If you have disabled Java for your browser, use the Site Map (linked in the header and footer).

Chessville
logo by
ChessPrints
 

Advertise
with
Chessville!!

Advertise to
thousands
of chess
fans for
as little
as $25.

Single insert:
$35
x4 insert:
@ $25 each.

From the
Chessville
Chess Store

From the
Chessville
Chess Store

In other words, Vasik was not aiming at perfection per se, but the strongest practical strength he could muster. Vasik was a very intelligent, sincere, charming, straightforward engaging guy.  I asked him why I was stuck in the 2200-2300 range despite reading so many Chess books.  Vasik - You play according to your level, like everyone else.

Although I am over-generalizing, his words made sense in the games Dzindi won.

3 - Amazing skill playing computers.  Roman would play some unrated games with the computer until he got a handle on its weaknesses, then go to town.  Roman had the highest blitz rating on ICC for a while, then Shirov, Nakamura and Grishchuk passed him.  To get to 3400 on ICC you must be able to beat 3000-rated computers.  Even when Kasparov and Kramnik were playing million dollar computer matches, I felt they were sending in the wrong guy.  Send in Dzindi!

4 - Grandmasters feel hounded by 1500s who want to improve - people like to speak their own language - Grandmasters tended to know who each other were and their handles.  Roman would have at least 60 people following him for any ICC game he played.  He changed handles like I change socks.

5 - I was impressed Roman could just call up Karpov and talk to him for 20 minutes if he felt like it . "Tolya?"

6 - Roman was a top-10 GM in 1980.

7 - A world traveler that could speak many languages.

8 - If someone beat Roman he roared like a Lion - "I win 10 games in a row now!"

9 - I helped Dzindi make five Chess videos, even lending him my Chess books as a backdrop and suggesting some games.

10 - Mostly Roman was the most entertaining man I ever met and his Chess seminars went very well in Denver.  I drove him all around town and he would repay me with great stories of the royal Chess life I never got to live.

11 - Dzindi's opinion of others always fascinated me.  He described me on the phone as an "ordinary master".  Even national Chess Champions were "not there" when it came to endings.  GM Karjakin "gets stronger every minute!"

12 - Roman helped Kamsky on his first rise to power, beating Anand, Salov, Kramnik and Nigel Short.  Since Karpov and Roman were such good friends, Kamsky did not use Roman against Karpov.  Kamsky lost to Karpov and withdrew from Chess for 10 years.

13 - Roman was with 38 year old GM Leonid Stein when he died practically in his arms.  Roman called for help but it was too late.

14 - Roman is one of the few that can boast that he beat Karpov in a simul - Roman gave the simul.

15 - One time Steve Sabean showed me a postal game he just drew.  I analyzed the final position for half an hour and stated Steve might still have some winning chances.  Roman, from across the room the next day, after analyzing for under five seconds, immediately stated he saw at least two ways to win.

16 - 2005 Utah Champion Josh Smith and I played a game where we could move the pieces and Roman was blindfolded.  Not only did we lose, we had to ask him where everything was.

When I heard Roman was playing an $11,000 match with Rybka with an extra $100 per point scored, I was very excited.  Finally the world's best anti-computer player versus the world's best computer despite the heavily funded Kasparov, Kramink Fritz matches.

Since Rybka has beaten GMs Ehlvest and Joel Benjamin and since no one knew if Dzindi's age (64) would play a factor, the odds were changed to - Roman gets White every game, the computer offers up a different pawn each game.

Legend has it Morphy offered the world pawn and move odds but no one stepped up.

Topic Rybka vs Dzindzichashvili - some comments By Vasik Rajlich Date 2008-03-08 09:25

The match between GM Roman Dzindzichashvili and Rybka has finished with a somewhat surprising (at least to me) score of 4-4.  Thanks to Larry for organizing everything and kudos to GM Dzindzichashvili for showing that this is still possible.

Without any deep analysis, here are my quick impressions:

1) The Rybka anti-human stuff wasn't especially effective.  Dzindzichashvili was regularly able to close the positions (game 7), trade at will (game 2 and others), especially the minor pieces, and generally do those things which we all agree the human wants to do.  Rybka's anti-human & contempt settings do penalize these things, but for complex reasons the penalties are not huge and in the right circumstances the human can have his way.  We saw already these tendencies in both matches against Benjamin, but Dzindzichashvili was able to exploit this even more frequently and more effectively.

