$20,000 ICUC - a report on wins

[Thomas A. Anderson]

Considering what Thomas and I labelled as "irregular wins", as the first tie-break rule, is a no-brainier, but I think we should be more ambitious, about the changes needed, and tackle the problem head on, at a much earlier stage.

Don't be to optimistisch what are no-brainers at IC. As I recently learned , even using a Swiss-Tournament software in the next tour, that fully complies to FIDE rules, isn't more than a hope for a program update. I think it make sense to continue the brainstorming offline with IC contribution. I'll ask Arno, if he is willing to setup a comunication plattform for interested people.

[Cumnor]

There is a forum here http://infinitychess.freeforums.org/portal.php

[Ozymandias]

Don't be to optimistisch what are no-brainers at IC. [...] using a Swiss-Tournament software in the next tour, that fully complies to FIDE rules, isn't more than a hope [...] I think it make sense to continue the brainstorming offline with IC contribution. I'll ask Arno, if he is willing to setup a communication platform for interested people.

I said that it was a no-brainer, not that it would be to them.

I was assured in the tournament chat, that the current software was already FIDE compliant, you're just confirming that my suspicions, about that not being the case, weren't just happenstance.

What's the benefit of going "offline"? I'm still waiting on an answer to my last two emails to him (dated April 21st and May 20th). In fact, the only way I have of knowing that I have the correct address, is because the very next day I pointed him to this post (before I actually went online with it), he proceeded to delete all the "[ct]" info on their DB, instead of accepting or debating about the corrections made.

With that kind of predisposition to discuss facts, I can't be anything else but skeptical, about discussing ideas.

At least in an open and active forum, like this one, we have the chance of hearing from other people (not much of a chance, but at least some). Discussion would be much more lively if this were the Rybka Forum of old, but as I've been claiming for years, that one's going the way of the dodo. The interest around here isn't exactly about Freestyle, but let's face it, we lack a proper home. The official forum never got of the ground.

[Thomas Zipproth]

Some notes about the current state of computer chess.

About one and a half year ago, while adding new calculated lines to Cerebellum, I noticed for the first time, that something would change in computer chess.
I realised that in contrast to what came before, there was no position problematic enough that could not be solved by calculating and adding some lost games and critical lines.
That includes positional and tactical problems.

The exceptions where very closed positions, fortresses and some wrongly evaluated endgames, but that did not happen very frequently, so it did not play a big role.
That meant that for all remaining notoriously difficult openings it became clear how to defend them as black or even get an advantage if white overplays the position.
The last one of this openings was the Giuoco Piano.

In the time that followed, I saw how this materialized also in the IC tournaments, till it could be finally seen in the Ultimate tournament.
Most games where won because of book errors or playing against someone out of the small group of weak players (Engine and/or weak hardware).
Some of the the weakest ones played most of their games as centaurs, so I'd leave this out.

I see basically three different ways to still win a game.

1.) The way Cerebellum + Brainfish does it:
Try to play as good as possible (good opening, strong engine and medium to good hardare), without too much consideration of statistics and engine tweaks.
This still leads to regular successes (I won two IC Tournaments this week), but it also becomes more and more difficult.
Therefore I'm already mixing a bit statistics (2.) into Cerebellum.

2.) The statistical way:
Play lines which where successful in the past, hoping that your opponent repeats errors from the past.
This still can be surprisingly successful, as it has been seen in the Ultimate tournament.

3.) The extreme hardware way:
Use a very fast computer, let's say with 16 - 44 Cores, and a strong engine.
Then try to lead your opponent into unknown terrain, best into a semi closed position where something like a slow developing king atack is possible,
An example for this is 1.e4 e5 2. Nc3 Nf6 3. g3

But of course, it still becomes more and more visible, that chess is finally a draw.
The best engines on fast hardware with enough time (rapid) are often playing nearly flawless games, finding their path into a drawn endgame.

Regarding the IC Ultimate Tournament, it becomes evident, that because of the high draw rate, you have to avoid any human or enviromental errors (like mouse slip, DC or even not playing), because otherwise these errors will dominate the engine results.

A possible solution to lower the draw rate in tournaments with a time control equal or above rapid could be to play some fixed openings changing every second round.
For example the Kings Gambit with white and black in the first two rounds, and then the Benko Gambit.
Those openings aren't lost for Black, the theory is not so advanced as in other openings, and it could lead to interesting and exciting games for some time.

[jefk]

We urgently need to find something to stop this development.

The main problem, is that some players, like Jef, still believe they can score better trough their own play, and most of the players who already acknowledge reality, remain silent.

well i also acknowledge reality ofcourse, in fact i was one of the first claiming that chess is a draw, which led to some opposition (eg by Bob Hyatt, but also by some math afficionado's who claim there's no proof).

