Purpose of correspondence chess in modern era

[eligolf]

I want to start by saying that I have very little knowledge in the correspondence chess world or how thing works with tournaments and stuff. Therefore I am sorry if I offend some correspondence chess player with my post

The main question: What is the point of correspondence chess in the computer era?

To me it seems completely pointless to play correspondence chess with the aid of computers, which I understand is allowed? Everyone have access to the same software and hardware (theoretically) so therefore it takes very little knowledge to draw a "master" correspondence chess player? It seems more to be a matter of "my thing is larger than yours", i.e. a competition of who can afford the best hardware for the purpose. Is even the slightest chess knowledge needed?

Sorry again for my complete lack of knowledge in this field. I don't want to offend anyone, just trying to understand and learn

[JohnW]

From what I have read, some correspondence chess tournaments consider it cheating to use a computer and if you were to win the tournament director can select one of your games and require you to provide analysis to prove you knew what you were doing on your own. There are also different types of correspondence chess called Centaur tournaments where human\computer teams compete against each other and you can also enter a stand alone computer. Often times the human\computer will beat the computer stand alone. I guess the theory is that the humans are better in the end game and certain positional situations and without time constraints will come up with a better decision.

[Ozymandias]

some correspondence chess tournaments consider it cheating to use a computer and if you were to win the tournament director can select one of your games and require you to provide analysis to prove you knew what you were doing on your own.

Which tournaments? What TD has done that?

[eligolf]

Often times the human\computer will beat the computer stand alone. I guess the theory is that the humans are better in the end game and certain positional situations and without time constraints will come up with a better decision.

That is cool John, I wouldn't have guessed that. To be able to play through the midgame and achieve a winning endgame vs a top engine must take some serious analysis and great chess player(s). In the era of NNUE, Lc0 and similar engines makes it even harder since many of their moves are hard to understand from the current human perspective.

[Dann Corbit]

Computers affect every sort of chess.
Preparation for any sort of tournament will involve computers.
Every game will get a computer post-mortem too.

So, shall we just give up then?
I think John Henry still has to try, until he lays down his hammer and he dies.

[cdani]

What is the point of correspondence chess in the computer era?

Fun
And with more time and human work, you have more possibilities to win games. Yes, human work. Engines tend to be simplifiers, or head to a drawish endgame where they think there is advantage, they don't prove the rivals,...
So you have to search ways to win. So you need more than just linear computer analysis.

[jefk]

cdani wrote:
'fun'

well in some ways, indeed; for the Iccf players it's probably a passion for chess, because it takes patience stamina and many years to get to the top, also (or maybe especially) now in this computer era. If you don't like (or are not knowledgeable about) chess, you also are more prone to enter the wrong (or sub-optimal) moves, when you have multiple games running simultaneously (yes this happens sometimes. The best Iccf players seem to come from Germany or Russia, which doesn't surprise me, computer chess is popular in Germany, and chess in general is very popular in Russia. Look at the last ICCF world championship:
https://www.iccf.com/event?id=66745
NB i looked at the won/lost games and indeed opening theory played an important role; simply playing opening moves from eg. the Chinese database isn't good enough at this level because it 1) is not the SF Nnue (yet?) and 2) doesn't have statistics (of real games, incl endgames played) and last but not least doesn't give you any verbal understanding (like in many chess books written by IM+about specific openings) and thus it you would enjoy it about as much as copying pages from a telephone book into a webserver.

[carldaman]

http://talkchess.com/forum3/viewtopic.p ... 50#p880605

I have already posted a similar question in the other thread.

"Can a test be run to see if top human corr players (playing as centaurs) can defeat unaided top engines running on some 'regular' hardware?"

Bare engine participation is something I had proposed a few years back, as well.

[jefk]

Can a test be run to see if top human corr players (playing as centaurs) can defeat unaided top engines running on some 'regular' hardware?"

Well the 'test' inmho is still simply the ICCF competition system itself; so
why running a simple test about something which (still) is obvious for more advanced chess players (at least for me (*))? Maybe SFNNue looks like approaching perfect chess but there still are areas where it's not perfect, e,g, complicated positions (look at the latest thread on video review of some deep analysis programs), practical endgames (maybe in such positions we would need another Nnue than for the opening/middlegame (just and idea), etc

For the 'centaur' it depends on his/her (advanced) chess knowledge, eg in practical endgames it's a matter of strategy how to win positions with double bishops (the famous bishop pair) vs bishop knight,SF- Nnue doesn't seem to give a large advantage for the bishop pair; indeed it's often difficult or impossible to win just based on such an advantage, while it depends on lots of other positional elements (usually pawns, king's position, and sometimes also other pieces).

