Why does TCEC use unbalanced opening?

[mvanthoor]

I'd like to see TCEC return to using balanced openings, with perhaps a 6 moves per side limit.

And 96 draws out of 100.

Do you want the best quality chess, or just artificial entertainment?

I assume it doesn't matter if the opening position is unbalanced. You can play a risky opening with white that would give you good chances against humans that don't know how to handle this opening, even if it was black who'd come out of it with a +0.7 advantage. Let's say the engine with black wins this game because of this imbalance.

If the engines are the same strength and then you swap colors and play the exact same opening, then black (now the OTHER engine) should win it again.

So, you can have a string of openings that are unbalanced; and if you keep the amount of unbalance equal, which means that on average all openings that favor white are +0.7, then the average of openings that favor black would need to also be +0.7 (otherwise one color would win a lot more than the other). If you do the color swap/each opening twice thing, two engines that are about the same strength could have a very interesting match with many wins/losses and less draws, that still ends with both sides having the same amount of points.

[Marcus9]

Do you want the best quality chess, or just artificial entertainment?

Best quality chess from positions that matter. Balanced positions don't matter anymore as they are too easily drawn by current engines on current hardware.

I would love someone to prove me wrong on this. I am willing to offer 100-1 odds. I will take black with draw odds, play main mainlines (Ragozin vs 1.d4, Ruy Lopez vs 1.e4, ...e5 lines vs 1.c4 etc.). I will play a few moves by hand and then just run Stockfish for 1-2 hours per move. My opponent gets unlimited time within reason so we can finish the game. I am willing to accept 5$ bet. That is I donate 500$ to the chess/AI/educational project of your choosing if you win and you donate 5$ to Lichess if I draw or win.

We can start probably next week. I will only accept challanges from people who seem to be serious enough to at least have a chance to make it a fight. I would love to lose btw but I don't believe it's possible anymore.

Alright stockfish 12 is very good and two different entity in LTC produce very likely a draw. But would you make the same bet using stockfish 12 against the best engine of 10 years engine from now? We will say that the chess is nearly solved when there is no more progress, Imagine having to play the same game against stockfish 12 with the best engine available 5 years ago: Komodo 9, Would you feel so sure of winning?

[mwyoung]

Do you want the best quality chess, or just artificial entertainment?

Best quality chess from positions that matter. Balanced positions don't matter anymore as they are too easily drawn by current engines on current hardware.

I would love someone to prove me wrong on this. I am willing to offer 100-1 odds. I will take black with draw odds, play main mainlines (Ragozin vs 1.d4, Ruy Lopez vs 1.e4, ...e5 lines vs 1.c4 etc.). I will play a few moves by hand and then just run Stockfish for 1-2 hours per move. My opponent gets unlimited time within reason so we can finish the game. I am willing to accept 5$ bet. That is I donate 500$ to the chess/AI/educational project of your choosing if you win and you donate 5$ to Lichess if I draw or win.

We can start probably next week. I will only accept challanges from people who seem to be serious enough to at least have a chance to make it a fight. I would love to lose btw but I don't believe it's possible anymore.

"Best quality chess from positions that matter. Balanced positions don't matter anymore as they are too easily drawn by current engines on current hardware."

Lets bet on what is being claimed! You said easily drawn. 2 hours a move does not sound easy. You get 3 mins a move, and I get as much time as I wish with in reason. And you can play any opening you wish, but so can I. And we play 24 games. Starting from this balanced position. A straight up match, no draw odds.

[d]rnbqkbnr/pppppppp/8/8/8/8/PPPPPPPP/RNBQKBNR w KQkq - 0 1

[Milos]

I'd like to see TCEC return to using balanced openings, with perhaps a 6 moves per side limit.

And 96 draws out of 100.

You are really an optimist, I would have been 98 draws out of 100, if we are lucky.

[AndrewGrant]

2x Stockfish wins in a row? Next season Leela won't make it to the SuFi?

I'm sure there will be some changes, openings included, worry not Leela friends

[OneTrickPony]

Alright stockfish 12 is very good and two different entity in LTC produce very likely a draw. But would you make the same bet using stockfish 12 against the best engine of 10 years engine from now? We will say that the chess is nearly solved when there is no more progress, Imagine having to play the same game against stockfish 12 with the best engine available 5 years ago: Komodo 9, Would you feel so sure of winning?

It just feels so hopeless today when analyzing. Difficult to find any kind of shot for white in main lines (I mean for computer chess, for humans it's more interesting than ever).
I would accept the bet if you had an engine from 2030 today. As it is we have to wait anyway, so the bet is not very interesting.

Lets bet on what is being claimed! You said easily drawn. 2 hours a move does not sound easy. You get 3 mins a move, and I get as much time as I wish with in reason. And you can play any opening you wish, but so can I. And we play 24 games. Starting from this balanced position. A straight up match, no draw odds.

