Topic The Rybka Lounge / Correspondence Chess / Are the following games the strongest games played to-date
![[us]](/mwf/v12/flags/us.png "United States")
Are the following games the strongest chess games played to this date in terms of the quality and strength of the moves (consider all computer chess games in the past, human supergrandmaster play, freestyle chess play etc.) :
http://www.iccf-webchess.com/EventCrossTable.aspx?id=25887
?
http://www.iccf-webchess.com/EventCrossTable.aspx?id=25887
?
![[es]](/mwf/v12/flags/es.png "Spain")
73 draws out of 75 games, sure looks like it.
![[mx]](/mwf/v12/flags/mx.png "Mexico")
Wow, with 97.33% draws it feels as if perfect chess was around the corner.
![[gb]](/mwf/v12/flags/gb.png "United Kingdom")
no more than i expected,if the game is a draw.We are finding this on WBCCC 2012.Thats why i forecasted 7.5/12 would win tournament.
Yes, I think that with current hardware and software resources, people already fully capable of playing perfect chess (turbojuice believed this back in 2007 with Rybka 3 + IDEA, I believe.)
![[de]](/mwf/v12/flags/de.png "Germany")
What do you mean by 'capable'?
That some player with some specific analysis method is able to play a perfect game every single time. That is, no opponent would be able to reach a position in where this theoretical player would make a game losing blunder (that is, I'd allow this theoretical player to make, say a -2.00 blunder in where all engines think he's in trouble, but I'd still consider it perfect play if he manages to save the game.)
Aha. I think we are really far away from this point.
Do you think that in most of those 73 games that ended in a draw players made game losing blunders that the opponents failed to punish?
My claim is the opposite, that if we had 32men tablebases to check those games, most would show that in none of the positions there was a mate in X that the opponent couldn't follow.
That we cannot know they're perfect doesn't mean they're not perfect.
My claim is the opposite, that if we had 32men tablebases to check those games, most would show that in none of the positions there was a mate in X that the opponent couldn't follow.
That we cannot know they're perfect doesn't mean they're not perfect.
> Do you think that in most of those 73 games that ended in a draw players made game losing blunders that the opponents failed to punish?
No, I think most of these games were perfect in the sense of always in the draw range.
However, this does not mean that much. Even many of my OTB games are perfect in this sense, although I am thousands of Elo points below perfect chess. Good players are better at provoking mistakes, that's why I would rarely play a perfect game if I were to play against 2700+ players.
Think of it this way: Suppose a strong correspondence player has to play with an engine, hardware and opening book from 10 years ago. I am sure you agree that he would lose quite frequently if he were to play in a strong tournament in 2012. Now you are basically proposing that if you provide that same player with top equipment from our time, he would never lose against someone from 10, 50, 100, or 1000 years in the future. This does not seem credible to me. It is a common pattern to believe that the present is a very special point in history in some respect. But what reason is there to believe this?
Remember how Vas said that he tested Elo gains from doubling the time spent at different depths? He indicated that the gains are very stable even at high depths. So unless you believe that the Elo gain will suddenly drop to zero at some particular depth, this seems to indicate that there is still a long way to go until we reach perfect chess.
> Now you are basically proposing that if you provide that same player with top equipment from our time, he would never lose against someone from 10, 50, 100, or 1000 years in the future.
Yes, my claim being that a player from today could be able to produce perfect moves on the fly, despite attempts of the opponent of "provoking mistakes". Since the game is draw with perfect play, the only thing a player from 1000 years in the future could do is going for a complex position in where the best move is really hard to find, and "hope" that the player of today misses it. That is, the player from the future, knowing that any try he makes could be futile (because he knows there's a move that refutes his plan, and even knows what move is it) would have to resort to hope chess.
If course, this would require the player from the present to ensure and play for a draw, and I could give that if the opponent from the future played a losing move in purpose just to make it very hard for the player to find the correct continuation, that the player from the present would decline the blunder and keep playing for the draw instead, playing imperfectly but still not losing.
And, I had a player in mind when posting that, he's turbojuice1122
, he has been running an IDEA tree since 2007(?) and has all major openings covered (?), I think that there could be some super-drawish variation he could play against any opponent of the future, so that, at 48 cumulative hours per move, he would not lose any game (if his wife allowed him to play, but since we have people 1000 years apart playing, let's say you offer turbo $100'000 if he doesn't lose any game). And I'll let turbo correct me if I disagree.>So unless you believe that the Elo gain will suddenly drop to zero at some particular depth, this seems to indicate that there is still a long way to go until we reach perfect chess.
