When will the chess programmers write an engine that plans ?

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: 17740
Joined: Thu Mar 09, 2006 5:40 pm
Location: US of Europe, germany
Full name: Thorsten Czub
Contact:

Re: When will the chess programmers write an engine that plans ?

Post by mclane » Sat Mar 21, 2020 4:44 pm

Chess is NOT all tactical.

E.g. LC0 plays very good chess without playing tactical chess.

You ask what it is good for.

So what is it good for to understand chess if you play it.

Right ?

You believe playing chess makes sense if the programs do rely on the search tree , millions of NPS and search depths of 40 plies ?
What seems like a fairy tale today may be reality tomorrow.
Here we have a fairy tale of the day after tomorrow....

JohnW
Posts: 201
Joined: Wed Nov 21, 2012 11:20 pm
Location: New Hampshire

Re: When will the chess programmers write an engine that plans ?

Post by JohnW » Sat Mar 21, 2020 5:07 pm

mclane wrote:
Sat Mar 21, 2020 4:44 pm
Chess is NOT all tactical.

E.g. LC0 plays very good chess without playing tactical chess.

You ask what it is good for.

So what is it good for to understand chess if you play it.

Right ?

You believe playing chess makes sense if the programs do rely on the search tree , millions of NPS and search depths of 40 plies ?
It's not all tactics? What moves could LC0 come up with that is not possible to see tactically?
It's good for me to understand chess if I play it because I am not a computer. It doesn't make any difference if I lost to a program that planned or lost to a program that played tactically. I still lost, they both played by the same rules.

bob
Posts: 20916
Joined: Mon Feb 27, 2006 6:30 pm
Location: Birmingham, AL

Re: When will the chess programmers write an engine that plans ?

Post by bob » Sat Mar 21, 2020 5:45 pm

Here's a question:

Suppose you are in the market for building a house. You have a pretty good idea of what you want. You find a guy that listens to you, and then builds exactly what you asked for. No blueprints, no drawings, he just builds it and ends up with a result you are happy with.

OR, you find a guy that listens, draws up blueprints, works out materials required, schedules the various crews. And you end up with EXACTLY the same house at the same cost.

Which is better, the one built without a plan or the one built with? Humans can't beat today's programs. So does it REALLY matter how the programs get to the end result (winning games)? This planning topic was a hot one back in the 70's, 80's and 90's. Not so much any more. For example, HOW can you tell an engine doesn't have any sort of plan? How can you tell that a human does? By asking either the human or the author? Is the lack of a plan obvious? IE the computer seems to be preparing to pile up on e5, but then it switches to f6. Is that a lack of planning by the program? Or perhaps a positional mistake by the opponent that can better be exploited with a "different plan"?

This is a common criticism that really makes no sense. If the goal is to build the fastest boat on the lake, does it really matter how you get there, so long as you DO get there? Fastest is fastest...

BTW chess is ALL about tactics. "positional play" is simply something used by humans that can't see deep enough to work out all the tactics. IE try to win a game by losing material when your opponent does not. The game will ultimately be solved by tactics, just as checkers was. Not by "positional play" which is pointless in the face of perfect tactics.

User avatar
mclane
Posts: 17740
Joined: Thu Mar 09, 2006 5:40 pm
Location: US of Europe, germany
Full name: Thorsten Czub
Contact:

Re: When will the chess programmers write an engine that plans ?

Post by mclane » Sat Mar 21, 2020 7:00 pm

Chess is NOT all about tactics.
In most positions there is absolutely no tactics.
What seems like a fairy tale today may be reality tomorrow.
Here we have a fairy tale of the day after tomorrow....

JohnW
Posts: 201
Joined: Wed Nov 21, 2012 11:20 pm
Location: New Hampshire

Re: When will the chess programmers write an engine that plans ?

Post by JohnW » Sat Mar 21, 2020 8:11 pm

mclane wrote:
Sat Mar 21, 2020 7:00 pm
Chess is NOT all about tactics.
In most positions there is absolutely no tactics.
If that's the case, how do engines that are based on tactics determine a move?

jonkr
Posts: 68
Joined: Wed Nov 13, 2019 12:36 am
Full name: Jonathan Kreuzer

Re: When will the chess programmers write an engine that plans ?

Post by jonkr » Sat Mar 21, 2020 8:18 pm

One thing I found interesting about modern search with the high search depths is that what it's doing seems actually more similar to how a human approaches chess to me. LMR/LMP and other reductions are usually tuned pretty aggressive, and there really are a lot of lines that will be reduced to almost nothing because they are evaluated early on as unlikely to be good.

