A 3000 could easily beat a 2000, but could a 4000 easily beat a 3000? - Chess Forums - Page 11

llama

Jun 7, 2018

0

#201

Elroch wrote:

Telestu wrote:

lfPatriotGames wrote:

What if a new computer program beats a 3600 9 out of 10 times? Wouldn't the rating have to go up?

Weak amateur human games end in a draw maybe 1 time out of 100. As the players (human or computer) get stronger, the draw rate increases. From memory (so it may be wrong) Carlsen draws about 60% of his games. Top engines can draw each other >80% of the games.

He's saying current programs are pretty close to perfect, so even a solved version of the game (a database that merely looks up whether a move is winning, draw, or losing) wouldn't be able to score 9 out of 10 against current engines.

As I have pointed out before, rating differences still lead to the same scores by definition. The difference is that the stronger player loses less and thus doesn't need to win as much to achieve a certain rating difference.

Some top computer matches have seen the extreme version of this, where all the wins are by one side.

What I mean by increased draw rate is, lets say, increasing ratings by 100 point steps.

1000 vs 1100 in a 100 game match is expected to score ~64%

1100 vs 1200 [ditto]

. . .

3400 vs 3500 [ditto]

But at each step, while 64% by the stronger player is maintained, there's a higher % of draws in the composition.

So the argument is increased playing strength will have diminishing returns, because results are tending towards 50%

All of this I'm sure you already know, but just to clarify what I was saying.

llama

Jun 7, 2018

0

#202

EndgameStudier wrote:

SmyslovFan wrote:

If you were talking about Go, which has a winner in almost every game, then the discussion of ever-increasing ratings would make sense. But in the context of chess, the conversation runs into the rules of the game.

If you ignore the nature of chess, it's easy to create a scenario with almost any Elo rating.

What rule of chess am I ignoring? The 50 move rule doesn't apply when it comes to engines.

I assume he's pointing out that in go draws don't exist and that in chess draws do exist.

In go even if you play imperceptibly better, you win. In chess you have to outplay your opponent by a fairly large margin to win.

EndgameEnthusiast2357

Jun 7, 2018

0

#203

Telestu wrote:

EndgameStudier wrote:

SmyslovFan wrote:

If you were talking about Go, which has a winner in almost every game, then the discussion of ever-increasing ratings would make sense. But in the context of chess, the conversation runs into the rules of the game.

If you ignore the nature of chess, it's easy to create a scenario with almost any Elo rating.

What rule of chess am I ignoring? The 50 move rule doesn't apply when it comes to engines.

I assume he's pointing out that in go draws don't exist and that in chess draws do exist.

In go even if you play imperceptibly better, you win. In chess you have to outplay your opponent by a fairly large margin to win.

True, in chess there is always the possibility of a draw; in Go, not so.

llama

Jun 7, 2018

0

#204

EndgameStudier wrote:

DjonniDerevnja wrote:

macer75 wrote:

EndgameStudier wrote:

DjonniDerevnja wrote:

Brixed wrote:

That's the point I was trying to make. There's a certain level, where the players are so good, that the game will be equal, regardless of the "difference". There's a finite number of moves in chess, but an infinite number of possible ratings. If one player can see a billion moves ahead, and the other can see a trillion moves ahead, the game will still probably be only 200 ish move anyway, so it doesn't really matter.

But why is 3000 elo anywhere close to that level of "good"? The Alpha-zero match demonstrated, if nothing else, that there is still a lot of room for improvement for modern engines. They aren't anywhere close to being perfect chess-playing machines, so why would 3000 - or even 4000, for that matter - be the point where someone can force a draw against perfect play?

I wish Anish Giri, one of the best ever to save a draw, could answer here. He is only ca 2800 rated, and holds draws against much, Or lets give Elias Hulleberg Sidali a try, when he was ca 1500 or 1700 he did draw an IM in a longchess tournament game, and also drew seveal other higher rated players.