Elo is broken anyway and it fails to predict outcomes properly once the elo becomes too high or a player becomes perfect. Uri Blass has posted thoughts on this:
"It is possible that at some level you simply do not lose with white so A can score 75% against B(by winning with white and drawing with black)
and B can score 75% against C(by winning with white and drawing with black) when A can score only 75% against C.
Rating is simply meaningless because if You play only A against B and B against C you get higher rating difference between A and C
relative to the case that you play only A against C."
But, anyway, yes, I have the idea that at some point in the future, strength reaches a point of "self refutation", in where the moves that defend perfectly against your attacks are evident, and chess engines rate 1.e4, 1.d4 and 1.f3 as 0.00. At this point, what do you play? Do you bother aiming for a Sicilian/Ruy Lopez or Queen Gambit/Indian, and what difference does it make if the engine can already show you what moves the opponent could play to force a draw?
Nowadays a player may go for a variation because he sees that if the opponent plays natural looking moves the evaluation jumps to 1.00 and he understands why, and he sees that the 1.00 position is dangerous and the opponent must be very careful. But if the 1.00 position is drawn with perfect play, engines of the future may score it correctly as close to 0.00, so the player of the future doesn't go for it and the extra knowledge doesn't help him to win more against players of today.
Also, comparing the level of today to the level of 10 years ago, to the strength of today to the one 10 years in the future doesn't work, as 10 years ago we didn't have strong corr tournaments ending with 97.33% draws. I don't claim that we had those resources 10 years ago, but I claim it's possible we do now.
So what is your argument exactly? The only reason you give to support your claim is the extraordinary draw rate in that one tournament. (Note that it is not 100%, so we already know that not all games are played perfectly.) There have been human tournaments with very high draw rates, which has even led some to talk about the draw death of chess, and so on. But we know that these players are really really far away from protection.
I would really like to offer you a bet (just like I was tempted to offer turbo a bet when he claimed that Rybka 3 + top hardware + top opening book is unbeatable. Do you still think that, turbo?). I'm only afraid that no-one will care for that bet in 10 years time.
I don't understand your 'Elo is broken' argument. If depth 28 is 50 Elo stronger than depth 27, this means that depth 27 loses games. I don't see the relevance of Uri's hypothetical scenario to our case.
I would really like to offer you a bet (just like I was tempted to offer turbo a bet when he claimed that Rybka 3 + top hardware + top opening book is unbeatable. Do you still think that, turbo?). I'm only afraid that no-one will care for that bet in 10 years time.
I don't understand your 'Elo is broken' argument. If depth 28 is 50 Elo stronger than depth 27, this means that depth 27 loses games. I don't see the relevance of Uri's hypothetical scenario to our case.
>So what is your argument exactly?
That with current software and hardware resources, in a time control of 48 hours per move a player can avoid playing game losing blunders in all positions that are drawn with perfect play.
Now, you've got to accept that at some time control a player can achieve perfection. Suppose, that you think that in 1000 years players can play perfectly on the fly, and that a player of today can survive 20 moves without playing a game losing blunder, then, a time control of 50 years per move would be enough to play perfectly, as by the time this player is done with the 20 moves, he can already play perfectly. I know this is farfetched but it's just an example. At what time control do you think perfection can be achieved today?
How many moves from the opening position do you think a player can play perfectly today, and what makes you think these positions are radically different from other positions? (so that such a player cannot play them perfectly anymore.)
> If depth 28 is 50 Elo stronger than depth 27, this means that depth 27 loses games.
No, one can have higher elo than the other just because one wins more games against weaker opposition than the other one, even if both draw all games against each other, and none lose.
> Now, you've got to accept that at some time control a player can achieve perfection.
Yes.
> At what time control do you think perfection can be achieved today?
I have no idea.
In your scenario, a player can upgrade hardware and software while playing, which is quite weird.
> How many moves from the opening position do you think a player can play perfectly today
I'm not sure this question makes much sense. In a normal case, or in the worst case, against what kind of opponent, ...? But anyway, I cannot sensibly answer any of these questions.
> and what makes you think these positions are radically different from other positions?
There is no single answer to the question in which positions one is not able to play perfectly. Mostly, it is worse positions, of course.
>> If depth 28 is 50 Elo stronger than depth 27, this means that depth 27 loses games.
> No, one can have higher elo than the other just because one wins more games against weaker opposition than the other one, even if both draw all games against each other, and none lose.
Now I see what you mean. I believe these numbers referred to self-play. (Apart from that, Uri's scenario way purely hypothetical.)
I am still wondering why you believe that today's players can play perfectly. I just do not see the evidence.