2) Dzindzichashvili played with a lot of confidence.  Going back to the days of Kasparov vs Deep Blue, psychology seems to play quite a role in these human vs machine matches, with the human often choosing to play cautiously.  Dzindzichashvili apparently wanted no part of that, turning down a draw by repetition in an equal position in game five and regularly sacrificing material.  To me the most eyebrow-raising example was the insanely bold 21.Nf6+ from game three.

3) This match was much wilder than previous matches.  When Benjamin played with roughly the same conditions, he had real losing chances in just two of the eight games.  By my possibly superficial count, in this match, Dzindzichashvili had real losing chances in five of the eight games.  He compensated for this with two very impressive wins, tripling Rybka's loss total in Larry's five matches against humans.

4) Once a position collapses, minimax search does not defend well.  This is well-known, although it rarely has any real significance.  We saw it again in game seven and especially in game two.  Of course, both games would most likely have been lost anyway without the rather absurd attempts at last resistance by Rybka at the end.  Other instances of this from Rybka's play are the loss to Ehlvest in the first human match and our loss to Cluster Toga in the last Paderborn.

It seems from Larry's emails that Dzindzichashvili had a lot of fun trying to crack this challenge, playing all sorts unofficial games under different conditions and making various suggestions, etc.

Vas

ps.  If somebody could volunteer a little write-up of the games for our web site, that would be great.

By Vasik Rajlich Date 2008-03-08 09:25

Since I was rooting for Roman let us examine his two wins.

Game 2
Dzindzichashvili-Rybka

[Event "Computer-Human Pawn and Move Match"] [Site "IM Larry Kaufman's house"] [Date "2008.3.4"] [Round "2"] [White "Dzindzichashvili"] [Black "Rybka"] [Result "1-0"] [Eco "minus-b7-pawn, odds match"] [Annotator "Brian Wall"] [Source "Gatorade"]

1.Nc3 Nc6 2.Ne4 Nb8 3.Nc5 Nc6 4.Nxb7 Nb8 5.Nc5 Nc6 6.Na4 Nb8 7.Nc3 Nc6 8.Nb1 Nb8

All this was played just to get to the "odds" position - no b7 pawn.  A 3000 rating is impressive but giving the double Soviet/American Champion pawn and move is pretty cheeky.  Roman was famous on ICC for killing computers better than anyone.

--------------------------------------------------------------------------
--------------------------------------------------------------------------

Rybka website

By Jim Walker Date 2008-03-04 13:13

Do you have to remove the pawns in strict order....a7,b7,c7 etc.. ?  Seems like knowing which pawn is missing next will help in preparation.  Just a thought.

---------------------------------------------------------------------------

By lkaufman Date 2008-03-04 15:30  We could specify any order, but it would not be fair for me to know the order and for the GM not to know it.

-------------------------------------------------------------------------
-------------------------------------------------------------------------

9.Nf3 Nf6 10.g3 Nc6 11.Bg2

It usually takes a long time in the English Opening to combine Rb1 and Bg2 pressure on the b7-pawn.  In this case that work has already been completed.  Rybka uses the open file.

11 ... Rb8 12.d4 e6 13.O-O d5

Grabbing a hunk of center but now c7 is a target.

14.b3 Be7 15.Ba3

Trading a 2540 bishop for a 3000 bishop plus getting closer to a winning ending.

15 ... Bxa3 16.Nxa3 Qd6 17.Nb1 O-O

Grandmaster Roman Dzindzichashvili is the best Chess storyteller in the world.

18.Nc3 h6 19.Re1

Roman is a pawn up and might pawn break with e4 or c4

19 ... Rd8

To discourage both pawn breaks if she was human.

20.e3

Looks like it is going to be c4.

20 ... Ba6

To slow down c4, if she was human.

21.Na4 Nd7 22.Qd2 once Nf6 once 23.Qc1 once Nd7 once 24.Qd2 twice Nf6 twice 25.Qc1 twice Nd7 twice 26.Rd1

Declining a three fold repetition and preparing c4, which can be played immediately.