This, and the puzzles I posted that the engines couldn't solve, proves engines could still be greatly improved.

I'm not trying to be rude, because I honestly don't know: I'm not sure if you guys are being serious.

Giri drew all his games in that one candidates tournament, but he wasn't at all playing drawish chess. He was drawing from good positions as well as bad positions, and playing enterprising openings.

Schlecter and Leko are good examples of draw addiction at a high level. At a somewhat lower level maybe someone like Ulf Andersson, who takes no risks, and is very strong in endgames.

But if we're talking drawing against God, why not mention someone like Petrosian, Kramnik, or Karjakin?

Giri is just one of those guys who made it to the top 10 only to discover he was not quite at their level. That's why he was drawing so much. He'd get good positions then couldn't convert, because he was simply outclassed. That's my opinion on his draws anyway... but opinion aside, his 14 draws at the candidates match were more an accident than playing like a Leko.

EndgameEnthusiast2357

Jun 7, 2018

0

#205

Again, I am just talking about whether engines can be perfected or not. Forget even the ratings. Rating only apply to normal games, not these crazy puzzles, but a truly perfect engine should be able to solve even a 1 Million Move puzzle, let alone a 15 move one.

SmyslovFan

Jun 7, 2018

0

#206

Grischuk is actually a better example. In one Candidate's match-play tournament, he strove for draws as White and often got the draw in under 25 moves. His strategy worked in the shorter matches, but when he faced Gelfand in a longer match he lost.

A 2800 player intent on drawing is very difficult to beat with perfect play. And less than perfect play runs the risk of actually losing.

All of the +2800s have several games where they didn't make any mistakes at all.

EndgameEnthusiast2357

Jun 7, 2018

0

#207

But is playing for a draw really perfect play?

llama

Jun 7, 2018

0

#208

EndgameStudier wrote:

Again, I am just talking about whether engines can be perfected or not. Forget even the ratings. Rating only apply to normal games, not these crazy puzzles, but a truly perfect engine should be able to solve even a 1 Million Move puzzle, let alone a 15 move one.

Improving engines and improving results may be more different than you think, and making the perfect engine is yet another different thing.

Making stronger and stronger engines is sort of like a goal of emptying a successively (and exponentially) larger swimming pools with buckets. Lets say the buckets you use are twice as big as mine, so you empty your swimming pool twice as fast i.e. your chess engine is stronger.

But a perfect engine is more like emptying the ocean. Your buckets may be twice as big, but after a year of dumping water, we measure ocean levels, and neither of us has made measurable progress. Getting a bigger bucket from a manufacturer or store will not help.

Meaning, sure, engines can be improved, but at some point it won't noticeably improve their results. We'd need a revolution in physics and technology before we can even entertain the idea of a machine that plays perfect chess. And while small improvements are always possible, if the pool is large enough, it wont change the results. Imagine telling me you added 1 cubic inch more space to your bucket. Yes, it's undoubtedly improvement, but it's meaningless.

llama

Jun 7, 2018

0

#209

SmyslovFan wrote:

Grischuk is actually a better example. In one Candidate's match-play tournament, he strove for draws as White and often got the draw in under 25 moves. His strategy worked in the shorter matches, but when he faced Gelfand in a longer match he lost.

A 2800 player intent on drawing is very difficult to beat with perfect play. And less than perfect play runs the risk of actually losing.

All of the +2800s have several games where they didn't make any mistakes at all.

Oh, I didn't know that.

But yes, when white plays for a draw it's nearly impossible to get any winning chances out of the opening (without accepting a decent level of losing chances). If a 2800 is playing for a draw with white it's almost hopeless.

