Reply
 
LinkBack Thread Tools Display Modes
  #1   Report Post  
Old June 16th 16, 12:47 PM posted to rec.games.chess.misc
external usenet poster
 
First recorded activity by ChessBanter: Jul 2010
Posts: 82
Default Human-unfriendly chess software

Are there any chess programs that take into account the way humans reason? I would think that whether a move looks "counter-intuitive" could be programmed to some extent. Examples of counter-intuitive moves are sacrifices, moves that move pieces to less active squares, moves that revisit squares that were visited recently etc. All these concepts are readily programmable.. Then, in a losing position, the program could show a strong bias towards lines that the human can only win by playing a counter-intuitive move.

I know that computers are fantastically strong in relation to humans. However, I'm not sure how good they are at playing losing positions. If you give the best programs a terrible position to start with, I'm not sure how good they are at making the human work.

Please note that I am using the phrase "not sure" in the literal sense. "I'm not sure how good they are" is not a tactful way of saying "They are bad...." It includes the possibility that they may be excellent.

Thanks for your thoughts.

Paul
  #2   Report Post  
Old June 16th 16, 03:11 PM posted to rec.games.chess.misc
external usenet poster
 
First recorded activity by ChessBanter: Mar 2009
Posts: 3,536
Default Human-unfriendly chess software

On Thursday, June 16, 2016 at 2:47:43 PM UTC+3, Paul wrote:
Are there any chess programs that take into account the way humans reason? I would think that whether a move looks "counter-intuitive" could be programmed to some extent. Examples of counter-intuitive moves are sacrifices, moves that move pieces to less active squares, moves that revisit squares that were visited recently etc. All these concepts are readily programmable. Then, in a losing position, the program could show a strong bias towards lines that the human can only win by playing a counter-intuitive move.

I know that computers are fantastically strong in relation to humans. However, I'm not sure how good they are at playing losing positions. If you give the best programs a terrible position to start with, I'm not sure how good they are at making the human work.

Please note that I am using the phrase "not sure" in the literal sense. "I'm not sure how good they are" is not a tactful way of saying "They are bad..." It includes the possibility that they may be excellent.

Thanks for your thoughts.

Paul


Various chess programs have things like "personalities" that you can adjust.. Some will play reckless too, sacrificing 2 pawns for knight.

You can get various programs off the web and play them in "Arena Chess".

RL
  #3   Report Post  
Old June 16th 16, 03:21 PM posted to rec.games.chess.misc
external usenet poster
 
First recorded activity by ChessBanter: Feb 2009
Posts: 57
Default Human-unfriendly chess software

On 16/06/16 12:47, Paul wrote:
Are there any chess programs that take into account the way humans
reason? I would think that whether a move looks "counter-intuitive"
could be programmed to some extent. Examples of counter-intuitive
moves are sacrifices, moves that move pieces to less active squares,
moves that revisit squares that were visited recently etc. All these
concepts are readily programmable..


Not as readily as you might think. Computers look at positions
rather than moves, whereas humans look at moves rather than positions.
In a grossly-oversimplified model, the computer is looking at the set
of positions [say] 20 ply away, evaluating each of them [apart from
those eliminated by alpha-beta pruning], minimaxing those values, and
playing the first move down the resulting principal variation. If
the transposition table is to work [and it has to, because a typical
position 20 ply away can be reached in 10x10x9x9x8x8x... ways], then
the evaluation can't include history of how the position was reached.

Trying to program up an "If I go there, he can go there, but
then I could play that, ..." approach would enable your ideas, but
the resulting program would not be able to do a full search to an
interesting depth.

Then, in a losing position, the
program could show a strong bias towards lines that the human can
only win by playing a counter-intuitive move.
I know that computers are fantastically strong in relation to humans.
However, I'm not sure how good they are at playing losing positions.

[...]

I think one of the snags is that most [all?] current programs
have little idea of what it is that makes a position difficult to win.
Eg, if the program is two pawns down in a complicated position, it
may [will?] swap down into an obviously lost K&P ending one pawn down
if it can win a pawn by the swaps. Of course, if it sees far enough
ahead, then "one pawn down" translates into "and my opponent can push
that pawn through to promotion" and becomes a position to avoid.

--
Andy Walker,
Nottingham.
  #4   Report Post  
Old June 16th 16, 04:56 PM posted to rec.games.chess.misc
external usenet poster
 
First recorded activity by ChessBanter: Jun 2016
Posts: 2
Default Human-unfriendly chess software

PyChess which runs in Linux tries to play like a human at the lower
levels.
  #5   Report Post  
Old June 17th 16, 12:08 AM posted to rec.games.chess.misc
external usenet poster
 
First recorded activity by ChessBanter: Mar 2009
Posts: 3,536
Default Human-unfriendly chess software

On Thursday, June 16, 2016 at 5:21:57 PM UTC+3, Andy Walker wrote:
If
the transposition table is to work [and it has to, because a typical
position 20 ply away can be reached in 10x10x9x9x8x8x... ways], then
the evaluation can't include history of how the position was reached.


