Centaur vs Unassisted Engine

Discussion of computer chess matches and engine tournaments.

Moderators: hgm, Dann Corbit, Harvey Williamson

Forum rules
This textbox is used to restore diagrams posted with the [d] tag before the upgrade.
jefk
Posts: 503
Joined: Sun Jul 25, 2010 8:07 pm
Location: the Netherlands
Contact:

Re: Centaur vs Unassisted Engine

Post by jefk » Sun Feb 21, 2021 6:21 pm

ok, now also seen comments by mrbbgian

First adding to my comment to Monte Carlo:

when looking deeper, apparently the latest Black move
(...Bxf3) after all wasn't so bad because after 15...Rb8,
Black gets an open line for his rook, and thus enough compensation
(in fact we get some pawn grabbing but then indeed difficulty
for White to get into a winning endgame).

The purpose of this testgame for me was to get an idea about the
sort of intrinsic advantage for White (bearing in mind the idea
to diminish draws in correspondence chess with comp assistance).
Now in this game, i will get a better pawn structure, that's for sure,
the c7 pawn probably becomes isolated; but then in comp vs comp
play, this probably won't be enough for a win. So be it...

So, although i will continue the game for quite some moves (endgames
sometimes aren't so dull as they are supposed to be), one
little comment (mainly to mrbb), it's not about me having
failed to 'steer' the game into desirable anti-comp positions,
because like i suggested (with the nnue nets), this is almost
impossible. My simple strategy was to play good moves (in fact i tried to play the best positional moves possible, helped by the comps ofcourse).
NB also with 1.e4 i wouldn't have been able to increase the initial
advantage into significant further advantage (in the Berlin with Re1 as in LK's book after a while he suggest d5 to increase space advantage, but then this also won't be enough to win the game with perfect play, nor against a fast comp with the Nnue stuff and the alfa/beta combined).

But then after all this isn't surprising to me because chess after all
is a draw with perfect play. concluding, for correspondence chess,
depth/ computer speed indeed is useful, but after a while
we reach a draw limit and you won't have much chances
into pushing your opponent into a worse position (which in my
situation i try (or tried) to do with positional considerations,
having several engines at my disposal with different evals
(but then also being less efficient in the alfa/beta ergo depth
area). If someone else would like to try it (winning with White
under the same conditions, be my guest (*), good luck :)

Anyway, after a few moves, i might offer a draw, without
being ashamed because i simply was some sort of Idea operator
with some chess knowledge sure, but not claiming i could win
a game which basically is fundamentally a draw anyway.

PS and mrbbgian tell me where i could have played better moves,
and then try it against your engine, it's probably (almost?) impossible.
PS2 as for going into a 'anti-computer' strategy (and playing strange
opening moves), in general this is just the way i won quite some
correspondence games, against people who thought they could
play any known (yet rather obscure) opening, after which i was
often able to refute it, even before the Nnue age, and yes,
even with my apparently rather 'slow' computer (not that it matters,
the current game has shown that a faster comp will be able
to find a draw a bit faster, that's all. As for winning chances for
Black now in current positions, this ofcourse also would be balderdash. :idea:
jefk

jefk
Posts: 503
Joined: Sun Jul 25, 2010 8:07 pm
Location: the Netherlands
Contact:

Re: Centaur vs Unassisted Engine

Post by jefk » Sun Feb 21, 2021 7:12 pm

anyway, lots of talk, and no moves,
so here's my move again:

14.Bxf3

now probably 14...Nc6 (excusez moi the haphazard pgn notation
but then in the pgn standard the input format doesn't have to
be as strict as the output format, believe it or not)
and anyway, i'll wait for the move.

PS as for the 'centaur' as topic in this thread, i'm not claiming i'm a
very special centaur, but then again, seeing the resistance Black is
posing, i'm still pretty confident that there wouldn't be many people
who could do it much better i.e. get a winning result. (well if you
think of trying 1.e4, antimarshall or so and then trying to out- calculating montechristo's (oops monte c's comp even that would not always
work and take quite some electricity, that's what i'm thinking)
No that's not a proof that chess is solved but then i like
to see the burden of proof reversed, i.e. let some (other)
people show they can win (always) with White (or Black @$#%')c1f^_!#?)
:)
jefk

MonteCarlo
Posts: 146
Joined: Sun Dec 25, 2016 3:59 pm

Re: Centaur vs Unassisted Engine

Post by MonteCarlo » Sun Feb 21, 2021 8:40 pm

14...Nc6 indeed...


jefk
Posts: 503
Joined: Sun Jul 25, 2010 8:07 pm
Location: the Netherlands
Contact:

Re: Centaur vs Unassisted Engine

Post by jefk » Mon Feb 22, 2021 12:07 pm

15.Bxc6

rewording my comments, i don't think i can win with current draw
rules, but again (using the secondary variation board in the Fritz
GUI) i'd like to get an idea of the (small) 'advantage' (in material (*))
White could get, even in this game (whereby in retrospect 11.Nc3 may
have been better than 'my' 11.Qc1, although latter up till know
used to be more common).