I'm sure e.g. Carlsen (and even lesser GMs, like 2600) would score some draws against stockfish in a long match... but it might take a number of painful losses to get there, so no one wants to play or watch such a match.

chesspuzzlerjunior

Jun 7, 2018

0

#210

well if the 10000 coiuld not beat the 3000 then how would he get to 10000 as he would lose so many points for every draw

Elroch

Jun 7, 2018

0

#211

SmyslovFan wrote:

Grischuk is actually a better example. In one Candidate's match-play tournament, he strove for draws as White and often got the draw in under 25 moves. His strategy worked in the shorter matches, but when he faced Gelfand in a longer match he lost.

A 2800 player intent on drawing is very difficult to beat with perfect play. And less than perfect play runs the risk of actually losing.

All of the +2800s have several games where they didn't make any mistakes at all.

A 2800 player might well be able to force a draw almost always against a 2800 player, based on all of their choices being based on minimising the opportunity for the opponent to get winning chances. If they can manage it against a 3400 player, they are not a 2800 player, they are a 3400 player. Grischuk is not a 3400 player.

Elroch

Jun 7, 2018

0

#212

SmyslovFan wrote:

Grischuk is actually a better example. In one Candidate's match-play tournament, he strove for draws as White and often got the draw in under 25 moves. His strategy worked in the shorter matches, but when he faced Gelfand in a longer match he lost.

A 2800 player intent on drawing is very difficult to beat with perfect play. And less than perfect play runs the risk of actually losing.

All of the +2800s have several games where they didn't make any mistakes at all.

Presumably that's "no mistakes at all" according to the engine that got 36% against AlphaZero? I have blitz and maybe even bullet games that meet that threshold (although this relies on ignoring small evaluation differences that are very sensitive to computation time, not to say that larger ones never are).

A 2800 player might well be able to force a draw almost always against a 2800 player, based on all of their choices being based on minimising the opportunity for the opponent to get winning chances. If they can manage it against a 3400 player, they are not a 2800 player, they are a 3400 player. Grischuk is not a 3400 player.

EndgameEnthusiast2357

Jun 7, 2018

0

#213

Telestu wrote:

EndgameStudier wrote:

Again, I am just talking about whether engines can be perfected or not. Forget even the ratings. Rating only apply to normal games, not these crazy puzzles, but a truly perfect engine should be able to solve even a 1 Million Move puzzle, let alone a 15 move one.

Improving engines and improving results may be more different than you think, and making the perfect engine is yet another different thing.

Making stronger and stronger engines is sort of like a goal of emptying a successively (and exponentially) larger swimming pools with buckets. Lets say the buckets you use are twice as big as mine, so you empty your swimming pool twice as fast i.e. your chess engine is stronger.

But a perfect engine is more like emptying the ocean. Your buckets may be twice as big, but after a year of dumping water, we measure ocean levels, and neither of us has made measurable progress. Getting a bigger bucket from a manufacturer or store will not help.

Meaning, sure, engines can be improved, but at some point it won't noticeably improve their results. We'd need a revolution in physics and technology before we can even entertain the idea of a machine that plays perfect chess. And while small improvements are always possible, if the pool is large enough, it wont change the results. Imagine telling me you added 1 cubic inch more space to your bucket. Yes, it's undoubtedly improvement, but it's meaningless.

Even quantum computers might be too slow in solving a problem like that.

lfPatriotGames

Jun 8, 2018

0

#214

SmyslovFan wrote:

Grischuk is actually a better example. In one Candidate's match-play tournament, he strove for draws as White and often got the draw in under 25 moves. His strategy worked in the shorter matches, but when he faced Gelfand in a longer match he lost.

A 2800 player intent on drawing is very difficult to beat with perfect play. And less than perfect play runs the risk of actually losing.

All of the +2800s have several games where they didn't make any mistakes at all.

Who decided that someone played several games where there were no mistakes at all? What happens if in 200 years when a computer looks at these games and they decide those games were full of all kinds of mistakes?

I also dont understand how a 2800 intent on drawing would be difficult to beat with perfect play because I dont think anyone has ever seen perfect play before. So who knows if it would be difficult or not. I think white has a big advantage and wanting to draw might be something black could never prevent, but I dont think we have any way of knowing that for sure right now.

