Not logged inRybka Chess Community Forum
Up Topic The Rybka Lounge / Computer Chess / The Problem with Chess Engines
- - By Bob Durrett (***) Date 2012-10-28 22:01 Edited 2012-10-28 22:04
The Problem with Chess Engines:

I am discovering that many interesting moves are simply rejected by chess engines because they search for the move that yeilds the most favorable score.

In one of my correspondence games, I immediately noticed a pawn sacrifice idea and spent a number of hours looking at it. My opponent could have played it and given me fits. Fortunately, he did not.

This is just one example of a chess engine ignoring many interesting ideas.

Note that it is normally not necessary to play the "best" (according to the chess engine) move to get an interesting game.

Opinions, anybody? Am I on the mark on this one?

Bob Durrett
Parent - - By kevsamiga (*) [gb] Date 2012-10-29 01:24 Edited 2012-10-29 02:50
The reason is selectivity, allowing a program to search deeper at the cost of missing something that at first look is unimportant.
So it discards unpromising moves, the trouble is how do you know they are unpromising unless you've checked them fully but then
your into the horizon effect and combinatoral explosion of the game tree and left with a program that doesn't see as deep.

The problem has excacerbated since engines became more "positional" and seeing deeper at the cost of missing some tactics and ideas.

The real reason is minimax choosing the best evaluated position even though subjectively it might not always be the best. But the evaluation
function says it is. GM's tend to disagree a lot of the time with computery moves based on a strategic standpoint.

You can add further knowledge to the eval to counteract it for different scenarios, but I still don't think it makes sense for a position to be represented by a single number
with the engines of today. Chess isn't only calculation or only 0's and 1's.

Maybe this is the way to go, who knows...

In any case, Computers for correspondance shouldn't be solely relied upon, because their positional understanding is most of the time over the search horizon.
And as a result can fall down into holes in static positions they don't or can't play well when there is nothing to calculate.

It needs a paradigm shift in computer chess to change that...

See : https://www.youtube.com/watch?v=DNMkF66VCow
Parent - By Bob Durrett (***) Date 2012-10-29 03:41
I just finished watching that video. It is EXCELLENT!

More tomorrow.

Bob Durrett
Parent - - By Christian Packi (****) [de] Date 2012-10-29 16:33
This video is probably 10 years old when Shredder still was the top engine. One thing GM Dzindzichashvili was right about though is that humans will soon stop playing the computer. We are already past that point.
Parent - By Geomusic (*****) Date 2012-10-30 04:39
I still enjoy playing (C)s @ most time-controls but 1 min chess...LOL
Parent - By InspectorGadget (*****) [za] Date 2012-11-01 17:31
I always pick on Piranha and Philemon :lol::lol::lol:
Parent - By Bob Durrett (***) Date 2012-10-29 17:04

> In any case, Computers for correspondance shouldn't be solely relied upon, because their positional understanding is most of the time over the search horizon.
> And as a result can fall down into holes in static positions they don't or can't play well when there is nothing to calculate.


I suppose this problem can be reduced to some extent by going to positions downstream in a line and then working backwards.

There may be a lot of room for improvement in how the computer uses hash tables. For example, when doing as in the above sentence, what is going on in the hash table(s)? Maybe things could be done better.

Bob Durrett
Parent - By NATIONAL12 (Gold) [gb] Date 2012-10-30 23:52
IMHO no engine understands positional chess.I know of no authors who claim anything on this,its just your perspective on the play which is based on search and eval.

atm Houdini is winning this with a narrower search than Rybka and a very fast exe which Rybka cannot match.

If Houdini was so perfect it would never lose a game against any engine,however this is simply untrue.

Dont get me wrong H3 is a superb product which i will be buying in next week or so.
Parent - By Razor (****) [gb] Date 2012-10-29 06:30
A program can be made to play any legal move.

My advice, if you want interesting then just play OTB chess against human opposition and study the game afterwards using your silicon friends.  Your games will have much more variety and as they are entirely 'your games', your study time with your silicon friends will be more focused on discovering 'opportunities to improve'; you get the added bonus of meeting lots of people too!  :smile:
Parent - - By Geomusic (*****) Date 2012-10-29 06:51
Does anyone play Corr chess without engines? It's interesting, but I would like to see some real chess not engine guided centaurism.
Parent - By Aliven (***) [us] Date 2012-10-29 06:55
I believe the correspondence chess on www.chess.com is for non-engine play.