>In your scenario, a player can upgrade hardware and software while playing, which is quite weird.
Not at all, it happens all the time, people aren't restricted to using the same hardware they started the game with, they can upgrade while the game is happening. In fact, I had many games going that I started with my 1core computer, and half-way I upgraded to my current 4core. If a game or match of games is going to last years, it's expected that the players in it do upgrade their hardware and software (and since software is released all the time, people upgrade it while the game is running on a regular basis for correspondence chess).
>I'm not sure this question makes much sense.
Let me rephrase it then. You are saying that a player from a distant future is certain to win a game against a player of today that does his best to draw. So, if both play the best moves, the future to ensure a win in the least amount of moves, and the present to draw as soon as possible (or to play the most forcing drawing variation), how many moves (the most that it can) do you expect the present player to last?
>But anyway, I cannot sensibly answer any of these questions.
It seems when you imagine a game of present player against future player you can envision the start of the game, and the end of the game, but the middle of it is a blur, yet you have for certain that the player of the future would win.
>Mostly, it is worse positions, of course.
And I've seen players losing games by being careless or playing for a win, I've never seen one lose when playing for a draw and doing their best to avoid worse positions. That's why I think you'd need to show the opening variation that the player of the future would need to use to ensure that the player from today cannot reach equality if he plays for a draw.
>(Apart from that, Uri's scenario way purely hypothetical.)
But it's a plausible scenario once we start reaching perfection (e.g. the first weakest perfect chess entity would be one that never loses as white, but as black may play imperfectly and thus prone to losing.)
>I am still wondering why you believe that today's players can play perfectly. I just do not see the evidence.
Looking for evidence for "players of today are able to avoid blundering with current software and resources" would be akin to having an hypothesis that all swans are white, and then having to look at all animals that aren't white and ensure that none of them are swans. Much easier is finding the black swan to prove the opposite, which would be your stance: "players of today are going to eventually make a game losing blunder with current software and resources".
For this a player from the future isn't needed,and not even playing a game is needed! All you need to do to prove your point is showing a position that you know only one move draws and the rest lose, and show that other people with current hardware and resources can't find the drawing plan and would lose. Then, a player from the future would be able to reach such positions and beat players from the present. You have infinite time for this task to simulate a player from the future, while a player of today would be restricted to 48 hours.
I just do not see evidence that such positions exist. In fact, the hardest positions I've seen were solved easily by unattended engines in times much shorter than 48 hours, and I can only assume that a strong centaur would do better.
Sorry, I am honestly quite lost in the discussion. You make so many points I don't really comprehend, or regarding which I don't understand how they relate to the issue at hand. I'll try my best to address your points nevertheless, apologies if I sometimes talk past the issues you raise.
What is weird about the possibility of an upgrade during the game is just that it's irrelevant to the topic we are discussing, which I thought related to players which have resources at their disposal which are available today.
No. I am only saying that he would sometimes win.
I'm not sure I understand this. For a start, the present player will not always make the best moves (otherwise he won't lose). Most games in which the future player wins will be rather long, I would guess.
What? Which set of players are you referring to here?
I have to show the opening line? That sounds crazy. (I am a player of today, you know.) Aside from that, I would guess that there is a large number of lines in which it is possible to beat a present player.
1) I don't think so. 2) It is still irrelevant to the point I made.
No, sorry. I did not ask you to prove your point, I was asking you for evidence that your claim is true. Why believe something for which there is no evidence?
I was about to ask this anyway. So you really think that a player of today will be able to draw any position which is objectively drawn against anyone? Sorry, but this just sounds crazy. (Note that it is a much stronger claim than the one you initially defended.)
I don't have such a position handy, and I suspect that it won't be easy to find one. But anyway, the challenge is interesting. If I happen to find such a position, would you be willing to represent the player of the present and play it out?
Edit: I am sure you are aware that positions have been found in which it takes several hundred moves to win, including incredibly long series of only moves. Do you think you would be able to find such series of moves if given 48 hours time per move?
What is weird about the possibility of an upgrade during the game is just that it's irrelevant to the topic we are discussing, which I thought related to players which have resources at their disposal which are available today.
> Let me rephrase it then. You are saying that a player from a distant future is certain to win a game against a player of today that does his best to draw.
No. I am only saying that he would sometimes win.
> So, if both play the best moves, the future to ensure a win in the least amount of moves, and the present to draw as soon as possible (or to play the most forcing drawing variation), how many moves (the most that it can) do you expect the present player to last?
I'm not sure I understand this. For a start, the present player will not always make the best moves (otherwise he won't lose). Most games in which the future player wins will be rather long, I would guess.