(*) if not a pawn eg four to three on king's side, then also queen vs
rook+knight, and both three pawns now seems possible (count the queen as nine points, and rook+knight as eight, then White would have won with new rules,
but then i claim this was simply due to the inherent first move advantage for White; ergo, material count in the endgame would be an option (in new correspondence rules, and be more effective than messing with the stalemate rule, but then as default one point might be subtracted for White as default...
my2cnts
jefk

MonteCarlo
Posts: 146
Joined: Sun Dec 25, 2016 3:59 pm

Re: Centaur vs Unassisted Engine

Post by MonteCarlo » Mon Feb 22, 2021 1:34 pm

15...bxc6



An idea like that (material as tiebreaker in case of draw) might work.

My only quibble is with using traditional piece values, which are a bit arbitrary.

I suspect it would be more interesting simply to count the pieces, regardless of "value". White could even start with -0.5, eliminating ties altogether, if desired.

Both versions of the suggestion would have the effect of discouraging/encouraging certain imbalances, and that effect would have to be carefully considered.

For general odds-giving, this scheme could even be extended to include non-drawn games, requiring the stronger player to be up a certain amount of material for a win to count.

This could have some weird implications though, like the weaker side ignoring early mate threats when the stronger side's material advantage isn't large enough.

It might be a fun kind of weird, though.

Cheers!

jefk
Posts: 503
Joined: Sun Jul 25, 2010 8:07 pm
Location: the Netherlands
Contact:

Re: Centaur vs Unassisted Engine

Post by jefk » Mon Feb 22, 2021 1:45 pm

ok, now 16.Rc1, and then I offer a draw.

Expecting 16..Rb8, after 17.Rxc6 Rxb2, after 18.Nc3 Rc2! etc.
Below the (sort of) endgame i expect in our game (especially
after a later 19.Rd1 (although 19.Rb1 also would be possible) :

https://ibb.co/xJcLVTj


Still tricky for humans sometimes, but not for engines with egtb.

Yes just counting pieces would be possible, but might change
the nature of the game a bit too much (unless engines would
start to adapt to it, possibly as option).
NB in possible new correspondence rules with just counting conventional
material values in the endgame, default one pawn for White may be a
bit too much, maybe indeed one half of a pawn is more reasonable.
Values for pieces for such adjudication purposes then also still
have to be worked out, e.g. bishop 3.5 pawn value.
jefk

MonteCarlo
Posts: 146
Joined: Sun Dec 25, 2016 3:59 pm

Re: Centaur vs Unassisted Engine

Post by MonteCarlo » Mon Feb 22, 2021 2:07 pm

Draw provisionally accepted.

Having said that, I will reveal I am not using tablebases, so if you want to continue to see if that can be exploited at this long TC, I'm up for it.

Indeed, both schemes would incentivize some types of imbalances, and the mere counting scheme's incentives would be different than the summing piece value scheme's incentives.

I'm not sure one actually affects the game substantially more than the other.

In the most common drawn material imbalance (one pawn more/less), they are equivalent.

They would then differ in draws with either same piece count/different value (value would give KR the win in drawn KRkb, for example), same piece value/different count (drawn theoretical endings like this are actually pretty rare with traditional values, but a middlegame repetition with rook vs minor plus 2 pawns would fit the bill, where counting would give the win to the minor piece side), or different piece value/different piece count (drawn KQkrp, where the schemes would each award a win, but to different sides, for example).

It's not obvious to me which changes gameplay more, and counting at least avoids that annoying step of arbitrarily assigning adjudication values to pieces :D

Cheers!

lkaufman
Posts: 4529
Joined: Sun Jan 10, 2010 5:15 am
Location: Maryland USA
Contact:

Re: Centaur vs Unassisted Engine

Post by lkaufman » Mon Feb 22, 2021 5:34 pm

jefk wrote:
Mon Feb 22, 2021 1:45 pm
ok, now 16.Rc1, and then I offer a draw.

Expecting 16..Rb8, after 17.Rxc6 Rxb2, after 18.Nc3 Rc2! etc.
Below the (sort of) endgame i expect in our game (especially
after a later 19.Rd1 (although 19.Rb1 also would be possible) :

https://ibb.co/xJcLVTj


Still tricky for humans sometimes, but not for engines with egtb.