Correct me if I'm wrong.
Parent - - By Bob Durrett (***) Date 2012-10-29 10:30
Engine guided chess is not chess at all. Only brain dead people play engine guided chess.

Using an engine does not require that the chess be "engine guided."

As I noted in an earlier thread, chess using engines as carpenters use hammers and saws is a valid variant of chess which can be enjoyed.

Restriction of use of chess engines to post-mortem analysis only is an unnecessary restriction.

On the other hand, playing chess at any time limits but without the use of engines &/or databases remains a popular form of chess.

Chess variants are not bad just because they are not the "original" form of chess.

Bob Durrett
Parent - - By leavenfish (***) [us] Date 2012-10-30 05:08 Edited 2012-10-30 05:11
I would agree that "brain dead" is too strong.
However, "using engines as carpenters use hammers and saws"...is indeed precisely a 'variant' - NOT real correspondence chess, a 'variant' of it. It is no more a 'competition' than someone with hypothetical Gun A against someone with hypothetical Gun B at a hunting meet when these hypothetical guns are the ones doing the vast majority of the work and are there to 'correct you' when you point it in the wrong direction. THAT is going to happen far more often than someone pointing the gun properly at a target. Similarly, someone piloting an engine is in the vast majority of cases actually being piloted by that engine when you consider the actual number of positions looked at during a game. It's doing the overwhelming amount of work.

These things make people with just ordinary chess skill look like Kasparovs. It's an illusion IMHO. But everyone is entitled to their illusions I guess.
Parent - - By Bob Durrett (***) Date 2012-10-30 11:19 Edited 2012-10-30 11:30
Imagine a hunter who refused to use a gun because he felt that "real" hunters only used spears or bows and arrows!

The "original" chess, according to the chess history books, did not allow capturing enpassant, castling and pawn promotion. Pieces could only move one square at a time.

Then, as I recall, in the late 1400s in France, some outrageously radical weirdoes started playing a variant of the "real" chess. The variant allowed capturing enpassant, castling and pawn promotion and the pieces were allowed to move more than one square at a time.

I am sure there were those who objected that the new variant was "no good" because it was not "real" chess!

What happened? The new variant became very popular whereas the original version fell into disuse.

Today we think of this radical new variant as being the "real" chess.

The same applies to correspondence chess too. Older chess players could view modern server based correspondence chess as being an abomination.

I guess I could be regarded as an "older correspondence chess player" but I find the servers to be extremely convenient. I really don't want to mess with those post cards.

We did not always have chess software. Those who resist change resist this new development and long for the "good old days" when there were no computers at all.

Modern hunters do use guns and I see nothing wrong with that.

Bob Durrett
Parent - - By leavenfish (***) [us] Date 2012-10-31 00:36
"The "original" chess.....etc".

Bob...the difference is none of these 'variants'  made use of 'generated moves'. When any person vs person competition is replaced by person with the biggest, baddest computer and engine vs someone attempting to do the same, it's a variant of the game of chess. Computer aided chess died when Kasparov gave it up. "Advanced Chess", I think it was called. He used it to get himself some $$$$ and publicity...and rightly let it die. It's a thing of the past to all but computer geeks (hey, I'm one myself in a sense) who for their own individual reasons like to undertake it. To the rest of the world, it is dead and holds no interested but to those I just mentioned.  RIP.

"The same applies to correspondence chess too. Older chess players could view modern server based correspondence chess as being an abomination."

No, 'server based correspondence chess'...is just that. Servers are nothing but places for storing information...in this case, the games in progress. That is all they do. I do agree with you though that those old post card days are gladly gone. Long live servers.
Parent - - By Bob Durrett (***) Date 2012-10-31 00:55

> it's a variant of the game of chess.


That is true.

But if that particular variant becomes popular, then it should be considered a "legitimate" game. A game, after all, is just a way to make people happy. What is wrong with that?

I exclude ugly games which are of an anti-social nature such as a competition between two shooters who try to see which one can shoot the most people.