I think when two grandmasters play and they are both about 2800 in ability it's probably not too difficult for one side to force a draw. But what happens when a 2800 who wants a draw plays a 3500 (computer) who doesnt want to draw. I think the 2800 then finds out how easy it is to lose even when they want to draw.

fischerrook

Jun 8, 2018

0

#215

EndgameStudier wrote:

fischerrook wrote:

I agree with EndgameStudier. After 3500 rating, they're all going to play the best move, so there will be no butts being whipped at the boards. Mostly draws and probably mostly pre-arranged agreed upon draws, so the competitive aspects will be lost.

Yes, but that's assuming chess is a draw when played perfectly. What if White Wins if both players play perfectly, then all the white players would win.

I think I tricked you in to debunking your own idea...lol.

SmyslovFan

Jun 8, 2018

0

#216

fischerrook wrote:

EndgameStudier wrote:

....What if White Wins if both players play perfectly, then all the white players would win.

I think I tricked you in to debunking your own idea...lol.

What if a bunch of people who know very little about chess started making suppositions about the game that just don't match reality?

SmyslovFan

Jun 8, 2018

0

#217

Elroch wrote:

SmyslovFan wrote:

Grischuk is actually a better example. In one Candidate's match-play tournament, he strove for draws as White and often got the draw in under 25 moves. His strategy worked in the shorter matches, but when he faced Gelfand in a longer match he lost.

A 2800 player intent on drawing is very difficult to beat with perfect play. And less than perfect play runs the risk of actually losing.

All of the +2800s have several games where they didn't make any mistakes at all.

Presumably that's "no mistakes at all" according to the engine that got 36% against AlphaZero? I have blitz and maybe even bullet games that meet that threshold (although this relies on ignoring small evaluation differences that are very sensitive to computation time, not to say that larger ones never are).

A 2800 player might well be able to force a draw almost always against a 2800 player, based on all of their choices being based on minimising the opportunity for the opponent to get winning chances. If they can manage it against a 3400 player, they are not a 2800 player, they are a 3400 player. Grischuk is not a 3400 player.

AlphaZero did not test itself against a fully functioning Stockfish. Stockfish was not optimized the way it is for the CCRL ratings. But let's agree that the version used was ~3300 and that AlphaZero scored 64%-36%. That's a 100 point difference. AlphaZero didn't dominate by 1000 points, or anything close to that. The vast majority of the games played in the match were drawn, and AlphaZero only published its wins.

The AlphaZero test was just a test. It wasn't rated by anyone. It's a very impressive test, and AlphaZero has shown some tremendous improvements over Stockfish. But it has not altered reality. It has not shown that chess is a win, and it has not shown any hint that a computer could reach 4000 elo. If anything, it demonstrated that there's a lot of space between 3300 and 3600. It probably performed at about 3400 strength, but we just don't know.

SmyslovFan

Jun 8, 2018

0

#218

Btw, Stockfish 8 64-bit is rated by CCRL at 3303. Alphazero used a version of Stockfish 8 that was not optimized.

http://www.computerchess.org.uk/ccrl/4040/rating_list_all.html

https://www.chess.com/news/view/google-s-alphazero-destroys-stockfish-in-100-game-match

EndgameEnthusiast2357

Jun 8, 2018

0

#219

Stockfish 9 is out now. But your operating system and RAM and processor speed affects how strong it is.

drmrboss

Jun 8, 2018

0

#220

Btw, SF improves "5" elo per month in average, or about "60" elo per year, since the start of fishnet in 5 years ago. Total improvement =300 elo after playing (836 million games), from 3200 to 3500 rating!!

If the progress is the same, Stockfish may probably reached 4000 rating after 8.33 years , from today. The improvement is still ongoing!!

SF 10 Developmental version= 28.2 elo stronger than SF 9!

https://github.com/glinscott/fishtest/wiki/Regression-Tests