> And I've seen players losing games by being careless or playing for a win, I've never seen one lose when playing for a draw and doing their best to avoid worse positions.
What? Which set of players are you referring to here?
> That's why I think you'd need to show the opening variation that the player of the future would need to use to ensure that the player from today cannot reach equality if he plays for a draw.
I have to show the opening line? That sounds crazy. (I am a player of today, you know.) Aside from that, I would guess that there is a large number of lines in which it is possible to beat a present player.
> But it's a plausible scenario once we start reaching perfection
1) I don't think so. 2) It is still irrelevant to the point I made.
> Looking for evidence for "players of today are able to avoid blundering with current software and resources" would be akin to having an hypothesis that all swans are white, and then having to look at all animals that aren't white > and ensure that none of them are swans
No, sorry. I did not ask you to prove your point, I was asking you for evidence that your claim is true. Why believe something for which there is no evidence?
> All you need to do to prove your point is showing a position that you know only one move draws and the rest lose, and show that other people with current hardware and resources can't find the drawing plan and would lose.
I was about to ask this anyway. So you really think that a player of today will be able to draw any position which is objectively drawn against anyone? Sorry, but this just sounds crazy. (Note that it is a much stronger claim than the one you initially defended.)
I don't have such a position handy, and I suspect that it won't be easy to find one. But anyway, the challenge is interesting. If I happen to find such a position, would you be willing to represent the player of the present and play it out?
Edit: I am sure you are aware that positions have been found in which it takes several hundred moves to win, including incredibly long series of only moves. Do you think you would be able to find such series of moves if given 48 hours time per move?
> What is weird about the possibility of an upgrade during the game is just that it's irrelevant to the topic we are discussing, which I thought related to players which have resources at their disposal which are available today.
And I assumed it includes software and hardware upgrades that happen while the game is going on. Since upgrading software and hardware is not disallowed in any correspondence game, I never thought it would be a special case. So, to clarify, my claim includes upgrades (because ALL players of today can and do upgrade).
>For a start, the present player will not always make the best moves (otherwise he won't lose). Most games in which the future player wins will be rather long, I would guess.
Define "best move". Since chess is a game that is drawn with perfect play, I wouldn't distinguish from one drawing move from another. If 1.e4 and 1.f3 draw with perfect play, both are best, there's no difference. It's only different, and e4 is better, if the white side is trying to win. As our present player only wants a draw, any move that suffices for this outcome is best, and I claim 48 hours are enough for the player to find the best move with resources of the time the move has to be made.
The onus would be on you to show a position in where the best move (any that doesn't lose) can't be found under 48 hours.
Also, I've heard this argument that all a perfect player has to do to win against an imperfect player is dragging the game long enough for the losing move to be played and then punish it, but I think that's fallacious. For instance, the perfect player could shuffle his pieces for 49 moves before forcing an exchange or moving a pawn, and force the opponent to do it by going into positions in where the opponent loses if he deviates from this strategy. This would produce a very long game that would be a draw anyway, because just making the game longer doesn't make it harder.
>What? Which set of players are you referring to here?
All the corr chess games that I've seen. You can actually debunk this by position a corr game in where one player wanted to draw, played for a draw, and lost anyway.
>I have to show the opening line? That sounds crazy. (I am a player of today, you know.)
But you'd get unlimited time to attempt it, unlike the player of today that is restricted to 48/hour move. Also, you are playing both sides for this, one side wanting to draw and the other to win, you could show weak moves of the drawing player and claim better moves can't be found in the 48/hours time frame.
At least this is something that could be done theoretically, unlike doing a real test against a player of the future.
>Aside from that, I would guess that there is a large number of lines in which it is possible to beat a present player.
And you only need to show one.
>1) I don't think so.
Why do you doubt it? How do you think elo figures will look like once we start reaching perfection and entities appear that will not lose any games? It is relevant to the discussion because my claims are based on the idea that at some point strength addition hits a block and runs into self-refutation and granularity loss (i.e. seeing e4 and f3 as equal). To think otherwise is to think that perfection will never be reached, while the number of positions that need to be analyzed may allow perfection (weakest solution to the game) to be reached as soon as year 2200.
>Why believe something for which there is no evidence?
I was pointing out you're doing the same. There's no evidence that a position exists in where the saving move can't be found in 48hours with current resources. If no such position exists, then a player from today can draw all his games against any opposition because the opposition would be unable to reach that position. Believing that such a position exists, and that a player of the future can sometimes force his way into it against a player from today seems like faith.
>If I happen to find such a position, would you be willing to represent the player of the present and play it out?
