent - - By skulibj (*) Date 2007-12-30 17:50
"the draw probability increases as the TB man size increases (i.e. 3-men EGTB vs 6-men pieces)"
really? do you have some info about that?
Parent - - By 8lrr8 (***) Date 2007-12-31 02:25
not explicitly, no.  but i think it's true if u take all 3-piece mates (their initial positions), divided by all possible 3-piece combinations, that probability will be larger than their 6-piece counterparts.  perhaps someone who has the complete 6-piece EGTB can give us the actual #'s?
Parent - - By skulibj (*) Date 2007-12-31 12:44
Are you talking about the probability of a random legal position being mating?
Parent - - By 8lrr8 (***) Date 2007-12-31 17:11
yes.  even if u take only the positions that have shown up in actual games, i believe my claim still holds.
Parent - - By skulibj (*) Date 2007-12-31 17:18
But even though the probability goes down the number of possible combinations might rise more rapidly. But then again the ability to choose moves... this is a hard question.
Parent - - By 8lrr8 (***) Date 2008-01-02 01:14
"But even though the probability goes down the number of possible combinations might rise more rapidly."

huh?  this makes no sense.  all that matters is the likelihood long forced mates actually show up in real games (rather than contrived positions) more often w/ many pieces still left on the board vs. when fewer pieces are on the board.  like i said before, i dont think this is true.
Parent - By Banned for Life (Gold) Date 2008-01-02 01:34
As the dimension of the TB increases, the average and longest distances to mate increase as well. Good luck finding these forced mates when DTM is greater than 100.
Parent - - By Lukas Cimiotti (Bronze) Date 2008-01-02 22:37
Shouldn't this thread be moved to the new OT forum, too? - i mean The Flip Side?

I never played a game of chess against god (would be kind of difficult, as I don't believe in any kind of god or any Supreme Being), and I don't think anyone here has ever done so.
Who ever played a game of chess against god, please tell us. ;)
If anyone claims he has done so, I'd like to hear what kind of psychotropic substance(s) he used.
Parent - By Uly (Gold) Date 2008-01-02 22:44

> Shouldn't this thread be moved to the new OT forum, too? - i mean The Flip Side?


I don't think so, since Rybka has been mentioned now and then :)
Parent - - By turbojuice1122 (Gold) Date 2007-12-30 15:51
I gave an argument in a different thread that lead to a result of 3850 for God's elo.  I think that the errors in my assumptions are larger than the resulting change in the number from a good correction for such errors.  The particular thread in question is one that some people would prefer to forget, so instead of linking to it, I'll simply copy and paste my argument here: (and then wait for the rebuttals that I'm sure will come).

