Lc0 51010

Discussion of anything and everything relating to chess playing software and machines.

Moderators: bob, hgm, Harvey Williamson

Forum rules
This textbox is used to restore diagrams posted with the [d] tag before the upgrade.
User avatar
mclane
Posts: 16197
Joined: Thu Mar 09, 2006 5:40 pm
Location: US of Europe, germany
Full name: Thorsten Czub
Contact:

Re: Lc0 51010

Post by mclane » Sat Oct 05, 2019 10:58 am

jp wrote:
Sat Apr 06, 2019 12:37 am
mclane wrote:
Fri Apr 05, 2019 1:26 pm
Exactly. Stockfish is artificial stupidity. It concentrates on search instead of knowledge. Take away the millions and millions of NPS and you see the artificial stupidity.

It is more a pocket calculator with enormous power then an entity that can play chess.
But all programs, including NN engines, rely on "millions and millions" of operations to play very strong chess. It sounds like you think number of nodes is the correct measure of how much calculation it's doing. That's not correct. So if you think doing lots of calculations is "artificial stupidity", then all engines are artificial stupidity. Why single out SF? We can agree that all chess engines are not AI.
If you believe it would be needed millions of millions of NPS to play chess you are wrong.
You can play chess with a few NPS. And you can play stronger chess with few NPS if you are an experienced GM.

What is the sense of all these millions of millions of NPS if it is not bringing chess and AI into a new step, a new level of intelligence ?

Therefore I think this approach is the wrong way.
NOT MORE NPS gives better chess.
The quality of the chosen moves must be increased. And this can (IMO) only be done if the chosen and followed NPS follow a plan.
The machine needs a plan.

In the Moment the Chess programs solve test suites.

For them there is no difference between solving nolot, BT2630 or colditz Test or bratko Kopec Test or a game of chess against a human Beeing.

It’s all the same solving a chess position.
And that cannot be doing chess.

To reach the next level in computerchess the machines need to create plans and search those NPS that helps to realize those plans on the chess board. It’s not important if the plan works or if the ELO is high or low.

But the machines need a plan. Otherwise they run arround like a blind man in a big city.

Of course they will find tactical combinations. And if they see a mate they will also mate.

But this is not chess.

Chess is not doing a move and hoping not to make a mistake and by change win a knight or by chance see a mate and win.
Chess is to create a plan to win the game.


Of course you can ignore this and continue to program a chess engine that increases strength from 3320 to 3325 and in a year reaching 3350.

But what is it good for ??
What seems like a fairy tale today may be reality tomorrow.
Here we have a fairy tale of the day after tomorrow....

jp
Posts: 825
Joined: Mon Apr 23, 2018 5:54 am

Re: Lc0 51010

Post by jp » Sat Oct 05, 2019 7:04 pm

mclane wrote:
Sat Oct 05, 2019 10:58 am
If you believe it would be needed millions of millions of NPS to play chess you are wrong.
You can play chess with a few NPS. And you can play stronger chess with few NPS if you are an experienced GM.

What is the sense of all these millions of millions of NPS if it is not bringing chess and AI into a new step, a new level of intelligence ?
<snip>

Of course you can ignore this and continue to program a chess engine that increases strength from 3320 to 3325 and in a year reaching 3350.

But what is it good for ??
I'm sure programmers would love to create an engine that didn't calculate much. It's just that no one knows how to do that, not that they don't want to.

Asking them what it's good for isn't helping them to come up with what you want.

zullil
Posts: 5670
Joined: Mon Jan 08, 2007 11:31 pm
Location: PA USA
Full name: Louis Zulli

Re: Lc0 51010

Post by zullil » Sat Oct 05, 2019 7:39 pm

mclane wrote:
Sat Oct 05, 2019 10:58 am

Chess is to create a plan to win the game.
You sound like a human. Maybe chess is simply to search a move tree. Or perhaps to perform "image processing" on chess diagrams and find intricate patterns that training has said are good.

Humans make plans. And they suck at chess. :D

duncan
Posts: 10388
Joined: Mon Jul 07, 2008 8:50 pm

Re: Lc0 51010

Post by duncan » Sat Oct 05, 2019 8:22 pm

mclane wrote:
Sat Oct 05, 2019 10:58 am


Chess is to create a plan to win the game.