Sure. If you find a position where I can't find the saving move after 48 hours, then you prove such a position exists and then my whole argument falls over, since there's no evidence that would say a player from the future would be unable to reach those kind of positions.
> I am sure you are aware that positions have been found in which it takes several hundred moves to win, including incredibly long series of only moves. Do you think you would be able to find such series of moves if given 48 hours time per move?
See above, my claim only includes avoiding losing. I explained it in this post, the part "If course, this would require the player from the present to ensure and play for a draw, and I could give that if the opponent from the future played a losing move in purpose just to make it very hard for the player to find the correct continuation, that the player from the present would decline the blunder and keep playing for the draw instead, playing imperfectly but still not losing", means my claim only carries to being able to find the draw in a losing position, not to finding a win if the future player made a game losing blunder.
> Define "best move".
I was using the expression in the same sense as you.
> How do you think elo figures will look like once we start reaching perfection and entities appear that will not lose any games? It is relevant to the discussion because my claims > are based on the idea that at some point strength addition hits a block and runs into self-refutation and granularity loss
The reason why I said that it is irrelevant is the following: If entity A2 has an Elo advantage over A1 in self-play, this entails that A1 does not play perfect chess.
I don't know if this is relevant, but I guess you would accept that a top engine of today plays perfect chess at 200 hours/move or so. I strongly doubt that an engine will score f3 and e4 as equal after 200 hours.
>> Why believe something for which there is no evidence?
> I was pointing out you're doing the same.
So you admit that you have no evidence for your claim?!
I have inductive evidence, for example.
> my claim only includes avoiding losing
Sure. But don't you think there are tablebase positions in which you have to find similarly long and difficult series of only moves not to lose?
If I'm not mistaken, then the hardest positions in the 6 piece tablebase are much harder than the hardest ones in the 5 piece bases. And so on. Now imagine the hardest positions in the 32 pieces tablebases. In my view, you really underestimate what a complex game chess is. (That latter claim was not supposed to be an argument.)
Okay. Let's see if I come back to the challenge.
> I don't know if this is relevant, but I guess you would accept that a top engine of today plays perfect chess at 200 hours/move or so
I don't think so, as I recall seeing positions in where some engine fails to find the best move (only saving move) within its depth limit (say, some engines reach depth 64 and call it a day, and the user can't analyze it further). It's easy to think that such positions exists where the depth limit isn't reached before 200 hours, but the best move isn't found anyway.
Such positions require centaur interaction, but then they may become trivial, e.g. if 1 out of 10 engines find the best move in 1 minute, an interacting monkey can find the move by switching engines after 10 minutes.
>So you admit that you have no evidence for your claim?!
>I have inductive evidence, for example.
I also have the inductive evidence of the chess games where a player only wanted a draw and got away with it. Practical examples:
natmaku vs keoki010 1/2-1/2
1.c4 e6 2.Nc3 d5 3.d4 Nf6 4.Bg5 Be7 5.cxd5 exd5 6.Qc2 O-O 7.e3 h6 8.Bh4 Be6 9.Bd3 c5 10.Nge2 Nc6 11.a3
parmetd vs Sebastian_Boehme 1/2-1/2
1.e4 c5 2.Nf3 d6 3.d4 cxd4 4.Nxd4 Nf6 5.Nc3 a6 6.Bg5 e6 7.f4 Qb6 8.Qd2 Qxb2 9.Rb1 Qa3 10.f5 Be7 11.fxe6 Bxe6 12.Nxe6 fxe6 13.Bc4 Nbd7 14.Bxe6 Nc5 15.Bb3 Rc8 16.O-O Nxb3 17.Rxb3 Qc5+ 18.Be3
tomski1981 vs Balabachi 1/2-1/2
1.c4 Nf6 2.d4 e6 3.Nf3 d5 4.Nc3 a6 5.e3 c5 6.a3 dxc4 7.Bxc4 b5 8.Ba2 Bb7 9.O-O Nbd7 10.Re1 Rc8 11.e4 cxd4 12.Nxd4 Ne5 13.Bd5 Nxd5 14.exd5 Rxc3 15.bxc3 Qxd5 16.f3 Nc4 17.a4 b4 18.Kh1 bxc3 19.Ne2 Bb4 20.Qb3 a5 21.Rd1 Nd2
In these 3 games, white was happy with a draw, and played perfect chess, and achieved the expected result. You'd need to show what black deviation would need to be played by a perfect future player to beat white (and you get infinite time to simulate it.)
>But don't you think there are tablebase positions in which you have to find similarly long and difficult series of only moves not to lose?