I think we currently have enough evidence to say that perfect play will certainly not result in a near-perfect score against today's top chess engines.  My guess is that it would result in scores around 95% or so against Rybka and Zappa Mexico (I originally pulled that number out of thin air, but now I see that it happens to be roughly 500 points of elo difference on today's scales).  The decided games in Mexico occurred when one or both engines "took risks" in the opening and/or game phase.  This doesn't always need to occur: game 6 is a good example.  One thing at which Zappa is excellent is drawing a drawn game; however, since its style is extremely different from Rybka's, and I predicted that this would occur based on watching games before the match, the number of decided games in that match was about the most that you will ever see between opponents at that level in match conditions.  Anyway, here is where I think is the problem with the idea that, say, a perfect chess-playing entity will have a near-perfect score against, say, a 3400-rated chess entity.  First, a 3400-rated chess entity will have a score of 85% against Rybka and Zappa Mexico.  Fine.  The point at which I become very suspicious is the notion that the same entity that is able to score 95% against Rybka and Zappa Mexico will also be able to score 76% against the 3400-rated chess entity.  I just don't see this happening--even if you had Erdo and Dagh/Jeroen play their two representative engines against each other and constantly fix the opening books of these 3100-rated engines, you would, after awhile, change the course of the game toward a sort of draw threshold.  This already changes things significantly, and the 3100-rated teams will be playing moves that, while possibly not the moves that give the "best" winning chances in the position, are still not far enough away to change the perfect, objective evaluation of the position.  To do that, they would probably have to produce a "sum of mistakes" in their random walk that equates to around +/- 1.00 in the evaluations of today's best engines (to wit: Rybka and Zappa).  When one of the opponents is perfect, this reduced to a one-directional "random" walk.  Today, if we talk about 50% of the games being decided at the 3100 level, it means that this threshold is crossed (and kept!) 50% of the time.  Pulling a number out of thin air (it turns out it doesn't matter WHICH number I choose, as long as it's small enough), let's assume that the "average" mistake is worth 0.20 units in the evaluation.  Thus, by the random walk statistics, the programs combine to make SQRT(N) mistakes, each worth 0.20 units, to get to this number.  Here, we see that N = 25.  Thus, each program makes roughly 12 mistakes, each worth 0.20 units.  Now let's look at a program that makes only 5 of these mistakes, therefore a program that will lose half of its games against a perfect chess-playing entity.  It makes such a mistake 5 times for every 12 times that the first program makes such a mistake.  With these being approximately Poisson statistics, the good program makes 5 +/- 2.23 mistakes per game, and the "bad" program makes 12 +/- 3.46 mistakes per game.  Thus, we have that on average, the good program wins a little bit over 95% of the games, say, 96%.  This is the same program that is able to score 25% against God.

Thus, I have basically just calculated that God has an estimated Elo of 3850.  I see this as an upper limit due to the opening book factors I mentioned above.

It's natural to want to look at my assumptions, in particular the one about an "objective" evaluation of +/- 1.00 meaning a decided game.  This might not actually be correct.  However, I do think that it's probable that only 25% of the moves made by each side in games between Rybka and Zappa Mexico are on the order of one-fifth of the amount required to throw the game.  While the rest certainly aren't perfect, most of the other mistakes would probably be of a noticeably smaller magnitude in actuality, perhaps a few hundredths of a pawn.  Lastly, we must worry about things like ...f4?? that blew game 9 completely.  I simply don't believe that a 3400-rated player will ever make such moves.
Parent - - By Mark (****) Date 2007-12-30 19:33

> I think we currently have enough evidence to say that perfect play will certainly not result in a near-perfect score against today's top chess engines.


I really disagree with the above statement, and it is the basis of your whole argument.  What evidence do we have?  Just that some of the best programs tend to get more draws against each other when they play on better hardware?  I don't believe that Rybka or any of the other programs could get a single draw in thousands of games against a perfect opponent.  Chess is just too deep and too complex.  Even the best chess players using the best programs many times can't even figure out what the losing move was when analyzing games.

Regards,

Mark
Parent - - By turbojuice1122 (Gold) Date 2007-12-30 19:37
Very simply, some drawing moves produce more winning chances than others.  Kramnik's drawing moves tend to decrease chances for both sides, while Kasparov's drawing moves (when they are such) often tend to increase chances for both sides, and since he was the best chess player ever, he was better able to wade through the complications and produce positive results than anyone else.
Parent - - By Hetman (*****) Date 2007-12-30 20:09
You have forgotten that Rybka is runing on the hardware, if it fails Rybka is losing :-). There is either the possibility that programm hangs and it looses ;-( .
Rgds
Hetman
Parent - - By turbojuice1122 (Gold) Date 2007-12-30 20:15
Thankfully, such "losses due to hardware" are more uncommon now than they used to be.  I am going to assume that rules in such matches would be put into place where God isn't allowed to disrupt the power flow to the CPUs :-).
Parent - By Hetman (*****) Date 2007-12-30 20:27
too many limitations  :-( .

even in the WCC matches it is not forbidden to influence psychologically the opponent :-)
Parent - - By 8lrr8 (***) Date 2007-12-31 06:39
"Thus, I have basically just calculated that God has an estimated Elo of 3850.  I see this as an upper limit due to the opening book factors I mentioned above."

