BrendanJNorman wrote: ↑Fri Sep 09, 2022 5:21 am
Just because two engines draw 100% of games, this doesn't mean chess is "solved".
It means the next Fruit/Rybka/Ippolit/Stockfish/AlphaZero/Lc0 hasn't come along yet to shake the game up.
And it *will* happen. Always does.
+1
absolutely...
Lots of draws simply means that top engines are close in strength.
It doesn't necessarily mean they're approaching perfect play, does it?
I don't believe for a second that SF 15 or Komodo or any engine play perfect chess...there's still a long way to go.
I remember when everyone thought Rybka 3 (2008) played perfectly...
now it's #64 on CCRL list:
Rybka 3 64-bit 4CPU 3114
I'd like to suggest that when 'perfect play' is achieved...
draw rate will = 100% (with millions of games played).
BrendanJNorman wrote: ↑Fri Sep 09, 2022 5:21 am
Just because two engines draw 100% of games, this doesn't mean chess is "solved".
It means the next Fruit/Rybka/Ippolit/Stockfish/AlphaZero/Lc0 hasn't come along yet to shake the game up.
And it *will* happen. Always does.
+1
absolutely...
Lots of draws simply means that top engines are close in strength.
It doesn't necessarily mean they're approaching perfect play, does it?
I don't believe for a second that SF 15 or Komodo or any engine play perfect chess...there's still a long way to go.
I remember when everyone thought Rybka 3 (2008) played perfectly...
now it's #64 on CCRL list:
Rybka 3 64-bit 4CPU 3114
I'd like to suggest that when 'perfect play' is achieved...
draw rate will = 100% (with millions of games played).
I don't recall anyone knowledgeable thinking that Rybka 3 played perfectly. There were plenty of events where players both used Rybka and lots of games ended decisively, correspondence play was still a real game then. But when the top engine draws every game against itself or a near-equal, that means that they are probably playing well enough to stay within the drawing margin if they aren't forced to play bad openings. That's not at all the same as "playing perfectly", it is certainly not the case that Stockfish or Dragon will always win if the opponent makes a losing error. I look at it this way: the margin needed to lose a game (in the opening) is about 70 true centipawns (not 0.70 Stockfish eval). Maybe the best engines on average are now good enough that the standard deviation of their cumulative errors in a game is something like 15 true centipawns (just a guess). So if White starts at about +20 centipawns, Black needs to lose about 50 centipawns to lose a game. If my estimate is right, that might happen in something like 1 out of a 1000 games, and maybe in another year or two (or even now with very long time control and big hardware) it might already be down to 1 in a million. But if the opponent makes a move that drops the true score (whatever that means) to +75 centipawns, then the engine may have only a 70% chance or so of winning the game, as it is well within one standard deviation from the win/draw line. That's why testing with unbalanced human openings or playing chess 324 produces a decent share of draw and wins even at long time controls, while playing normal chess with best openings leads to 99% plus draws. Bottom line: "Perfect play" has two very different meanings; never losing a game is much easier than always winning a "won" game. The first goal is near, the second is still far away.
lkaufman wrote: ↑Fri Sep 09, 2022 6:22 am
Bottom line: "Perfect play" has two very different meanings; never losing a game is much easier than always winning a "won" game. The first goal is near, the second is still far away.
The definition of a "won" game today would likely depend on statistics from today's (imperfect) top engines...
or does an empirical truth already exist of which you are aware?
lkaufman wrote: ↑Fri Sep 09, 2022 6:22 am
Bottom line: "Perfect play" has two very different meanings; never losing a game is much easier than always winning a "won" game. The first goal is near, the second is still far away.
The definition of a "won" game today would likely depend on statistics from today's (imperfect) top engines...
or does an empirical truth already exist of which you are aware?
I mean "won" game in the game-theoretical sense. I know that we have no general way to determine it today. But a perfect player should never miss a move that someone later can prove would have won the game. In other words, we can't prove that an engine is perfect, but we can prove that it is not.
kranium wrote: ↑Fri Sep 09, 2022 5:43 am
Lots of draws simply means that top engines are close in strength.
I have no evidence, but I suggest that if you use imbalanced time control (give 2x time for one side) you will see many Wins/losses (as expected for such ELO difference).
I also suggest, that many draws is not sign of perfection, but rather similarities of development and engine knowledge. Two noobs cannot outplay each other if they have IDENTICAL (but very poor) knowledge/skills.
yurikvelo wrote: ↑Fri Sep 09, 2022 10:53 amI also suggest, that many draws is not sign of perfection, but rather similarities of development and engine knowledge. Two noobs cannot outplay each other if they have IDENTICAL (but very poor) knowledge/skills.
We know that, assuming players are good enough to get a checkmate (without which they probably wouldn't get a rating), then the higher the rating, the more likely they are to draw.
It seems very likely to me that if you keep improving, you eventually become almost unbeatable. I think in previous threads somebody charted draw rate against rating, and, as best I remember, the draw rate would be expected to go to 100% at something like 4200 elo. Of course, the curve may flatten as it approaches 100%. So... the evidence suggests that it's still possible to beat today's computers, and probably today's correspondence chess players as well. If chess programs continue to strengthen, there will come a time when it's almost impossible to beat them.
Human chess is partly about tactics and strategy, but mostly about memory
This is interesting since I always thought that if you give a top engine lets say 4 minutes per move while the other top engine only get about 15 seconds per move the engine getting 4 minutes per move is equivalent to having a CPU that is capable of processing16 times faster than the engine that is getting only 15 seconds per move, if both engine are using the same CPU. Therefore, for Dragon 3.1 to get such an impressive result is almost unbelievable
Chessqueen wrote: ↑Wed Sep 07, 2022 7:42 pm
With good Opening and super strong hardware The Top 2 engines are telling us that Standard Chess is reaching the point where most games end up in Draws...
I have no evidence, but my intuition is that there might be a way to beat today's top engines at standard chess: play for complex positions. Complex positions would offer two advantages:
1. throttle search depth by increasing the branching factor
2. weaken the value of knowledge because the position is so tactical
I am assuming, of course, that it's possible to:
a) find moves that lead to positions becoming complicated with a large number of lines the opponent will be obliged to investigate
b) outplay top chess programs in these positions
I could be wrong about all this: the evidence isn't strong for any of it except that I have seen search depth get throttled by complex positions on older computers (it has been many years since I seriously investigated this). I would be genuinely interested in what other people think.
What I am getting at is that Standard chess very soon will be solved probably in the next 5 years where the top 2 engines will draw 1000 games as long as the Opening is balanced and NOT tweaked in favor of one side like in TCEC
Just because two engines draw 100% of games, this doesn't mean chess is "solved".
It means the next Fruit/Rybka/Ippolit/Stockfish/AlphaZero/Lc0 hasn't come along yet to shake the game up.
And it *will* happen. Always does.
But when and it will happens in the future you see the top 2 engines Draw 1000 times out of 1000 games, then you know the time has arrived
. Book was 8 moves. With reversed colors 94 draws in 100 games. Dragon lost one game 
.