I think that's irrelevant, for one, people from the present can already take a look at those positions' solutions and find the saving move in zero time, and tools exist to solve positions near them (see FinalGen). Also, if those positions rely on material imbalances (say, rook and pawn beating two bishops long-term) a present player can just avoid such material imbalances, and that kind of ending cannot be reached by a future player.
Besides showing there's a material imbalance position that current players can't defend, you'd also need to show it can't be avoided, because the evidence I've seen shows that if a player sees a line goes for material imbalance, then he can avoid it from the distance. Those don't happen unless both players want them, regardless of strength (e.g. there's a deviating previous move that would allow material to remain balanced.)
>Now imagine the hardest positions in the 32 pieces tablebases.
How does the future player quantify such a position? If anything, he'd need to downgrade to software and hardware of our era to see if we would be able to find the saving line, and he can't be certain of winning sometimes due to asymmetry of analysis methods (basically, if people analyzed the same, I'd agree with you, but as they analyze differently, the future player has no idea what kind of positions the present player would be unable to solve. After all, the least amount of moves that can be played to save the game in a position, the easier it may be to find them, just look at recaptures as an example, those where there's only one saving move, but even patzers can find it.)
I still believe that there are diminishing elo returns with greater depth, and that the degree to which they diminish is asymptotic. Uly's argument concerning the draw rate is important due to the fact that one can use this to analyze what happens to chess-playing level as time increases. The greatly increased draw rate seems to imply that doubling the thinking time, say, from 1 day per move to 2 days per move has nowhere near the effect that would occur if you doubled it from 1 hour per move to 2 hours per move. This shows that we are already approaching the draw limit, even with our current technology, and have been doing so for quite a long time. I did a more mathematical analysis on this some years ago and concluded that the elo of a chess player who always plays perfectly and strategically is less than 4000 elo (I've forgotten the exact number that I estimated, though it had a large uncertainty--the main point is that it was on the order of 4000, not 5000, 10000, etc.), so I certainly disagree that you are many thousands of points below this.
On the other hand, it is clear that a perfect human player who is also extremely strong strategically will force a weaker player into positions that are more difficult to play perfectly. My statement before about playing current chess was referring to the current players and technology available. Future technology will be better not just with speed, but also with chess strategy, which is where things will change.
I'm not sure of the necessary time control for which today's top software (cluster Rybka software) plus top hardware (cluster Rybka plus top opening book (whatever that is) is unbeatable. I think that Cluster Rybka in such a situation would, if thinking for one week per move, lose very few games to a strategically perfect player (one playing perfect moves, as well as moves that attempt to make things very difficult for the opponent). What "very few" is, I don't know--if it played an opening that is quite drawish, it might never lose a game in such a situation.
On the other hand, it is clear that a perfect human player who is also extremely strong strategically will force a weaker player into positions that are more difficult to play perfectly. My statement before about playing current chess was referring to the current players and technology available. Future technology will be better not just with speed, but also with chess strategy, which is where things will change.
I'm not sure of the necessary time control for which today's top software (cluster Rybka software) plus top hardware (cluster Rybka plus top opening book (whatever that is) is unbeatable. I think that Cluster Rybka in such a situation would, if thinking for one week per move, lose very few games to a strategically perfect player (one playing perfect moves, as well as moves that attempt to make things very difficult for the opponent). What "very few" is, I don't know--if it played an opening that is quite drawish, it might never lose a game in such a situation.
> I still believe that there are diminishing elo returns with greater depth, and that the degree to which they diminish is asymptotic.
In another post, I mentioned a thread in which Vas said that he made extensive tests about Elo gains from doubling time (or, roughly equivalently, from going one ply deeper). There, Vas roughly said that he did not observe significantly diminishing returns even at very high depth. Unfortunately, he did not reveal any specifics about this test.
> Uly's argument concerning the draw rate is important due to the fact that one can use this to analyze what happens to chess-playing level as time increases. The greatly
> increased draw rate seems to imply that doubling the thinking time, say, from 1 day per move to 2 days per move has nowhere near the effect that would occur if you
> doubled it from 1 hour per move to 2 hours per move.
Why is that? As I mentioned before, there were discussions about the 'draw death' of chess due to the players' strength approaching perfection already hundred years ago. High draw rates might conceivably be a sign for engines reaching perfection, but we have seen extremely high draw rates among players who are very far from perfection.
> I did a more mathematical analysis on this some years ago and concluded that the elo of a chess player who always plays perfectly and strategically is less than 4000 elo
Based on which data?
> so I certainly disagree that you are many thousands of points below this.
I certainly am. :) Anyway, I did not make any specific claim about how far we are from perfection.