26 c4 dc 27 Nd2 or after 26 Nb2 e5 27 c4 or 26 Nd2 e5 27 c4  Not sure why IM Vasik Rajlich considered this pawn down position equal for Rybka.

26. ... Ne7 27.Nd2 Be2 28.Re1 Bb5 29.c4

Interesting - rather than waste time moving the knight the Grandmaster allows doubled rook-pawns.  Somehow he makes it work.  Roman has other advantages - nice g2-bishop, weak c7-pawn.

One advantage GM Dzindzichashvili doesn't have is his name - I saw him almost tear his hair out when he tried to install cable and had to spell his name out 10 times to the same company in one afternoon.  Roman is only patient on the chessboard.

29 ... Bxa4 30.bxa4 Qb4 31.cxd5 Nxd5 32.Bxd5

Keeping it simple reduces losing chances - Roman wants a position he can understand.

32 ... exd5 33.Nb3

Uncovering an attack on c7 buys time to save the silly doubled rook pawn.

33 ... Rb7 34.a5 c5

Eliminating the key weakness.

35.Nxc5

Trades reduce losing chances and still keeps some edge.

35 ... Nxc5 36.Qxc5 Qxc5 37.dxc5 Rc7

Roman cannot save the c-pawn but the position is comprehensible and the GM can counterattack other pawns.

38.Rec1 Rdc8 39.Rab1

No wild computer tactics to confuse the human - just simple stuff.  C-pawn falling - counterattack begins.

39 ... Rxc5 40.Rxc5 Rxc5 41.Rb8+ Kh7 42.Rb7

Dzindi is still up a pawn.

42 ... Rxa5 43.Rxf7 Rxa2 44.Rd7

A key moment.  I don't see a good move for Rybka.  44.Rd7 Rd2 hoping for 45.R:a7 d4= fails to 44.Rd7 Rd2 45.Kf1 a5 46.Ke1.  I think 44.Rd7 Ra5 should fail to a Dzindi King march to e5 after suitable pawn preparation.  44.Rd7 d4 45.R:d4 does not seem like any improvement on the game.  Maybe Rybka is just lost here.

------------------------------------------------------------------------

Rybka website

By lkaufman Date 2008-03-04 01:41 Roman said that 43...Rxa2 was a serious mistake. Instead pitching the "d" pawn first by d4 offered much better drawing chances, he said.

----------------------------------------------------------------------
----------------------------------------------------------------------

44 ... a5 45.Rxd5 a4

The a-pawn is not enough counterplay to offset Roman's pawn massive Kingside pawn roller.

46.Ra5 g5 47.g4

To stop ... g4, fixing pawns.

47 ... a3 48.Kg2 Kg7 49.Ra6 Kf7

I don't see the win after 49 ... Kh7 50 Kg3 Kg7 51 f4 Ra1 52 f5 a2 53 Kg2 Re1 53 R:a2 R:e3

----------------------------------------------------------------------------

Rybka website:

By lkaufman Date 2008-03-04 01:39

The move 49...Kf7? is very suspicious, even if Rybka was already losing.  Rybka does not make this move with contempt turned off.  I will investigate.

By lkaufman Date 2008-03-04 06:02

Despite my earlier comment, this turned out not to be so.  Apparently the choice here just depends on the time taken, not much on the contempt setting.  The basic problem is that Rybka is not materialistic enough in the endgame, the opposite of what everyone used to say about computers!  But as usual, fixing this gives worse test results.

------------------------------------------------------------------------------

By lkaufman Date 2008-03-04 15:28

Yes, the ...Kf7 move is one such position to correct.  The release won't be quite as soon as we thought, not because of this match but just because certain changes are taking longer than expected.

-----------------------------------------------------------------------------

By lkaufman Date 2008-03-04 16:04

Every change helps in some positions and hurts in others.  We only keep changes that help on balance, but there will always be some situations where the newer, better program makes a worse move than the older one.  Anyway the timing of release is based on completing certain work and then testing; we are long past the point of wondering whether 3.0 will be better than 2.3.2a, that is not in doubt.