Anyway, reason i could play better as centaur was:
1) that i have my own book, in ChessPartner format (not with these .ctg statistics), and with own analysis and some comparisons with Cerebellum; to choose the proper book moves i couldn't play in engine mode.
2) stockfish still has some positional flaws; as centaur i could look at other engine's evals and thus at certain critical points set up an attack by selecting an aggressive engine move (one of many) which agreed with my hunch/intuition that this could lead to an attack;

For the rest not going to reveal more about my method(s), ofcourse it will become more and more difficult to beat the best engines including (or especially) this mysterious 'Zor', and it becomes similar as the draw problem in correspondence chess (where also multiplte engines are used together with own knowledge btw). Someone as Tsvetkov here would be a grandmaster on ICCF (correspondence chess) if everything he claims would be true; in reality ofcourse top engines as K11 nowadays simply refult many of the moves suggested by him; sometimes i have ideas a bit similar as Tsvetkov, but being an active correspondence player (and recently participated in this ultimate tourn) i'm not going to teach you how to do it ofcourse..
Concluding, i still would welcome another tourn as this Ultimate 2017 tourn, Cerebellum also is still improving, and we still could see some interesting games at high level. In the meantime we can start thinking about methods to reduce the increasing draw rate problem..

[Ozymandias]

I see basically three different ways to still win a game.

1.) The way Cerebellum + Brainfish does it:
Try to play as good as possible (good opening, strong engine and medium to good hardare), without too much consideration of statistics and engine tweaks.

2.) The statistical way:
Play lines which where successful in the past, hoping that your opponent repeats errors from the past.

3.) The extreme hardware way:
Use a very fast computer, let's say with 16 - 44 Cores, and a strong engine.
Then try to lead your opponent into unknown terrain, best into a semi closed position where something like a slow developing king atack is possible,
An example for this is 1.e4 e5 2. Nc3 Nf6 3. g3

But of course, it still becomes more and more visible, that chess is finally a draw.
The best engines on fast hardware with enough time (rapid) are often playing nearly flawless games, finding their path into a drawn endgame.

There's very little to add to this part, I've summarised your three approaches as to how to win a game, and then bolded the part which pretty much admits none of them will work, unless your opponent "cooperates".

Regarding the IC Ultimate Tournament, it becomes evident, that because of the high draw rate, you have to avoid any human or environmental errors (like mouse slip, DC or even not playing), because otherwise these errors will dominate the engine results.

A hundred times this.

A possible solution to lower the draw rate in tournaments with a time control equal or above rapid could be to play some fixed openings changing every second round.
For example the Kings Gambit with white and black in the first two rounds, and then the Benko Gambit.
Those openings aren't lost for Black, the theory is not so advanced as in other openings, and it could lead to interesting and exciting games for some time.

Before forcing players to play some specific lines, it'd be a first step to force them not to play at least one: The Marshall Attack. Selecting a prearranged set of openings, could work, but only if announced way before the tournament begins, and not with the usual one month notice.

i was one of the first claiming that chess is a draw, which led to some opposition (e.g. by Bob Hyatt, but also by some math afficionado's who claim there's no proof).

You can't fault them for speaking the truth, you can however, accuse them of focusing on just one interpretation of what chess really is. Is it a problem or a game? If you don't admit the dichotomy, you can end up, making statement's like those of Bob.

Anyway, reason i could play better as centaur was:
1) that i have my own book, in ChessPartner format (not with these .ctg statistics), and with own analysis and some comparisons with Cerebellum; to choose the proper book moves i couldn't play in engine mode.
2) stockfish still has some positional flaws; as centaur i could look at other engine's evals and thus at certain critical points set up an attack by selecting an aggressive engine move (one of many) which agreed with my hunch/intuition that this could lead to an attack;

All that, and more, sounds good and I would agree, that a centaur plays better than an engine, but the difference is so small right now, that's become a moot point. Just look at the performance gap between the first two players, centaurs, and the best engines... negligible. Right now, playing as centaur is more a question of style than one of strength.

Regarding your wins, if your nick is the one I suspect, at least one of the two could almost be considered as a book loss: the first 4 black moves are suboptimal at best, and move 9 avoids a couple of previously played, drawish choices, to go down a route that leaves the engine in a pretty bad position (move 11). The other win had more merit, but still, black's choice for move 18 (the last book move), was simply bad.

None of the two, can be categorised as book losses, because there's still not enough data points to establish a correlation, but they're good examples of what I called earlier as giving "the winning side, ample advantage".

[Thomas A. Anderson]

Before forcing players to play some specific lines, it'd be a first step to force them not to play at least one: The Marshall Attack. Selecting a prearranged set of openings, could work, but only if announced way before the tournament begins, and not with the usual one month notice.