The chess championship is decided by the one who wins the games. Not the one who deserved to win the game by having the best plan. As an idealist you may want a rule change. That is not what we have at the moment. Same with all sports. Bolt is the best sprinter even though he has been described as having the elegance of a cupboard and there are others more deserving with far better style who should have won.

User avatar
mclane
Posts: 16197
Joined: Thu Mar 09, 2006 5:40 pm
Location: US of Europe, germany
Full name: Thorsten Czub
Contact:

Re: Lc0 51010

Post by mclane » Sat Oct 05, 2019 9:23 pm

jp wrote:
Sat Oct 05, 2019 7:04 pm
mclane wrote:
Sat Oct 05, 2019 10:58 am
If you believe it would be needed millions of millions of NPS to play chess you are wrong.
You can play chess with a few NPS. And you can play stronger chess with few NPS if you are an experienced GM.

What is the sense of all these millions of millions of NPS if it is not bringing chess and AI into a new step, a new level of intelligence ?
<snip>

Of course you can ignore this and continue to program a chess engine that increases strength from 3320 to 3325 and in a year reaching 3350.

But what is it good for ??
I'm sure programmers would love to create an engine that didn't calculate much. It's just that no one knows how to do that, not that they don't want to.

Asking them what it's good for isn't helping them to come up with what you want.
And that is why the quality and the paradigm has to change.
I am sure there is a way.
Maybe we have to go where botwinnik was trying to go.
What seems like a fairy tale today may be reality tomorrow.
Here we have a fairy tale of the day after tomorrow....

User avatar
mclane
Posts: 16197
Joined: Thu Mar 09, 2006 5:40 pm
Location: US of Europe, germany
Full name: Thorsten Czub
Contact:

Re: Lc0 51010

Post by mclane » Sat Oct 05, 2019 9:28 pm

duncan wrote:
Sat Oct 05, 2019 8:22 pm
mclane wrote:
Sat Oct 05, 2019 10:58 am


Chess is to create a plan to win the game.

The chess championship is decided by the one who wins the games. Not the one who deserved to win the game by having the best plan. As an idealist you may want a rule change. That is not what we have at the moment. Same with all sports. Bolt is the best sprinter even though he has been described as having the elegance of a cupboard and there are others more deserving with far better style who should have won.
If you have a car that has 100 PS and drives good, it does not make much sense to drive another car that has 200 Ps or 300 Ps.
Of course the more Ps the faster you can drive. But it does not give you much more satisfaction or comfort.

The programmers try to make their programs 10-20 elo points stronger. But why ?
The programs are very strong.

Instead of programming the 10-20 elo points you could buy a new computer that is twice as fast, that would give the same progress.

If you don’t change the method, nothing will change at all.
What seems like a fairy tale today may be reality tomorrow.
Here we have a fairy tale of the day after tomorrow....

User avatar
mclane
Posts: 16197
Joined: Thu Mar 09, 2006 5:40 pm
Location: US of Europe, germany
Full name: Thorsten Czub
Contact:

Re: Lc0 51010

Post by mclane » Sat Oct 05, 2019 9:33 pm

zullil wrote:
Sat Oct 05, 2019 7:39 pm
mclane wrote:
Sat Oct 05, 2019 10:58 am

Chess is to create a plan to win the game.
You sound like a human. Maybe chess is simply to search a move tree. Or perhaps to perform "image processing" on chess diagrams and find intricate patterns that training has said are good.

Humans make plans. And they suck at chess. :D
If you think that doing a search tree that needs 98% nodes to find 2% important moves is the way to go... feel free to think so.
Any power plant is more efficient.
Yes I am a human,
And humans are more efficient when it comes to be efficient.

The method you suggest us an ineffective method,
It is only succesfully if you have much power.
What seems like a fairy tale today may be reality tomorrow.
Here we have a fairy tale of the day after tomorrow....

chrisw
Posts: 2142
Joined: Tue Apr 03, 2012 2:28 pm

Re: Lc0 51010

Post by chrisw » Sat Oct 05, 2019 10:14 pm

mclane wrote:
Sat Oct 05, 2019 10:58 am
jp wrote:
Sat Apr 06, 2019 12:37 am
mclane wrote:
Fri Apr 05, 2019 1:26 pm
Exactly. Stockfish is artificial stupidity. It concentrates on search instead of knowledge. Take away the millions and millions of NPS and you see the artificial stupidity.