---------------------------------------------------------------------------
---------------------------------------------------------------------------

50.Rxh6

We finally reach a position I can win.

50 ... Rb2 51.Ra6 a2 52.Kg3 Re2 53.f3 Ke7

If 53 ... R:e3 54 R:a2-f2 and pawn to h4 creates two connected passed pawns, a basic win.

54.e4 Kd8

Going after Dzindi's rook

55.h4 gxh4+ 56.Kxh4 Kc8 57.Kg5 Kb7

Rybka's plan is to win the Roman's rook but of course Rybka's rook is no match for three connected passed pawns.

58.Ra3 Kb6 59.f4

59 Kf5 with a slower pawn march is also winning, not offering Rybka any pawns - the swimming buddy system.

59 ... Kc5 60.e5 Kb4 61.Ra8 Kb3 62.Kf5 Rc2

62 ... Re1 63 e6 a1(Q) 64 R:a1 R:a1 65 e7 is too fast for her anyway.

63.g5 Rc8 64.Ra6

No need to be fancy - 64 R:a2 works just fine.

64 ... Rf8+ 65.Kg4 Re8

Larry Kaufman and Roman Dzindzichashvili wanted to spell it out for their international audience.

66.g6 Re6 67.Rxa2 Rxg6+

I wonder if Roman felt any pity for the silicon box?

68.Kf5 Rb6 69.Rd2

Cutting Rybka's King out of the action.

69 ... Rb5 70.Kf6 Rb6+ 71.e6 Kc4 72.f5 Kc3 73.Rd8 Ra6 74.Kf7 Ra5 75.f6 1-0 Rybka resigns.

I was thrilled.

Game 7
Dzindzichashvili-Rybka

[Event "Human-Computer pawn and move match"] [Site "IM Larry Kaufman's house"] [Date "2008.3.7"] [Round "7"] [White "*Dzindzichashvili"] [Black "*Rybka"] [Result "1-0"] [Eco "minus g-pawn, odds match"] [Annotator "Brian Wall"] [Source "Ole Man River"]

1.Nh3 g6 2.Na3 g5 3.Nxg5 Nc6 4.Nh3 Nb4 5.Nb1 Na6 6.Ng1 Nb8

All this was played just to get to the "odds" position - no g7 pawn.

7.d4 Nf6 8.Nf3 d6 9.Bg5

All trades are good when you are winning - GM Roman Dzindizchashvili.  However 11.Bxf6 frees Rybka's Queenside and gives up the two bishops.

Larry Kaufman, a good friend to Roman and also Rajlich's evaluation function helper stated that part of the handicap was not creating a special opening book for odds play, i.e., moves that avoided trades and kept the game complex.  Larry even admitted he was not sure what the right strategy for no g-pawn is.

9 ... Nbd7 10.Nbd2 h6 11.Bh4 c6 12.e4 Bg7 13.c3

The position is a little bit funny since Roman produces videos on this opening.  Not being able to play ...g5 seems odd.

13 ... Nh5

I think the idea here is ... Nf8-g6 winning the two bishops or maybe planting a knight on ... f4.

14.Bc4 b5 15.Bb3 a5 16.O-O a4

Gaining some space for his pawn but ...

17.Bc2 a3

This reminds me of the early computers that would capture on rook three after 1.h4-h5-h6 and a4-a5-a6.  Kasparov called this type of pawn "the alien."

18.b4

18.b3 makes sense to me to avoid weaknesses, but the square b3 might be a good post for a Dzindi piece.

18 ... Nf4 19.Re1 Nf8 20.e5

A very strong mobility increasing move to pressure e7 and open up e4 for the White pieces.

20 ... d5

To shut down Dzindi's attacking lines but now 18.b4 looks brilliant.  The c5-square is a great outpost for a knight and Dzindi might win a second pawn with Qb1-b3 and Nb1:a3.

Rybka is probably programmed to prefer open lines, but what is she to do when opening lines makes things worse?

21.Bg3 Nh5 22.Nh4

Another problem with no g-pawn crops up - there is no g6 pawn to prevent Nf5 - Dzindi's other knight can also play Nf1-e3-f5.

Another point - Dzindi is controlling the tactics, typically with closed pawn structures and trading pieces.

