When will the chess programmers write an engine that plans ?

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

Moderators: hgm, Rebel, chrisw

jefk
Posts: 626
Joined: Sun Jul 25, 2010 10:07 pm
Location: the Netherlands
Full name: Jef Kaan

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

Post by jefk »

With 'Father' and 'Tsvetkov' being silent, its
time again for some human input imho:

Talking about computer plans and 'strategy' :
What engines do *not* have is the subtle strategy of a human;
knowing about the possible fallacies and weaknesses
of his opponent....

Watch here (in horror) the subtle strategy below ; where White (yours truly)
applies a trick against another human, almost equal in (tactical)
strength, but not aware of opening traps, two examples"
(with Black playing the sometimes thought to be most solid,
strategic defence (Caro-Kann) possible against White...
(lol)
[Site "Chess.com"]
[Date "2020.09.30"]
[White "jefk"]
[Black "Dxxxhxxd"]
[Result "1-0"]
[TimeControl "900+10"]
[EndTime "7:04:12 PDT"]
[Termination "jefk won by checkmate"]
1. e4 1... c6 2. Nc3 {[%clk 0:15:10]}
2... d5 {[%clk 0:15:10]} 3. Nf3 {[%clk 0:15:20]} 3... dxe4 {[%clk 0:15:20]} 4.
Nxe4 {[%clk 0:15:30]} 4... Bg4 {[%clk 0:15:30]} 5. h3 {[%clk 0:15:40]} 5... Bh5
{[%clk 0:15:40]} 6. Qe2 {[%clk 0:15:50]} 6... Nd7 {[%clk 0:15:50]} 7. Nd6#
{[%clk 0:16:00]} 1-0
:)

Example 2:
[Site "Chess.com"]
[Date "2020.08.24"]
[White "jefk"]
[Black "jaxxrxk"]
[Result "1-0"]
[ECO "B10"]
[WhiteElo]
[BlackElo "1466"]
[TimeControl "900+10"]
[Termination "jefk won by checkmate"]
1. e4 1... c6 2. Nc3 {[%clk 0:15:10]} 2... d5 {[%clk 0:15:10]}
3. Nf3 {[%clk 0:15:20]} 3... dxe4 {[%clk 0:15:20]} 4.
Nxe4 {[%clk 0:15:30]} 4... Bf5 {[%clk 0:15:30]} 5. Qe2 {[%clk 0:15:40]} 5... Nd7
{[%clk 0:15:40]} 6. Nd6# {[%clk 0:15:50]} 1-0

lol
:)
yes i know, i set a trap, with a suboptimal move (Qe2)
hoping that the opponent will make (fatal) mistake;
in other words 'speculative' play. Neither Lco nor NNue
are doing that; you only can do that with 'high contempt'
(and chess knowledge).

Now you say, you cannot do this against computer programs ???
Yes we can ! Here's an example, accomplished against an engine
on Fics (no NNUE, simple alfa beta), rapid rating 2000+ plays
like a toddler:

[Event "Computer chess game"]
[Site "freechess.org"]
[Date "2020.10.03"]
[White "jefk"]
[Black "SpeckEngine"]
[Result "1-0"]
[ECO "C28"]
[Opening "Vienna"]
[Time "19:03:16"]
[Variation "3.Bc4 Nc6 4.f4"]
[TimeControl "840+1"]
[PlyCount "49"]
[WhiteType "human"]
[BlackType "program"]
[Comment "SpeckEngine checkmated 1-0"]
1. e4 e5 2. Nc3 Nf6 3. Bc4 Nc6 4. f4 exf4 5. Nf3 Na5 6. Be2 Nc6 7. d4 Bb4
8. e5 Nd5 9. Bd2 Ne3 10. Bxe3 fxe3 11. d5 Ne7 12. Qd3 O-O 13. h4 Bxc3+ 14.
bxc3 d6 15. c4 Nf5 16. Ng5 h6 17. g4 hxg5 18. gxf5 gxh4 19. O-O-O dxe5 20.
Rdf1 Bd7 21. f6 h3 22. Rhg1 g6 23. Rxg6+ Kh7 24. Rg7+ Kh8 25. Qh7# 1-0

Well that was simple, wasn't it (don't tell me i played lousy; i know
but that was not the point, i'ts the last point that counts...)
:)
As for 'simple' ? well actually not, i just played with intuition
(see PS below). Bottom line, for humans know the playing style of your
opponent, try to exploit weaknesses, and know that chess is a game
(where the opponent can make 'strategic' mistakes); you don't have to
play perfectly dull positional chess (often ending in a draw with strong
opponents) to make a fun game out of it; possibly true at higher levels
as well (Lco and Nnue are still machines, they don't have creativity).