> died when Kasparov gave it up. "Advanced Chess", I think it was called.


Kasparov gave up more than "Advanced Chess." He gave up your favorite version of chess too, but OTB chess has not died. What Kasparov did or quit doing is interesting but he did not kill anything IMHO. I think Kasparov "screwed up."

I see the use of chess software in correspondence chess as far more relevant than its use in a variant of OTB chess.

Bob Durrett
Parent - - By leavenfish (***) [us] Date 2012-11-01 00:06
Tossing a string along side a cat to see if it can grab it before you can whisk it away can be considered a 'legitimate game' I guess, you I agree. Nothing wrong with either, it's a competition.

Not only did Kaspy give up Advanced Chess (it was just marketing tool and quick $$), everyone has...organizers and promoters. Like I said, it is dead.

It only lives in the world of those I mentioned...with no intrest on the outside. Rather like the game I play with my cat...
Parent - By Bob Durrett (***) Date 2012-11-01 02:54
I agree if you are talking about OTB with conventional time limits but with the availability of computers.

I would not agree if you were talking about correspondence chess since it appears that correspondence chess using chess software is becoming very popular worldwide.

If the day ever comes that chess playing computers become 3 or more orders of magnitude (base 10) faster, then Advanced Chess may come back.

We should talk more about correspondence chess where the time is available for the chess engine to reach good search depths.

Incidentally, I would love to have a much faster chess computer, but primarily for post mortem analyses and for analyses of GM games.

Bob Durrett
Parent - - By Master Om (Bronze) [in] Date 2012-11-06 12:49
What u r talking is shatranj or chaturanga which originated from india where it represented the war. Pawns representing pyaada or soldiers, bishops representing senapati or head of soldiers , Knights as horses, rook as elephants and Queen as Wazir or the minister. Only one house for pawns per move and no castling concepts .Hence indian name to openings is still there. And pawn promotion was for the piece of the house of pawn i.e. queen pawn reaches queen comes out, rookh pawn brings out rook and with exception to king pawn. I still dont understand what the heck QUEEN HAS TO DO IN CHESS???
Parent - - By Bob Durrett (***) Date 2012-11-06 14:44
I think women have been in the world as long as men have. :-)

Agreed that women probably had no place in the wars you mentioned other than to be spoils of war. In any case, the "queen" moved only one square at a time. Perhaps in that original version the piece we call a queen actually represented something other than a woman? What is a wazir?

Incidentally, I wonder whether or not anybody still plays that game.

Maybe the originator of that game had a wife or concubine or lover who talked him into including a woman?

We are assuming that the game was made up by a man. Could it be, instead, that it was a woman who gave birth to that game?

Bob Durrett
Parent - By Chess_Rambo (***) [at] Date 2012-11-06 15:26

> What is a wazir?


An advisor. http://en.wikipedia.org/wiki/Vizier

In Xiangqi (Chinese Chess) there are two advisors and there is no queen.
Parent - By Master Om (Bronze) [eu] Date 2012-11-06 16:05
Wazir is in persian and in sanskrit is 'Mantri' or minister who is next in power in battle after king. There is no place of women in battle as they are the cause of war!! :smile:
We indians still play it in villages as shatranj. As the story goes it was first devised by a brahmin named sisa on the order of king in order to replace 'pasa' which is a gambling play with dice, but with similar addiction. He came up with a war game on a chaturanga means  8x8 board them used to make maps. Legend says king was so happy that he asked sisa a prize. He said he said he needs a grain for every square and doubled every day for 64 days. Amazingly king found no rice in his godown!! In eight days.
When persian came india they took algebra and shatranj and modified it. Now queen was wazir, pawn was pheela, bishops were oont or camel, knights were ghoda or horses and rook as rookh or elephant. When britishers came they knew about arithematic and shatranj and changed it to chess modifying it and adding dumb queen. We indians give importance to women only inside house and not outside it.
Parent - By Uly (Gold) [mx] Date 2012-11-07 06:43

> Imagine a hunter who refused to use a gun because he felt that "real" hunters only used spears or bows and arrows!