Yes just counting pieces would be possible, but might change
the nature of the game a bit too much (unless engines would
start to adapt to it, possibly as option).
NB in possible new correspondence rules with just counting conventional
material values in the endgame, default one pawn for White may be a
bit too much, maybe indeed one half of a pawn is more reasonable.
Values for pieces for such adjudication purposes then also still
have to be worked out, e.g. bishop 3.5 pawn value.
For tiebreaking by material count, the question of whether bishop = knight or bishop > knight is not a minor detail, but is absolutely the critical question. If we assume that Black wins with equal count but White wins with superior count, then normal 1-3-3-5-9 count (or anything similar with bishop = knight) will mean that Black will be a huge favorite, because it is very unlikely that White can actually win a pawn or the equivalent with perfect play. But if bishop > knight, then White might even become a favorite, as it is easy to imagine that White's advantage is at least enough to force Black to trade a bishop for a knight to avoid something worse. The details aren't very important, it could be 1-3-3.5-5-9.5 (queen shouldn't be less than R+B+P). But the nature of the game will change.
Komodo rules!

mbabigian
Posts: 189
Joined: Tue Oct 15, 2013 12:34 am
Location: US
Full name: Mike Babigian

Re: Centaur vs Unassisted Engine

Post by mbabigian » Mon Feb 22, 2021 5:51 pm

Jefk, to semi-answer your questions, first I must say I'm mostly in the Uri Blass camp. So if you want me to provide moves to win, that's a tall order as I would not suggest that even with my resources, that I can force a significant numbers of wins.

1) Chess is likely a draw with best play.
2) Unassisted top engines playing on good hardware and with tablebases will score very high draw rates against the CC world champions if they were to play hundreds of games to get a good statistical sample and were not allowed to repeat the same "won" game repeatedly (either through a rule prohibiting it or a variety book).
3) There are opening lines you can play hoping to push the engine into the small game space where engines under perform. This is what top CC players do to score their few wins. Most of the time they fail, but occasionally their opponents get led down the merry path. These few wins make the difference over the long haul between the champs and the chumps. It takes enormous resources to find such lines now days and once you find one, you tell no one about it until you get lucky and someone plays into it. The chess game space is so large, there will always be such rare opportunities, but you are exploiting holes in engine understanding, not finding a win for white (see point 1).

Considering point 2 above, you had a low probability of steering MonteCarlo into a winning position even if you had a line prepared ala point 3 above. If Montecarlo's PC just happened not to play into your prepared line simply because of multi-processor nondeterminism, you end up with a draw. I think this is why folks like Uri believe that computers have sucked the fun out of CC chess. I think the CC players of today like the challenge of finding the rare win and fiddling with their software/hardware everyday. This is a different type person than the CC players of the 1980s and before.

Finally, you must realize that the hardware to compete in CC today (at the top level), is likely very extreme. You have mentioned you use Aquarium so I'm sure you spin up workers (engines) to build trees of moves you want to investigate and understand that process. So with that in mind, my main computer has enough PCIE 4.0 ultra fast SSD storage to hold the WDL 7 piece Syzygy files, has a huge amount of RAM to do long Leela searches, and as you are aware, is orders of magnitude faster than your PC or MonteCarlo's, but that is just one computer. I can turn up numerous computers to add cores (workers) to do tree building. By poking a few buttons, I can have a 100 cores or more expanding an opening variation I'm interested in and if I so choose, turn up another 100 cores. If it is easy for me to do, I suspect there are many other folks that do the same.

So to be clear, I was never questioning your skill or methods, only that at CC time controls scoring something other than a draw is relatively rare. If you tell me you have a roulette wheel that lands on 1 about 2% of the time, 2 about 96% of the time, and 3 about 2% of the time, I'm going to only bet on 2 every time! ;)

Mike
“Censorship is telling a man he can't have a steak just because a baby can't chew it.” ― Mark Twain

jefk
Posts: 503
Joined: Sun Jul 25, 2010 8:07 pm
Location: the Netherlands
Contact:

Re: Centaur vs Unassisted Engine

Post by jefk » Mon Feb 22, 2021 6:49 pm

MC

ok, since you don't accept my draw (yet), let's continue
for -at least- a few more moves, so it's your move now again
(expecting 16...Rb8). As it stands now, i still expect an interesting
endgame after 19.Rb1, and then i'm curious what your 20th
move will be. The egtb's (or your lack of it) probably don't
matter yet at this stage, but maybe the 'network' (nnue file)
you're using isn't as good in the endgame as in the opening/
middlegame, thus leaving me with some slight chances
(at least again, for a 'better' endgame position).

mrbbg

roughly agree, those lines where still sometimes a game
can be won are most likely in 1.e4 openings where Black
is trying something which is less drawish (avoiding Berlin or
RL Marshall), but then ends up being out-calculated by
a faster comp after one sub-optimal move; my 2 cnts.

mr LK
ok, so 1,3,3,5,9 would be the way to go in material count,
but then still White would have the first move advantage
which means in a proper tourn two players A/B should
play each other both with White and Black...
While i was agreeing with MC that a penalty of 0.5 for
White would be reasonable, when there are no halves in
above counting system, such a penalty would be meaningless
(maybe the queen then can be counted 8.5 to change that, lol)
jefk

Post Reply