I said too easily. That is it's too easy for them to make the match boring not to draw 100% of game with TCEC time controls (although it would be very close to 100%). I think it would be 100% already with a bit longer time controls or better hardware. I don't have TCEC hardware at home but I might accept a game when I run SF only for 15 minutes per move on 3970x, which would be equivalent of about 5-7 minutes per move in TCEC (and the engines have enough time there to think that long in key moments).

I don't really have time nor interest to play 24 very long correspondence games. I might play a few if it's interesting. I would really love to lose one as I love chess and it would open my eyes to new possibilities.
As it is right now I believe the only interesting ground left (for cc/computer chess) are very unbalanced starting positions and TCEC does great job of showing that to the world.

[Uri Blass]

I'd like to see TCEC return to using balanced openings, with perhaps a 6 moves per side limit.

And 96 draws out of 100.

You are really an optimist, I would have been 98 draws out of 100, if we are lucky.

I do not care about the number of draws and I would like to see 100 games with TCEC hardware and TCEC time control to see how close engines are to be unbeatable.
No opening book so people cannot blame the opening book for a loss.

Stockfish with many cores is not deterministic so we are not going to see the same game again and again.

[duncan]

I'd like to see TCEC return to using balanced openings, with perhaps a 6 moves per side limit.

And 96 draws out of 100.

Do you want the best quality chess, or just artificial entertainment?

Best quality chess is agreeing to a draw right after starting the clock (both 32 men tbs). And frankly, I can hardly follow the games of 2600 GMs without a help, I am to "profoundly admire" a 4000 Elo level 96% drawfest? Risible endeavor to me.

You can still learn a lot from draws, and I doubt that they'd mainly be boring.
However, everybody to their own taste.

We need to ask ourselves why are we seeing draws.

1. TCEC hardware running Lc0, and Stockfish 12 are so strong that they have nearly solved chess to a draw.
2. As with top level human GM games. When 2 strong and equal players play. The results as theory suggest. The game will end in a draw.

For 1 to be true, then Stockfish 12. In 5 years time could play the best engine in 5 years, or 1,000,000 years. On today's TCEC hardware, and still score a 96% draw rate. That is the logical claim that is being made here by some.

and if (2) then why are we seeing a higher percentage of draws as computer get more powerful?

[CMCanavessi]

Do you want the best quality chess, or just artificial entertainment?

Best quality chess from positions that matter. Balanced positions don't matter anymore as they are too easily drawn by current engines on current hardware.

I would love someone to prove me wrong on this. I am willing to offer 100-1 odds. I will take black with draw odds, play main mainlines (Ragozin vs 1.d4, Ruy Lopez vs 1.e4, ...e5 lines vs 1.c4 etc.). I will play a few moves by hand and then just run Stockfish for 1-2 hours per move. My opponent gets unlimited time within reason so we can finish the game. I am willing to accept 5$ bet. That is I donate 500$ to the chess/AI/educational project of your choosing if you win and you donate 5$ to Lichess if I draw or win.

We can start probably next week. I will only accept challanges from people who seem to be serious enough to at least have a chance to make it a fight. I would love to lose btw but I don't believe it's possible anymore.

"Best quality chess from positions that matter. Balanced positions don't matter anymore as they are too easily drawn by current engines on current hardware."

Lets bet on what is being claimed! You said easily drawn. 2 hours a move does not sound easy. You get 3 mins a move, and I get as much time as I wish with in reason. And you can play any opening you wish, but so can I. And we play 24 games. Starting from this balanced position. A straight up match, no draw odds.

[d]rnbqkbnr/pppppppp/8/8/8/8/PPPPPPPP/RNBQKBNR w KQkq - 0 1

Are you and I some 3800 ELO chess masters?

[Alayan]

I'd like to see TCEC return to using balanced openings, with perhaps a 6 moves per side limit.

And 96 draws out of 100.

You are really an optimist, I would have been 98 draws out of 100, if we are lucky.

I do not care about the number of draws and I would like to see 100 games with TCEC hardware and TCEC time control to see how close engines are to be unbeatable.
No opening book so people cannot blame the opening book for a loss.

Stockfish with many cores is not deterministic so we are not going to see the same game again and again.

Navratil has been making games that should be to your liking.

Hardware : 10980XE + 2 x RTX2070
TC : 1h + 1m increment
Engines : SF and Lc0 SuFi versions.
Openings : 8 plies balanced GM openings (+2-2=96), bookless (100 draws, extremely similar games throughout).

This doesn't actually show what a future stronger engine could do, but it's telling when two engines, different in strength at short TC or unbalanced openings, different in playing style, end up drawing everything.