No, no, spears and bows and arrows are tools equivalent to chess engines, real men hunt beasts with their bare hands!
Parent - By Moz (****) Date 2012-10-29 07:21
It's always going to be a problem unless there's some paradigm shift in chess software. The move that scores best will always get the most attention and the recommendation. Engines always have trouble evaluating any position where it's necessary to take one step back, in order to take two steps forward.

Fortunately for people that enjoy correspondence chess, they don't understand the benefits of complications, so there's still a valuable role for humans to play.
Parent - - By Siren (*) Date 2012-10-30 23:22
There's just no way for engines to examine all possible variations because of hardware speed limitations and pruning quality limitations, so they have to make cuts to the tree.. large cuts.. Actually, there's more they dont see than they do see. However, most of what they dont see is not worth seeing anyway. I think overall, these machines "see" more than what the top humans see on average. If you go over top GM games with a microscope like Houdini, you'll see they often miss a lot of opportunities to capitalise in many instances. Why? Because humans dont see everything either. And that's part of life.
Parent - By Bob Durrett (***) Date 2012-10-30 23:52 Edited 2012-10-31 00:17
Several of the correspondence chess players here have expressed their belief that always playing the move found by the chess engine is a good way to lose correspondence chess games.

One person, a correspondence chess GM, said that doing so would result in too many draws.

Either way, if you wish to win you must not allow yourself to become a slave of your chess engine. Not even if it is Houdini 3.

Bob Durrett
Parent - - By titanium cranium (***) [us] Date 2012-11-05 17:50
Bob,
While it's true that engines do not play perfectly, it's also true they play in the +3300 ELO range, or 500 pts above GM. They may miss some tactics but more than make up for it in overall play.
Parent - - By Bob Durrett (***) Date 2012-11-05 19:35
Well, in any case I am very happy with my new Houdini 3 software. Nevertheless, I can't wait for the next leap forward in chess engine strength. Ultimately, I want an engine several orders of magnitude stronger and faster than Houdini 3.  :-)

Houdart, take note!

Bob Durrett
Parent - - By titanium cranium (***) [us] Date 2012-11-05 20:40
you just need better computer hardware to produce 30-40,000 kn/s
Parent - By Bob Durrett (***) Date 2012-11-05 21:45
I want both! Six or eight orders of magnitude would be even better! Three for the software and three for the hardware. :-)

Bob Durrett
Parent - By Uly (Gold) [mx] Date 2012-11-07 06:46
It's not a "problem", that human guidance can still find better moves than what engines suggest, no matter how strong engines are, is what makes the sport interesting. If any idea that you could have in the position was already covered by chess engines without space for your own ideas, the whole exercise would be pointless.
- - By kevsamiga (*) [gb] Date 2012-11-05 16:26
On another note, I'm curious. Is there any engine out there either WB or UCI that uses *no* selectivity at all and just brute force searches everything to the maximum depth it
can using only basic alpha-beta and nullmove to save examining useless moves, so that ALL moves are checked without missing anything important that having selectivity would
negatively affect ?!?
Parent - By dgordon (*) [ca] Date 2013-01-03 13:24
Re. brute force engines, Rybka Winfinder was a tactical problem solver engine.  Houdini 3 has a tactical mode which spends much of its time analyzing moves unlikely to bear much fruit.  One thing that people were remarking on with Houdini's tactical mode is that it wasn't that much weaker than the regular Houdini even though it was brute force searching.

A lot of engine gaffs seem related to a poorly developed sense of danger.  A human looks at a position, and thinks 'there's a pawn storm coming on the kingside,' whereas Houdini will quite happily hunt for tactical shots until it gets mated.  King safety evaluation needs to be improved.

Another area where humans can have an impact is in keeping the engine focused on lines that seem promising.  Pretty well all engines prune some lines that ultimately turn out to be best, so even in tactics, they make mistakes.  As a chess player though, you can feed your engine the first few moves of a critical line until it finally sees what you're getting at, and its evaluation changes to favour that line.  If the engine is bad at backward propagation though, when you move back up the line, it may end up losing the critical moves, and going back to its old first choice, so you have to keep an eye on it, and store the critical analysis in some form.
Up Topic The Rybka Lounge / Computer Chess / The Problem with Chess Engines

Powered by mwForum 2.27.4 © 1999-2012 Markus Wichitill