Generally speaking, apart from a couple of other issues, I am suspicious about that 'presence bias', i.e. the belief that just now is a special point in history, which people seem prone to. (I mentioned this in a response to Uly before) Precisely now is the time when perfection is achievable with the best resources we currently have. (Note that you said the same thing already a few years ago, although back then you seemed to be more confident.)
> Precisely now is the time when perfection is achievable with the best resources we currently have.
I don't think we're at a significant point, as I think at all points in time, perfection was achievable at some time control, this was just the time in where I thought it was achievable in a reasonable time. As time advances, the time control needed to play perfect chess with resources of the time gets shorter.
Okay, after reading turbo's post it seems my thoughts about reaching perfection when playing for a draw while analyzing 48 hours/move games in my 4core seems unrealistic, but here you said that unattended engines could reach perfection at 200 hours/move, and I disagreed. Perhaps we can agree that a top centaur from today with top software and hardware resources can reach perfection at 9/days [216 hours] move?
Then, the argument that it's 'presence bias' would not apply, as, say, I would claim that with resources of 2011 a player could reach perfection at 18/days [432 hours] move, and a player of 2010 could reach perfection at 36/days [864 hours] move, etc. (or some other curve; remember that as we'd allow upgrades, a game slow enough would reach its middlegame in our present) because 9 days/move is not significant and it's an arbitrary point.
> here you said that unattended engines could reach perfection at 200 hours/move, and I disagreed
Why are you repeating this? I already said that I did not mean to say this and do not think it is true. Here is what I said in the post you link to:
"I guess you would accept that a top engine of today plays perfect chess at 200 hours/move or so"
This was a guess about you, not about me.
I believe 9 days/move is still not sufficient.
I don't see a presence bias issue. We are still very far from having programs play perfectly every time against a strategically perfect player in all positions. However, I am quite sure that a skilled centaur using a computer and who wants only a draw would have a high rate of drawing success against a strategically perfect player, and that Cluster Rybka would generally be stronger than such a player. (This strength would show in having a higher success rate against weaker players.)
The data to which I came to that conclusions were the data from the various rating lists in which the draw percentage increases significantly as the time controls increase and as the level of play increases. There was a very noticeable trend in this direction and still is. Any tests that Vas has done would not have the necessary amount of data--you need something to which groups of people have contributed with most of their computing power over a period of years. Currently, the top 10 programs in CEGT 40 moves in 4 minutes average 33% draws. At 40 moves in 20 minutes, it is 40% draws. At 40 moves in 2 hours, it is 48%. With quads on software 2 years old, it was 49.3%. WBCC games decided so far: 70%, with a wide range of computer experience and a number of the non-drawn games decided by factors other than overall strength.
As for high draw rates among weaker players, the key is not taking specific players, but taking large groups of games among high-level players.
The data to which I came to that conclusions were the data from the various rating lists in which the draw percentage increases significantly as the time controls increase and as the level of play increases. There was a very noticeable trend in this direction and still is. Any tests that Vas has done would not have the necessary amount of data--you need something to which groups of people have contributed with most of their computing power over a period of years. Currently, the top 10 programs in CEGT 40 moves in 4 minutes average 33% draws. At 40 moves in 20 minutes, it is 40% draws. At 40 moves in 2 hours, it is 48%. With quads on software 2 years old, it was 49.3%. WBCC games decided so far: 70%, with a wide range of computer experience and a number of the non-drawn games decided by factors other than overall strength.
As for high draw rates among weaker players, the key is not taking specific players, but taking large groups of games among high-level players.
> We are still very far from having programs play perfectly every time against a strategically perfect player in all positions.
Okay, I guess I misunderstood you here then.
> a skilled centaur using a computer and who wants only a draw would have a high rate of drawing success against a strategically perfect player
What is a strategically perfect player? If this just means someone who wins any winning position and at least draws any drawn position, then I am inclined to agree with you. (It may not even be that difficult to draw against such a player.)
Did you try this with human games as well? I would be curious to see the draw rates among 2700+ players in classical time control.
In any case, I do not deny that draw rates between roughly equal players increase with playing strength. (In human games, for instance, it is obvious that draw rates increase with longer time controls.) I do not even deny that this has something to do with the play approaching perfection. (It is even trivial that we get closer to perfection when strength increases.) However, suppose we witness 99.9 % or even 100 % draws between the best entities. Does this mean that these entities play perfect chess, or are just about to reaching perfection? Maybe yes, but so far I haven't found a conclusive reason to think so.
> What is a strategically perfect player?