This can't be right. Though I understand transpositions, it's very unlikely that a position 20 ply away can be reached by so many different moves, since each move 'constrains' the position. There are more ways of reaching a position that's 1 ply away than 100 plys away.

It's like an opening: in the beginning, many transpositions, but after 30 moves, it's very unlikely you'll get a transposition.

What else don't you know Andy Walker?

RL


  #6   Report Post  
Old June 17th 16, 01:15 AM posted to rec.games.chess.misc
external usenet poster
 
First recorded activity by ChessBanter: Feb 2009
Posts: 57
Default Human-unfriendly chess software

On 17/06/16 00:08, raylopez99 wrote:
If
the transposition table is to work [and it has to, because a typical
position 20 ply away can be reached in 10x10x9x9x8x8x... ways], then
the evaluation can't include history of how the position was reached.

This can't be right. Though I understand transpositions,


That doesn't seem to be true.

it's very
unlikely that a position 20 ply away can be reached by so many
different moves, since each move 'constrains' the position. There
are more ways of reaching a position that's 1 ply away than 100 plys
away.


If the position is 1 ply away, there can be only one way to
reach it. On the other hand, the number of legally-possible games is
very much larger than the number of legally-possible positions, so
the number of ways of reaching each position must, on average, be
extremely large. In particular, the number of different games of
100 ply, order of magnitude 30^100, is much larger than the number
of legal positions, shown in "http://tromp.github.io/chess/chess.html"
to be less than 10^46. Go figure.

As I wrote, but you snipped, I was considering "a grossly-oversimplified model", in particular in which moves don't interfere
with each. This is not generally the case, though it is more often
the case in computer analysis than in actual games. To compensate
for this, in cases where [eg] it takes a queen two or three moves
to get from one square to another, there are often several routes
the queen could take; and if a move is wasted, there are usually
many ways in which it could be wasted to reach the same position.

It's like an opening: in the beginning, many transpositions, but
after 30 moves, it's very unlikely you'll get a transposition.


You're confusing the number of ways in which a position is
likely to be reached in a real game [of interest to humans but not
to computer analysis] with the number of routes in a given number
of moves from one given position to another, which is what matters
for the transposition table, and which is typically very large in
cases of interest.

--
Andy Walker,
Nottingham.
  #7   Report Post  
Old June 17th 16, 09:31 AM posted to rec.games.chess.misc
external usenet poster
 
First recorded activity by ChessBanter: Jul 2010
Posts: 82
Default Human-unfriendly chess software

On Thursday, June 16, 2016 at 3:21:57 PM UTC+1, Andy Walker wrote:
On 16/06/16 12:47, Paul wrote:
Are there any chess programs that take into account the way humans
reason? I would think that whether a move looks "counter-intuitive"
could be programmed to some extent. Examples of counter-intuitive
moves are sacrifices, moves that move pieces to less active squares,
moves that revisit squares that were visited recently etc. All these
concepts are readily programmable..


Not as readily as you might think. Computers look at positions
rather than moves, whereas humans look at moves rather than positions.
In a grossly-oversimplified model, the computer is looking at the set
of positions [say] 20 ply away, evaluating each of them [apart from
those eliminated by alpha-beta pruning], minimaxing those values, and
playing the first move down the resulting principal variation. If
the transposition table is to work [and it has to, because a typical
position 20 ply away can be reached in 10x10x9x9x8x8x... ways], then
the evaluation can't include history of how the position was reached.

Trying to program up an "If I go there, he can go there, but
then I could play that, ..." approach would enable your ideas, but
the resulting program would not be able to do a full search to an
interesting depth.

Then, in a losing position, the
program could show a strong bias towards lines that the human can
only win by playing a counter-intuitive move.
I know that computers are fantastically strong in relation to humans.
However, I'm not sure how good they are at playing losing positions.

[...]

I think one of the snags is that most [all?] current programs
have little idea of what it is that makes a position difficult to win.
Eg, if the program is two pawns down in a complicated position, it
may [will?] swap down into an obviously lost K&P ending one pawn down
if it can win a pawn by the swaps. Of course, if it sees far enough
ahead, then "one pawn down" translates into "and my opponent can push
that pawn through to promotion" and becomes a position to avoid.

--


Excellent brief snapshot of some of the issues, Andy! Many thanks -- I'm really glad I raised the topic, here.

Endgames with few remaining pieces (perhaps 6 is the threshold for an ordinary PC but I'm just guessing wildly) can be played perfectly by using a database, without any need for minimax and related techniques. The program simply stores every reachable position where White has checkmated Black, and every position where Black has checkmated White. Then it stores every position where White can move into a position where White has checkmated Black and every position where Black can move into a position where Black has checkmated White, and goes on for as many ply as needed, bounded only by the number of reachable positions which is small.

Do you think human-unfriendliness is practical for these endgames, or are there (as previously) issues of great difficulty that not I'm taking into account?