It is more a pocket calculator with enormous power then an entity that can play chess.
But all programs, including NN engines, rely on "millions and millions" of operations to play very strong chess. It sounds like you think number of nodes is the correct measure of how much calculation it's doing. That's not correct. So if you think doing lots of calculations is "artificial stupidity", then all engines are artificial stupidity. Why single out SF? We can agree that all chess engines are not AI.
If you believe it would be needed millions of millions of NPS to play chess you are wrong.
You can play chess with a few NPS. And you can play stronger chess with few NPS if you are an experienced GM.

What is the sense of all these millions of millions of NPS if it is not bringing chess and AI into a new step, a new level of intelligence ?
is very likely all the AI “steps” that can be sucked out of chess have already been sucked. AI has moved on to way more interesting and challenging problems, leaving chess pretty much as a backwater.

Therefore I think this approach is the wrong way.
NOT MORE NPS gives better chess.
You drone on a lot about humans this and humans that. What actually humans do in decision making and categorising is operate to the principle of “this is good enough”, not “this is 100% correct”.
Chess engines are good enough. Sensible humans would stop there and put their resources into something else. Idiot humans can’t stop and instead try for an impossible and unnecessary perfection. When you get to 99.99999% it gets to be not too useful to try to get to 99.999999%
The quality of the chosen moves must be increased. And this can (IMO) only be done if the chosen and followed NPS follow a plan.
The machine needs a plan.

In the Moment the Chess programs solve test suites.

For them there is no difference between solving nolot, BT2630 or colditz Test or bratko Kopec Test or a game of chess against a human Beeing.

It’s all the same solving a chess position.
And that cannot be doing chess.

To reach the next level in computerchess the machines need to create plans and search those NPS that helps to realize those plans on the chess board. It’s not important if the plan works or if the ELO is high or low.

But the machines need a plan. Otherwise they run arround like a blind man in a big city.

Of course they will find tactical combinations. And if they see a mate they will also mate.

But this is not chess.

Chess is not doing a move and hoping not to make a mistake and by change win a knight or by chance see a mate and win.
Chess is to create a plan to win the game.


Of course you can ignore this and continue to program a chess engine that increases strength from 3320 to 3325 and in a year reaching 3350.

But what is it good for ??

jp
Posts: 825
Joined: Mon Apr 23, 2018 5:54 am

Re: Lc0 51010

Post by jp » Sat Oct 05, 2019 10:26 pm

chrisw wrote:
Sat Oct 05, 2019 10:14 pm
mclane wrote:
Sat Oct 05, 2019 10:58 am
What is the sense of all these millions of millions of NPS if it is not bringing chess and AI into a new step, a new level of intelligence ?
is very likely all the AI “steps” that can be sucked out of chess have already been sucked. AI has moved on to way more interesting and challenging problems, leaving chess pretty much as a backwater.
I'm not sure about calling it a "backwater", but yes, it has not been a part of AI for decades.

User avatar
mclane
Posts: 16197
Joined: Thu Mar 09, 2006 5:40 pm
Location: US of Europe, germany
Full name: Thorsten Czub
Contact:

Re: Lc0 51010

Post by mclane » Sat Oct 05, 2019 10:37 pm

Hi Chris. Thanks for calling me an idiot.

I remember that we once worked on a program that was your chess program.

While you gave up chess and concentrate to show that crafty copied Fruit, I am still interested in computerchess.
Ok.
This is of course because I am an idiot. I am an idiot not because I worked with you, as some people mentioned, but because I still do computerchess while you concentrate on something else.

IMO i see no sense in doing progress in search tree bringing Stockfish, komodo or Houdini from 3500 elo into 3520 elo in 2 months.

It makes no sense to me. I can understand that mark tries to give komodo another 20 elo increase , maybe some day he will be as strong as Stockfish is, but what is the sense of doing this other then to have an outcome.
The race is nice, but is is only working for feeding capitalism.


I would appreciate that the next level is started.

And IMO the next level is doing a plan.

A plan differs an engine that is doing a test suite from an engine that is trying out ideas,

An engine that has absolutely no idea what to do when there is no KEY MOVE to find will play like a blind man.
What seems like a fairy tale today may be reality tomorrow.
Here we have a fairy tale of the day after tomorrow....

Post Reply