So search is concentrated on lines the program evaluates as promising, and the search of earlier positions also gets widened as search depths increase to include more moves to better verify the pv line is actually good. So maybe it sees a move supports a passed pawn advance, then couple ply later that it can advance the pawn, then can advance again, so keeps this line with an increasing eval. Or that this move increases pressure on a weak piece, then next move can increase pressure more, then next move can capture the piece. Maybe it finds a refutation then switches to the next-most promising line, and expands that since it's now scored the best.

I don't really have an opinion on whether or not that could be called planning, and wouldn't even object if planning was defined as something only a human can do, but there are some similarities.

Evaluation and evaluation fine tuning was also more important than I initially expected (partially because it helps direct search.) In my program evals can occasionally get to 3 or 4 without material gain (and without significant passed pawn or immediate king safety issues.) So in that sense it's more than twice as positional as my 2005 version. In theory the eval should lead to material gain or checkmate, although there isn't always something concrete, then the program looks a bit silly. But play is still much stronger and I think more interesting than the 2005 version.

User avatar
mclane
Posts: 17740
Joined: Thu Mar 09, 2006 5:40 pm
Location: US of Europe, germany
Full name: Thorsten Czub
Contact:

Re: When will the chess programmers write an engine that plans ?

Post by mclane » Sat Mar 21, 2020 8:24 pm

JohnW wrote:
Sat Mar 21, 2020 8:11 pm
mclane wrote:
Sat Mar 21, 2020 7:00 pm
Chess is NOT all about tactics.
In most positions there is absolutely no tactics.
If that's the case, how do engines that are based on tactics determine a move?
They do not determine. That’s the problem. They make a move.
Because they have to.
That’s why they lose against LC0.
What seems like a fairy tale today may be reality tomorrow.
Here we have a fairy tale of the day after tomorrow....

User avatar
Ovyron
Posts: 4392
Joined: Tue Jul 03, 2007 2:30 am

Re: When will the chess programmers write an engine that plans ?

Post by Ovyron » Sat Mar 21, 2020 8:47 pm

MonteCarlo wrote:
Sat Mar 21, 2020 3:52 pm
In all seriousness, the claim as it stands is clearly wrong.
Indeed, and the opposite is true: Software has increased much more significantly than hardware.

How do I know? Because I've had the same computer since 2009, and 8 years ago I was completely lost against the top hardware of that era, losing most of my games, with rare draws, and no wins. Currently it's a totally different picture, with me drawing most of my games and losing rarely, and getting a few wins. And this is against people that have gone through 8 years of hardware improvements! So how's this possible? Because software has improved so much that it nullifies hardware improvements, so the gap has been abridged and it's way more important what book you have than your hardware.

The improvement has been ridiculous, just fire up Stockfish 8 against 11 and see how the former plays like a patzer...

acase
Posts: 734
Joined: Wed Mar 15, 2006 12:14 am
Location: Columbus, Ohio USA
Full name: Andrew R Case

Re: When will the chess programmers write an engine that plans ?

Post by acase » Sat Mar 21, 2020 9:19 pm

mclane wrote:
Fri Mar 20, 2020 10:22 pm
I mean that creates a plan and develops a main line that leads to something.
Not the usual engines we have today. That play chess within a Horizont of search depth.
Because Tactics are FAR more important than plans and strategy. If you (or a computer) make a strategical mistake against an opponent you could possibly save the game and not lose, however if you make one tactical mistake it could lose the game for you immediately. This is absolutely why tactics must be the main focus of all computer chess.

JohnW
Posts: 201
Joined: Wed Nov 21, 2012 11:20 pm
Location: New Hampshire

Re: When will the chess programmers write an engine that plans ?

Post by JohnW » Sat Mar 21, 2020 9:50 pm

mclane wrote:
Sat Mar 21, 2020 8:24 pm
JohnW wrote:
Sat Mar 21, 2020 8:11 pm
mclane wrote:
Sat Mar 21, 2020 7:00 pm
Chess is NOT all about tactics.
In most positions there is absolutely no tactics.
If that's the case, how do engines that are based on tactics determine a move?
They do not determine. That’s the problem. They make a move.
Because they have to.
That’s why they lose against LC0.
Isn't their move choice determined by applying a score to a position? I mean if it wasn't determined then their choice would have to be by randomly selecting an available move which would result in hanging pieces.

Post Reply