How to build in 'creativity' (exploiting possible mistakes of
the opponent (like eg Carlsen is doing in the endgame even
at GM level) for comps ? It requires another sort of contempt
and some knowledge about the sort of opponent (rating (once
crafty had a feature to know things about the opponent on ICC,
eg human/comp, not sure about rating), playing style, etc).
Will it add rating points ? mwah only for the best (lco of SF,
with such an extra feature); will it solve chess ?
Definitely not (it's a draw!); will it add some more fun
to the game(s)? Yes it will, imho.

best regards

jef


PS simple? well only because looked beyond the horizon (Nnue and lco ofcourse
would be more difficult), and i have some recent experience with the Vienna:
see eg this one :

[Site "Chess.com"]
[Date "2020.09.30"]
[White "jefk"]
[Black "Vexxkoxxj"]
[Result "1-0"]
[ECO "C25"]
[TimeControl "900+10"]
[EndTime "6:35:28 PDT"]
[Termination "jefk won by checkmate"]
1. e4 1... e5 2. Nc3 {[%clk 0:15:10]} 2... Bc5 {[%clk 0:15:10]} 3. Bc4 {[%clk 0:15:20]} 3... h6 {[%clk 0:15:20]} 4. f4 {[%clk 0:15:30]} 4... Nc6 {[%clk 0:15:30]} 5. Nf3 {[%clk 0:15:40]} 5... exf4
{[%clk 0:15:40]} 6. d4 {[%clk 0:15:50]} 6... Bb6 {[%clk 0:15:50]} 7. Bxf4 {[%clk
0:16:00]} 7... d6 {[%clk 0:16:00]} 8. Qd2 {[%clk 0:16:10]} 8... Nf6 {[%clk
0:16:10]} 9. h3 {[%clk 0:16:20]} 9... O-O {[%clk 0:16:20]} 10. O-O-O {[%clk
0:16:30]} 10... Ba5 {[%clk 0:16:30]} 11. Bxh6 {[%clk 0:16:40]} 11... gxh6 {[%clk
0:16:40]} 12. Qxh6 {[%clk 0:16:50]} 12... Nh7 {[%clk 0:16:50]} 13. Nd5 {[%clk
0:17:00]} 13... Be6 {[%clk 0:17:00]} 14. Bd3 {[%clk 0:17:10]} 14... Bxd5 {[%clk
0:17:10]} 15. exd5 {[%clk 0:17:20]} 15... Nb4 {[%clk 0:17:20]} 16. Qxh7# {[%clk
0:17:30]} 1-0

PS2 chessdb.cn now only displaying xianchi apparently ?
ah well. it getting a bit boring nowadays in corresp chess..
jefk
Posts: 626
Joined: Sun Jul 25, 2010 10:07 pm
Location: the Netherlands
Full name: Jef Kaan

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

Post by jefk »

analyzed my (recent) Vienna game against 'speckengine' with komodo ,
and indeed, it was not perfect; Komodo gives a ? for my move 15 c4
But Speckengine plays Pf5 instead of dxe5 !
And later makes the mistake 20...Ld7??
(20...Qf6 was better, but even then White
ie yours truly was already strategically winning)
Did i see all that ?
nope not in such detail. LIke i said, intuition.
Now i don't have the chess abilities of Lco and SFNnue i admit
but they don't have 'intuitive' capabilities (knowing about
your opponent, tricks, traps, opening strategies, etc) and
there's still imho some room for improvement.
Top of 4000 lo ? mwah dunno, 3800 maybe. we'll see.
In general there's still are hybrid (human/comp) improvement
possible in the game of chess, once known to be the 'fruitfly' of AI.
With nowadays the hype of coming socalled AGI (and (much) later ASI
it's still interesting to see how this field is doing in this relatively
simple (but in fact not so simple) game of 'chess''...
;-)
jef
PS my whole change in playing style (with the Vienna examples
was a result of the Lco revolution; i already have a solid 1.d4
repertoire, but against weaker players (or weak comps like
this speckengine), i sometimes try 1.e4 (and then deviating
from known theory, or at least comp opening books)
if i'm in a strong/confident mood... yeah i know Capablanca
advised to kiddies to get strong in the endgame but i never
bothered coz i like it in amateur chess when i win within 20 moves
or so which deserves me another beer at least...
:)