but see, the thing is God may not be able to score +600 against a 3200 entity, due to: white's advantage, opening book strength (particularly if one was created w/ the objective of playing for the draw as black), and the nature of the elo rating system.
Parent - By turbojuice1122 (Gold) Date 2007-12-31 06:45
Yes, I think this (an argument Larry also just gave) helps to balance out the ideas about God potentially using omniscience concerning future opponent moves (a point whose existence is debatable, even in principle) to choose His own moves based on those that would place the opponent on the most treacherous path to keep God's elo below or at least close to 4000.
Parent - - By lkaufman (*****) Date 2007-12-30 15:55
    The problem  with this question, I believe, is that everything depends on the rating difference between the opponents. It is quite possible that the current rybka with a good enough book can draw pretty often with White against anyone, assuming a book written expressly for this purpose. If you will all your games with White and draw all your games with Black against someone, you end up with a bit under 200 more points than him. So if Rybka is 3100, then it might be possible to create a book that will limit any future opponent to around 3300 if they are always paired. But if a succession of stronger opponents are created (as they surely will be), each one can easily outrate its predecessor by 50 points or so by drawing with Black and getting some wins with White. As long as pairings are limited to close ones, this can go on almost forever, until the way to draw with Black is clear. In this fashion, it is easy to foresee a 4000 rating as being possible, as long as the 4000 rated player doesn't have to defend it against mere 3600 players for example.
     Also, the current CCRL and CEGT ratings used short, randomized opening books, and on this basis it is much easier to imagine a 4000 rating than would be the case with 30 move deep opening books.
Parent - - By skulibj (*) Date 2007-12-30 17:35
How does Rybka score against 10 year old chess programs?
Parent - - By lkaufman (*****) Date 2007-12-30 20:38
I just ran off fifty bullet games on my quad between Rybka Paderborn version and Fritz 5.32, using 25 randomized openings with each playing one white and one black. The score was 46 wins for Rybka and four draws. This works out to a 552 rating differential. Fritz 5.32 is rated 2608 on the CEGT blitz rating list, so this would indicate a rating of 3160 for Rybka Paderborn quad. This is completely consistent with taking the CEGT blitz rating for Rybka 2.3.2a dual and adding 35 points or so for quad and another fifty for my estimated improvement of Rybka Paderborn over Rybka 2.3.2a. In fact, the agreement is almost perfect, which is obviously just a coincidence. Earlier tests indicated that Rybka was a good match for Fritz 5.3.2 in blitz with Rybka giving knight odds; maybe I'll rerun that test now.
Parent - - By skulibj (*) Date 2007-12-31 12:51
Interresting. Are the drawn games interresting?How did Frits 5 manage a draw?
Parent - - By lkaufman (*****) Date 2007-12-31 15:40
Of the four draws, three were endgames with Rybka having an extra pawn which just wasn't enough to win for various reasons. The other was even all the way until the end (king and queen each only).
Parent - - By skulibj (*) Date 2007-12-31 16:07
Have you results against even older chess programs?
I must say I am a bit surprised by those draws. But they make perfect sense considering the elo difference. It just sounds strange if Rybka makes draw against Fritz 5. 10% chance.
Parent - - By lkaufman (*****) Date 2007-12-31 16:20
No, although I'm running a knight odds match with Crafty 19.01, which I believe is a bit newer but a bit weaker than Fritz 5.32. Bear in mind that I didn't set a contempt factor for these games. When Rybka wins a pawn she normally tries to trade down (as all the books recommend), which is usually correct but occasionally just leads to a draw. Being able to tell when an extra pawn wins the ending and when it doesn't is rather difficult. Rybka would probably score  even better against such old programs without the incentive to trade when ahead, but she would score worse against close rivals.
Up Topic Rybka Support & Discussion / Rybka Discussion / empirical estimate of God's elo rating
1 2 3 Previous Next  

Powered by mwForum 2.27.4 © 1999-2012 Markus Wichitill