If a computer is losing one of these endgames, it can (and probably does) maximise the number of ply the human needs. I think that, for these few-piece database endgames, the computer might be able to improve by taking into account whether human moves are counter-intuitive (for example, greatly reducing the mobility of the human's own pieces), and show a bias towards moves which force the human to play counter-intuitively.

I understand how computers think in terms of positions rather than moves, but for few-piece endgames, memory and time should be amply sufficient for considering moves too.

Thanks again,

Paul
  #8   Report Post  
Old June 17th 16, 09:38 AM posted to rec.games.chess.misc
external usenet poster
 
First recorded activity by ChessBanter: Mar 2009
Posts: 3,536
Default Human-unfriendly chess software

On Friday, June 17, 2016 at 3:15:41 AM UTC+3, Andy Walker wrote:

As I wrote, but you snipped, I was considering "a grossly-oversimplified model", in particular in which moves don't interfere
with each. This is not generally the case, though it is more often
the case in computer analysis than in actual games. To compensate
for this, in cases where [eg] it takes a queen two or three moves
to get from one square to another, there are often several routes
the queen could take; and if a move is wasted, there are usually
many ways in which it could be wasted to reach the same position.


I see. You are talking about transpositions for purposes of analysis of legal moves, not the moves that appear in the move list as the best moves lines (the lines that move to the top of the tree when you do the alpha-beta and pruning). I suppose you are correct, as your 'only one way to reach one ply' example shows. I stand corrected, a rarity.

It's like an opening: in the beginning, many transpositions, but
after 30 moves, it's very unlikely you'll get a transposition.


You're confusing the number of ways in which a position is
likely to be reached in a real game [of interest to humans but not
to computer analysis] with the number of routes in a given number
of moves from one given position to another, which is what matters
for the transposition table, and which is typically very large in
cases of interest.


I don't notice that big a drop in Elo when I limit the transposition tables to a small memory limit, though I'm not paying too much attention. If you limit them to a very small number like 64 MB then it does seem to impact performance, but I think it's an asymptotic curve--you know what that is, yes?

RL
  #9   Report Post  
Old June 17th 16, 04:17 PM posted to rec.games.chess.misc
external usenet poster
 
First recorded activity by ChessBanter: Feb 2009
Posts: 57
Default Human-unfriendly chess software

On 17/06/16 09:31, Paul wrote:
Endgames with few remaining pieces (perhaps 6 is the threshold for an
ordinary PC but I'm just guessing wildly) can be played perfectly by
using a database, without any need for minimax and related
techniques.


The 5-piece tablebases are around 7GB, 6-piece around 1TB,
7-piece around 140TB. ATM, therefore, 5 is "easy", 6 is just about
within the space range of an ordinary PC [if you don't mind devoting
a significant part of your discs to it] but is quite difficult to
download [though coming within range], and 7 is still a bit much for
a normal domestic PC, to download, to store or to compute.

Do you think human-unfriendliness is practical for these endgames, or
are there (as previously) issues of great difficulty that not I'm
taking into account?


I don't think there is anything insuperable. But there are
issues. The tablebases are of most use not when you're in a table-
base position, but when it's a few ply away and the computer has to
decide whether to play into it. The problem then is that the computer
will do almost anything to stay out of a lost tablebase position; it
would rather lose a queen to keep 7+ men on the board than swap off
into a really tricky [but technically lost] Q&P ending. This is
probably curable, but see the last comment below.

If a computer is losing one of these endgames, it can (and probably
does) maximise the number of ply the human needs. I think that, for
these few-piece database endgames, the computer might be able to
improve by taking into account whether human moves are
counter-intuitive (for example, greatly reducing the mobility of the
human's own pieces), and show a bias towards moves which force the
human to play counter-intuitively.


Yes. To some extent, the computer would need to know the
human's background; for some people, the ending KBNvK is drawn,
while for others it's a simple win! You can account to some extent
for human frailty by counting the number of "only move" positions
in the winning line, but that's still fraught as many "only move"s
are obvious and others really subtle.

There's probably a research paper or three in these areas!

--
Andy Walker,
Nottingham.
Reply
Thread Tools
Display Modes

Posting Rules

Smilies are On
[IMG] code is Off
HTML code is Off
Trackbacks are On
Pingbacks are On
Refbacks are On


Similar Threads
Thread Thread Starter Forum Replies Last Post
What are some good beginner books? John Salerno rec.games.chess.misc (Chess General) 17 August 24th 08 11:26 PM
rec.games.chess.misc FAQ [2/4] [email protected] rec.games.chess.misc (Chess General) 0 April 23rd 06 05:21 AM
rec.games.chess.misc FAQ [2/4] [email protected] rec.games.chess.misc (Chess General) 0 April 7th 06 05:30 AM
rec.games.chess.misc FAQ [2/4] [email protected] rec.games.chess.misc (Chess General) 0 February 4th 06 05:25 AM


All times are GMT +1. The time now is 04:54 PM.

Powered by vBulletin® Copyright ©2000 - 2019, Jelsoft Enterprises Ltd.
Copyright 2004-2019 ChessBanter.
The comments are property of their posters.
 

About Us

"It's about Chess"

 

Copyright © 2017