Page 2 of 5
Re: Where did all these 3400 engines come from?
Posted: Sun Aug 18, 2019 6:51 pm
by mclane
Chess is not about elo but about mating the opponent.
There is no chess rule that says the game is won because you lead by +3 or your elo is better.
What do you do if you are ahead 3 pawns but the opponent is not resigning ?
You try to win another material ? Or to mate ?
Why not plan the mate.
Move the pieces . Create a mate net.
Even sac material for that mate.
Re: Where did all these 3400 engines come from?
Posted: Sun Aug 18, 2019 7:26 pm
by zullil
mclane wrote: ↑Sun Aug 18, 2019 6:29 pm
IMO even those strong chess programs have no clue about chess and run around waiting for the opponent to make a mistake.
We have so many draws in computerchess because the engines have no plan.
There are many draws in computer chess because chess is a theoretical draw and engines, unlike humans, make very few horrific errors. Humans suck at chess. We've sucked at it long enough that the chess literature is full of "brilliant" wins (i.e., horrific blunder-filled losses).
If chess engines had been invented in time, humans might have realized how much they sucked at chess and stopped playing altogether.
Re: Where did all these 3400 engines come from?
Posted: Sun Aug 18, 2019 7:43 pm
by hgm
OTOH, those so-called top engines suck at Chess too: fortress draws usually go undetected, even when they are reglementary draws, they do not detect perpetual trapping, they are very poor at recognizing winning lines that have multiple zugzwangs... While for humans this is often easy.
So Elo only correlates weakly with Chess capability. This is also demonstrated by the NN engines, which have not much trouble beating these high-Elo conventional engines, despite performing quite poorly themselves on test suites.
Re: Where did all these 3400 engines come from?
Posted: Sun Aug 18, 2019 7:49 pm
by bob
jdart wrote: ↑Sun Aug 18, 2019 4:50 pm
Code: Select all
But do they play really chess ? I mean, do they have a plan or do they make moves that do not lose material?!
Well, they don't usually lose material to shallow tactics. The NN engines used to do that but they have improved to the point where mostly they don't.
All these programs rely on heuristics: techniques that, on the average over many games, lead to good results.
There is usually no plan in the sense that is easy to explain to a human.
Actually chess programs DO have a plan. Win material and checkmate the opponent.
Re: Where did all these 3400 engines come from?
Posted: Sun Aug 18, 2019 7:50 pm
by zullil
hgm wrote: ↑Sun Aug 18, 2019 7:43 pm
OTOH, those so-called top engines suck at Chess too: fortress draws usually go undetected, even when they are reglementary draws, they do not detect perpetual trapping, they are very poor at recognizing winning lines that have multiple zugzwangs... While for humans this is often easy.
So Elo only correlates weakly with Chess capability. This is also demonstrated by the NN engines, which have not much trouble beating these high-Elo conventional engines, despite performing quite poorly themselves on test suites.
Play a match against Stockfish and see how many fortress draws you get. Pablo seems to have spent his life mastering this skill. Maybe talk to him.
I do appreciate your point. Stockfish still needs depth in the 60's to find mate-in-5 here:
[d]rk6/p1r3p1/P3B1Kp/1p2B3/8/8/8/8 w - - 0 1
viewtopic.php?f=2&t=66577&start=10#p753058
Re: Where did all these 3400 engines come from?
Posted: Sun Aug 18, 2019 8:21 pm
by mar
zullil wrote: ↑Sun Aug 18, 2019 7:50 pm
I do appreciate your point. Stockfish still needs depth in the 60's to find mate-in-5 here:
[d]rk6/p1r3p1/P3B1Kp/1p2B3/8/8/8/8 w - - 0 1
This seems to be one of the positions where disabling nullmove helps,
I bet SF would then see the mate immediately.
Re: Where did all these 3400 engines come from?
Posted: Sun Aug 18, 2019 8:28 pm
by Rebel
mclane wrote: ↑Sun Aug 18, 2019 6:29 pm
The idea behind chess is to mate. For that you need a plan. Otherwise you run around on the board without understanding and IMO you do not play chess but survive.
If an engine plays 2600+ ELO but is not planning to mate, it has IMO no understanding of the game.
Ok it plays 2600 elo.