You do not really propose to use opening blacklists, do you? You might have your personal nightmare with the Marshall, but you know that such a list would be endless. Different start positions sound more reasonable, but if the intention is to lower draw rates, why announcing in advance?

[Ozymandias]

Before forcing players to play some specific lines, it'd be a first step to force them not to play at least one: The Marshall Attack. Selecting a prearranged set of openings, could work, but only if announced way before the tournament begins, and not with the usual one month notice.

You do not really propose to use opening blacklists, do you? You might have your personal nightmare with the Marshall, but you know that such a list would be endless. Different start positions sound more reasonable, but if the intention is to lower draw rates, why announcing in advance?

As I mentioned earlier, chess is both a game and a problem. What's the outcome from the initial position? Everything points out to a draw, and currently, the Marshall Attack nullifies e4 Nf3, which has always been the main try for white to create an advantage. It's nothing personal, everyone is shying away from playing 3 Bb5, just because they're anticipating it. No one has found a suitable way to proceed. Which means that, at least from a practical point of view, that part of the problem has already been solved. It makes no sense to keep trying new ways of drawing, or even losing, like Werner did against Uwe in the ICUC. Chess is also a game, and refuting the Marshall Attack has proven to be a most boring exercise, which is why the draw rate keeps getting higher and the game gets getting more boring.
At least Christopher Lutz has been honest enough to accept this fact, and has focused on three draws, as possibly the most interesting games out of the 781 played. Arno, on the other hand, has repeatedly skipped the opening phase, to highlight "beautiful" games. Let me be clear, without truth, there's no beauty. If you want that, you can just compose a position, but a game needs to be well played from the first move, in order to be considered of artistic value.
The list of banned defences would grow, that's the nature of solving the problem aspect of chess, but at least it would keep the game aspect interesting for some years to come. In the end, you'd have a number of start positions which would still be broader and less arbitrary, than what Thomas proposes. As for announcing them in advance, let's not forget that members of the IC team do take part in the event, if they were to be kept secret until the last moment, how would you know they weren't privy to them, beforehand? Remember: "Caesar's wife must be above suspicion"; in this case it's the wife, the mother and the sister, who must watch out for their reputations.

[carldaman]

Before forcing players to play some specific lines, it'd be a first step to force them not to play at least one: The Marshall Attack. Selecting a prearranged set of openings, could work, but only if announced way before the tournament begins, and not with the usual one month notice.

You do not really propose to use opening blacklists, do you? You might have your personal nightmare with the Marshall, but you know that such a list would be endless. Different start positions sound more reasonable, but if the intention is to lower draw rates, why announcing in advance?

As I mentioned earlier, chess is both a game and a problem. What's the outcome from the initial position? Everything points out to a draw, and currently, the Marshall Attack nullifies e4 Nf3, which has always been the main try for white to create an advantage. It's nothing personal, everyone is shying away from playing 3 Bb5, just because they're anticipating it. No one has found a suitable way to proceed. Which means that, at least from a practical point of view, that part of the problem has already been solved. It makes no sense to keep trying new ways of drawing, or even losing, like Werner did against Uwe in the ICUC. Chess is also a game, and refuting the Marshall Attack has proven to be a most boring exercise, which is why the draw rate keeps getting higher and the game gets getting more boring.
At least Christopher Lutz has been honest enough to accept this fact, and has focused on three draws, as possibly the most interesting games out of the 781 played. Arno, on the other hand, has repeatedly skipped the opening phase, to highlight "beautiful" games. Let me be clear, without truth, there's no beauty. If you want that, you can just compose a position, but a game needs to be well played from the first move, in order to be considered of artistic value.
The list of banned defences would grow, that's the nature of solving the problem aspect of chess, but at least it would keep the game aspect interesting for some years to come. In the end, you'd have a number of start positions which would still be broader and less arbitrary, than what Thomas proposes. As for announcing them in advance, let's not forget that members of the IC team do take part in the event, if they were to be kept secret until the last moment, how would you know they weren't privy to them, beforehand? Remember: "Caesar's wife must be above suspicion"; in this case it's the wife, the mother and the sister, who must watch out for their reputations.

'Banned' is an ugly word with negative connotations in this context and that would turn many people off. Why not use thematic openings that would, of course, leave out the Marshall and other troublesome boring/drawish openings? It is just a subtle nuance, but it may matter to some.

CL

[Ozymandias]

Why not use thematic openings that would, of course, leave out the Marshall and other troublesome boring/drawish openings?

Mainly, because (without thousands of engine games to back it up), the decision about which openings to label as boring/drawish, would be somewhat arbitrary.

Also, there's the thing about not every edgy opening being equal. Depending on what the chances are for a sound defence, getting an opening against a strong or a weak opponent would en up making all the difference.

The only way to make it fair, would be to play a thematic tour, like those in correspondence chess. But of course, you know what the opening is going to be, in advance.