22...Nxg3 23.fxg3

I would have played 18.b3 and 23.hg following my simple rules.  I assume the point of fg is to increase the pressure on f5 and f7.  23 fg almost looks like a bughouse strategy of covering your King with pawns.

23...e6

---------------------------------------------------------------------------

Rybka website

Topic Game 7. Dzindzichashvili - Rybka 1-0 (3.5-3.5 before game 8) By oudheusa Date 2008-03-09 12:20 i am surprised about the move 23... e6 as it closes out the black c8 bishop

By lkaufman Date 2008-03-09 14:13

Roman said that this move was unavoidable.

---------------------------------------------------------------------------

Hmmm.  Perhaps because of the f5 and f7 pressure.  23...e6 certainly looked like a stupid computer move to me.  Dzindi now manages to aim both knights at the outpost c5.

Roman has a habit of reducing Black's QB to misery - he likes 1.e4 e6 2.d4 d5 3.Nc3 Bb4 4.e5 c5 5.Bd2 and 1.e4 c5 2.Nc3 Nc6 3.f4 e6 4.Bb5 a6 5.B:c6.

I have a vivid memory of Dzindi pointing to a c8-bishop trapped inside a pawn chain of black pawns ...e6, ...d5, ...c6, ...c5 and white pawns a2, b3, c4, d3 laughing and asking where will it go?  The general strategy against a trapped piece is to open lines on the other side of the board where you are a "piece up".

Dzindi is making Rybka look like a $20 1600-rated 1980-version handheld joke in this game.

24.Qh5

Not only does Dzindi have the c5 knight outpost but also a winning plan of doubling rooks on the f-file.

24 ... Qg5

Rybka tries to bail out with a Queen trade.

25.Qe2

Roman sees no reason to empower Rybka's h8-rook and h6-pawn.

25 ... h5

Opening up the h6-square for Rybka pieces and supporting ... Qg4.  Rybka doesn't see the long-term effects of advancing a weakness deeper into an opponent's territory.

26.Nhf3 Qg4 27.Nb3

Immediately suggesting Rf1 Ne1-d3-c5.

27 ... Ra7

Supporting f7.

28.Nc5

I have a diagram in my book of the Zebra Formation - 1.a4 a5 2.c4 c5 3.e4 e5 4.g4 g5 with knights in all the outposts.  Dzindi can now reach a Queenside Zebra formation with Nd2-b3-a5.

28 ... Nd7 29.Nxd7 Kxd7

Rybka is still down a pawn, stuck with a hideous, trapped c8-bishop, a backward f7-pawn, while Roman can plant another knight on c5.

The proper way to trade pieces when you are winning is to post them on great squares that will win more material if the losing opponent doesn't trade.

30.Rf1

Better King, better rooks, better minor pieces, better pawns.

30 ... Bh6

Finally a smidgen of activity.

31.Rf2

To increase f7-pressure.

31 ... Kc7 32.Ne1

Knights are drawn to outposts like magnets.

32...Kb8 33.Nd3 Qg7 34.Raf1 Rc7 35.Nf4

Dzindi fakes left and goes right.  Roman told me he plays super-safe against computers.

Roman - "If anyone on ICC plays me like I play against the computers, I immediately censor them!"

A knight on c5 would just sunbathe, looking pretty but not accomplishing anything.  Early computer knights had a habit of just sitting on outposts, attacking nothing but enjoying the view.

To win Roman must pick up the a-pawn or mobilize his Kingside pawns.  Roman is a planner.

35 ... Bg5

If 35 ... B:f4 Roman might try gf-f5 or R:f4-h4, Rf1-f4, and Bd1 to annex the h5-pawn.

36.Nxh5!!!!

Beautiful - handing Black the exchange so he can roll with the kingside.  A small price to pay for progress.

36...Qh6 37.g4 Be3 38.h3 Bxf2+ 39.Qxf2

There is no defense to the inexorable march of the toy soldiers.

39 ... Kb7 40.Qf4

Now that he has a winning plan the Grandmaster is willing to trade Queens.

There are only four people I know of that were, at some point in their careers, both Soviet and US Chess Champions, and two of them are married - to each other!:  Dzindi, Alburt, Mr. and Mrs. Gulko.