But is chess running around and winning because the opponent resigns or has not enough material or is chess the art of mating the opponent ?
IMO even those strong chess programs have no clue about chess and run around waiting for the opponent to make a mistake.
We have so many draws in computerchess because the engines have no plan.
Yes humans have plans.
Why would machines NOT have a plan ??
It’s Intelligent to have a plan for the day.
Machines having NO plan are stupid.
IMO the next step in computerchess are programs that plan what to do to mate from the given position.
It’s what intelligent entities do. No matter if dog, human being, alien or AI machines.
Chess in essence is search, if you can think 50 moves ahead you don't need a plan.
Fortunately we can't, hence humans need a plan, comps don't.
Re: Where did all these 3400 engines come from?
Posted: Sun Aug 18, 2019 8:45 pm
by mclane
Ed, ist difficult for me to disagree with you. Your chess programs dominated computerchess from 1986 until whatever.
My full respect for this. As usual.
But i want to disagree because I think we need another level now.
We need chess engines doing a plan.
IMO its bringing computerchess into another level.
It is said humans need a plan because they can’t search 40 searches deep like stockfish.
But humans have eyes and can see while stockfish needs 40 searches to identify.
Planning is IMO the next step.
Why not trying it out.
We have nothing to lose.
Re: Where did all these 3400 engines come from?
Posted: Sun Aug 18, 2019 8:48 pm
by Uri Blass
mclane wrote: ↑Sun Aug 18, 2019 6:29 pm
The idea behind chess is to mate. For that you need a plan. Otherwise you run around on the board without understanding and IMO you do not play chess but survive.
If an engine plays 2600+ ELO but is not planning to mate, it has IMO no understanding of the game.
Ok it plays 2600 elo.
But is chess running around and winning because the opponent resigns or has not enough material or is chess the art of mating the opponent ?
IMO even those strong chess programs have no clue about chess and run around waiting for the opponent to make a mistake.
We have so many draws in computerchess because the engines have no plan.
Yes humans have plans.
Why would machines NOT have a plan ??
It’s Intelligent to have a plan for the day.
Machines having NO plan are stupid.
IMO the next step in computerchess are programs that plan what to do to mate from the given position.
It’s what intelligent entities do. No matter if dog, human being, alien or AI machines.
2600 engines knows to mate even if the opponent does not resign so I disagree.
plan what to do to mate is not something that humans do in every chess positions.
There are cases when humans have no idea how to mate and they plan to win material because they believe that if they have material advantage there is a good chance that they will be able to mate the opponent and it is impossible to win in another way.
They do not need to see how they mate when they win material.
suppose black play against a weak player and the game begins
1.Nf3 c5 2.d3 e5 3.Nxe5??
[d]rnbqkbnr/pp1p1ppp/2p5/4N3/8/3P4/PPP1PPPP/RNBQKB1R b KQkq - 0 3
Black plays Qa5+
Does black has a plan how to mate?
No.
The plan of black is win a knight and later think what to do with the idea that there are good chances that black is going to find a way to mate
later.
Re: Where did all these 3400 engines come from?
Posted: Sun Aug 18, 2019 9:00 pm
by Uri Blass
mclane wrote: ↑Sun Aug 18, 2019 8:45 pm
Ed, ist difficult for me to disagree with you. Your chess programs dominated computerchess from 1986 until whatever.
My full respect for this. As usual.
But i want to disagree because I think we need another level now.
We need chess engines doing a plan.
IMO its bringing computerchess into another level.
It is said humans need a plan because they can’t search 40 searches deep like stockfish.
But humans have eyes and can see while stockfish needs 40 searches to identify.
Planning is IMO the next step.
Why not trying it out.
We have nothing to lose.
I think that chess engines are already at high level that humans do not understand them.
We do not need stronger engines for humans but need something that humans can understand.
For example humans can understand problems like winning material in X moves.
Engines do not understand it.
If I give them to analyze they may say +4.3 but we do not know based on their score how much of the 4.3 is material and how much is positional score and we do not know how many plies forward they calculated.
I would like to see for analysis information like white wins 2 pawns based on 10 plies search with no pruning and extensions except extending non quiet positions(meaning extending captures after 10 plies when they are possible).