This is someone who not only always plays perfect moves, but is also extremely adept at trying to take the game into difficult positions. My position is that such a player would have to play some semi-optimal, though still perfect, moves in order to do this, and that current technology is such that one would be able to avoid losing in such situations because by playing such semi-optimal moves, the drawing path has been widened.
> However, I am quite sure that a skilled centaur using a computer and who wants only a draw would have a high rate of drawing success against a strategically perfect player, and that Cluster Rybka would generally be stronger than such a player. (This strength would show in having a higher success rate against weaker players.)
turbo, are you really talking here about correspondence games in where the strong centaur analyzes for 48 hours (or more, including pondering) per move? Because it sounds as if you're talking about 90 minute games with 30 seconds increment (or such) in where your arguments seem to be accurate.
I've never seen Rybka Cluster take at it as Corr chess, but I'd expect it would be much weaker than top centaurs, mainly because it's unassisted and not helped by other engines, so that even I, in my 4core, could defeat it by going to a position where, say, it analyzed a position as +1.00 where black is actually winning.
It's not realistic that I can do so in a short game, but at corr time controls I'd expect my chances to be fine, and to manage to be ahead of the cluster by pruning useless variations that the Cluster wouldn't while extending critical variations that the Cluster wouldn't deem as such, so that I'd expect the Cluster to pick a move on the first few hours of analysis and waste the rest of the time.
We wouldn't be talking about 40 moves in 2 hours, but about 40moves in 2months and a half.
I think that Cluster Rybka is going to be stronger than the great majority of correspondence players. It will not be as strong as the top ones, perhaps the top 10%, but it will be up there. This strength will primarily be shown by having better results against noticeably weaker players. If I recall, for example, it didn't have too much trouble in finding Highendman's 27.d5!! when left analyzing that position. However, if everyone else had the same software as Cluster Rybka, I would agree with you.
> I think that Cluster Rybka is going to be stronger than the great majority of correspondence players. It will not be as strong as the top ones, perhaps the top 10%, but it will be up there.
I think you overestimate the Cluster, or assume its strength escalates as good in corr time control as it does in <2 hours/game time controls. For instance, I think I'd perform better than the cluster at corr time controls, and I'm nowhere near the 10% of strongest players.
As I said, this mainly would be the result of low move change chances at the low end of the 48 hours, and that it'd probably play as strong (weak) as if it only had two hours per move (or 10, but the point is that it would decide on the move early on and not make a change to a better move due to branching factor. Such a move is very strong to beat a centaur at Freestyle time controls but would be easily nullified in corr chess).
i dont remember cluster finding HEM's 27.d5,i must have missed post on this.
do you honestly think that game is a ''strong chess game''????
is one of the 2 lost game of the tuournament...lossing a game in 16 moves,at least for me,it isnt to play a ''strong game'',is blunder horribily
[Event "WC27/final"]
[Site "ICCF"]
[Date "2011.6.10"]
[Round "-"]
[White "Dronov, Aleksandr Surenovich"]
[Black "João, Névio"]
[Result "1-0"]
[WhiteElo "2680"]
[BlackElo "2569"]
1.Nf3 Nf6 2.c4 c5 3.Nc3 e6 4.g3 b6 5.Bg2 Bb7
6.O-O Be7 7.d4 cxd4 8.Qxd4 d6 9.Bg5 a6 10.Bxf6 Bxf6
11.Qf4 Be7 12.Ne4 Bxe4 13.Qxe4 Ra7 14.Nd4 b5 15.cxb5 Qb6
16.Nxe6 1-0
is one of the 2 lost game of the tuournament...lossing a game in 16 moves,at least for me,it isnt to play a ''strong game'',is blunder horribily
[Event "WC27/final"]
[Site "ICCF"]
[Date "2011.6.10"]
[Round "-"]
[White "Dronov, Aleksandr Surenovich"]
[Black "João, Névio"]
[Result "1-0"]
[WhiteElo "2680"]
[BlackElo "2569"]
1.Nf3 Nf6 2.c4 c5 3.Nc3 e6 4.g3 b6 5.Bg2 Bb7
6.O-O Be7 7.d4 cxd4 8.Qxd4 d6 9.Bg5 a6 10.Bxf6 Bxf6
11.Qf4 Be7 12.Ne4 Bxe4 13.Qxe4 Ra7 14.Nd4 b5 15.cxb5 Qb6
16.Nxe6 1-0
Where can I get past games of this site?
Thanks.
Are all games long time matches using engines?
Are all games long time matches using engines?
Topic The Rybka Lounge / Correspondence Chess / Are the following games the strongest games played to-date
Powered by mwForum 2.27.4 © 1999-2012 Markus Wichitill