40 ... Qf8

If Rybka traded Queens GM Dzindzichashvili (Gin-Jee-<cough>ash-vill-ee) would bring his King up to push the pawns.  Rybka is frozen on the Queenside and getting overrun on the Kingside.

Most humans seem delighted that Dzindi drew a line in the sand, a roadblock to the computers - Pawn and move against a GM?  -  you may go no further.

41.Nf6

I think we should vote on which side of the board Roman is making Rybka look sillier, his locked up Queenside or his imploding Kingside.

41 ... Kb6

Rybka's plan is to imitate a turkey on Thanksgiving.

42.g5

Here comes the axe.

42...Qg7 43.Rf3 Bb7

Feeling no fear, shame or humiliation is a major Rybka benefit at this stage.  A human would be squirming.

44.h4

Baby steps.  The old fox makes Chess look easy.

44 ... Qf8 45.h5

O' man river,
Dat ol' man river,
He mus' know sumpin'
But don't say nuthin'
He jes' keeps rollin'
He keeps on rollin' along.

45 ... Ka7

The Black King, QB and QR just walk around the prison yard on lockdown bummin' for cigarettes while the SWAT team tears up the other compound.

46.h6

c7-rook- You want me to help out?

h8-rook - Don't bother, I can barely breathe in this phone booth as it is.

46 ... Ba8 47.h7

Total triumph of strategy versus a tactical god.  The pawn and move helped.

47 ... Rb7

a8-bishop - Can you guys see anything?

b7-rook- Nope, prison wall's too high.

48.Qh4

Roman's king is waiting for a Queen trade.

48...Rc7

c7-rook- getting that Deja Vu feeling again.

49.Qh6

The Rybka basement is getting flooded - 49 Qh6 Rc8 50 Q:f8 Rc:f8 51 Nd7 or 49 Qh6 Qd8 50 Qg7 threatening 51 Ng8.

49...Qxh6 50.gxh6

Roman's plan - Rf3-g3-g8, Bc2-d1-h5:f7, g4-5-6

Rybka's plan - Kb6-a7-b6-a7-b6-a7-b6

50...Kb6 51.Rg3 Bb7 52.Rg8 Rc8 53.g4 c5

Rybka sees his f-pawn falling in all lines and tries desperately to break his chains.  Only problem is - Dzindi saw the problem 20 moves ago.

54.dxc5+ Kc6 55.Kf2

To snuff out the garbage can fire on d4 - nothing wrong with g5 and Bd1-h5:f7 either.

55 Ra8 56.Ke3

Stamping out the last vestige of hope.

56...Kc7 57.Kf4

New plan - waltz into the Kingside.  Other crushing ideas - Kd4 completing the lockdown, Rg7-Bg6 - releasing the f7-pawn from its failed and useless existence.

57...Kc6 58.g5 Kc7 59.g6

The Grandmaster could just win the f7-pawn with 59.Rg7 but sees no defense to a King march.

59...fxg6 60.Rxa8 Bxa8 61.Kg5

The best win is the simplest win - Bobby Fischer

61...d4

Rybka shows his sense of humor, finally freeing his bishop.  That must have cost Vasik lots of programming time.

62.cxd4 Bd5

63 Bb1 or K:g6 also works.  Maybe Dzindi wouldn't let Rybka resign.

63.Nxd5+ exd5 64.Kxg6 Re8 65.Kf7 Kd8 66.Bg6

Mating.

66 ... Re7+ 67.Kf6 Ra7 68.h8Q+ 1-0

Rybka resigns.  Mate in 6.

A fairly straightforward crush.

Maybe the programmers are not too upset since Rybka started each game with a losing position anyway.  I bet they make some adjustments from this match before releasing Rybka 3.0.

It's good to see we humans have a limit to our stupidity.  The match ended in a tie with 2 wins each.
 

Back to Going to the Wall

search tips

The
Chessville
Chess Store
 

Reference
Center

The Chessville
 Weekly
The Best Free
Chess
Newsletter
On the Planet!

Subscribe
Today -
It's Free!!

The
Chessville
Weekly
Archives

Discussion
Forum

Chess Links

Chess Rules

Visit the
Chessville
Chess Store