This also shows that correspondence chess still has a 'purpose', namely to enhance your understanding of chess, especially openings (although Nnue is pretty good, i expect new/updated opening trees will emerge as result of the new engines), but also middlegame/endgame positions, and 'practical' endgames.

Sure we are approaching at top level draw equality, but i don't think that running an SF engine for an hour without human assistance/understanding will give top level correspondence chess and will guarantee a draw. Especially not when some rules would be modified, to decrease the (increasing) draw score at *top* level correspondence chess (as was discussed in the 'Kaufman' thread about correspondence chess in the Nnue area).

If you think otherwise, try it, you may get to Iccf 'expert' level, and in the meantime also increase your chess understanding (although there may be more time effective ways to achieve this, i agree), but i still think you won't get
beyond 'expert' level without more in-depth understanding & experience.

Or accept following my challenge, I start with White (**), you can replay with your engine move (let it run for one or two hours, as Uri B suggested should be sufficient). Yes i'm serious, i hereby challenge you, and subsequently you also get your 'test'; five rules, 1) you are not allowed to update the Neural net in later stages (presuming you are going to use SF Nnue, 2) nor your current hardware (at least for the game you now can accept with Black, if you want and 3) you will not increase engine thinking time (beyond the 2 hrs) , 4) being honest and only plugging in the engine move, you (ofcourse) will not accept any human help or change strategy after reading my (or others) comments during the game, and 5) time forfeit for me is impossible (unless i would have abandoned the game or so eg. due to disease, in such a case it may count as draw), but generally i will try to respond with a maximum of -average- one week per move or so (probably faster).
Good luck !
NB although this forum is not for correspondence chess, such a test should be allowed imho, especially also with the comments/discussion which it can generate.

(*) for the opening stage, Uri Blass (GM correspondence player) wrote that SF-Nnue nowadays is good enough to get openings moves which cannot be beaten; but i don't think this is always true; make one suboptimal move (for Black) and still may lose the game, especially if for whatever reason you (ie the engine) got into a sharp, highly unbalanced line.
(**) probably 1.Nf3 or so, but i will decide on that after you accept my challenge.

[MonteCarlo]

This is intriguing to me. I may be willing to play the other side, but there are a few things I'd want to clarify since I'm not familiar with the details of the referenced claim by Uri Blass.

1) Is the claim that a top level engine can draw with an unassisted 2 hours per move, even if the centaur opponent knows the exact engine build, configuration settings, and hardware specs? I'm not sure I'd be interested in defending that claim, as even at the current level, there are enough examples of positions where they make big mistakes that I suspect it could be led down a bad path by a skilled centaur if the centaur knows exactly the build/specs and has the time to predict its responses.

If, on the other hand, the claim is just that under normal correspondence conditions, where you don't know exactly what engine the opponent is using, or what build of that engine it's using, or the configuration, or the level of search being obtained, top engines are enough to draw a game unassisted, that I think is interesting.

2) I'm sure there's a good reason for it, but it seems the conditions you cite give you decent time odds to the centaur. With 2 hours unassisted per move, even if I were unable to analyze multiple games in parallel, I'd be able to take on a load of 84 simultaneous games at 1 week per move. If you're playing fewer than that, I'd be giving you time odds (as if I were playing in the ICCF I could take on the same workload and use more time than the 2 hours allotted).

This goes back to the exact claim being defended. If the claim is literally that 2 hours unassisted is enough always to hold a draw no matter how much time the centaur is given, that seems to me incredibly likely to be false, so I'm not too interested in exploring that claim.

If the claim is that using an unassisted engine to play in the ICCF will get you draws against skilled centaurs assuming similar workloads, then we'd need to do something to make sure that the centaur's not getting big time odds.

3) Assuming we're talking about the to-me-interesting scenarios from items 1 and 2, then there's just the question of what hardware is acceptable. It's probably easiest just to say max nodes allowed for the various top engines (SF traditional, SF NNUE, LC0 with big net), that way no matter what hardware I have, I could just scale to the desired level of play.

Cheers!