The one with Deep Fritz...
What happens when you discover that your opponent is using a very strong chess program to help him with his game? One which can crank up to 20+ ply moves deep while your cranky brain coughs out 3 or 4 shallow and dubious lines?
It all happened in an ICCF Email Master Norm tourney...
Playing 16 moves of theory in a relatively unexplored English Opening gambit, I was quite taken aback to find the following email from my opponent, Hans van Unen...
1.c4 e6 2.g3 d5 3.Bg2 c6 4.Nf3 dxc4 5.0-0 Nd7 6.Qc2 Nb6 7.a4 a5 8.Na3 Qd5 9.Rd1 Nf6 10.Ne1 Qc5 11.d3 cxd3 12.Qxd3 Nbd7 13. Nc4 Ng4 14. e3 Nge5 15. Nxe5 Qxe5 16. Nf3 (16. Qc2 Qh5 17. Bd2 Be7 18. Bc3 O-O 19. Rd2 Deep Fritz -0.59/19})
So I made some conjectures,
1) My opponent is using Deep Fritz
2) He's cranking it up on a multi-processor computer which means blazing speed
3) Deep Fritz thinks my opponent's position is better
4) He might be trying to out-psyche me
5) I definitely need to win this one!
Moreover, I was very tempted to add (17.b3 0-0 19.Bb2 +1.27/28 Deep Blue Junior) for good measure.
What to do with an opponent who is assisted by a strong chess playing program? I looked hard at the position and was quite certain I had sufficient compensation for the pawn and was a trifle worried that my assessment might be faulty. I don't have Deep Fritz but I do possess a Fritz 6.0 engine which could be cranked up on my Pentium 233 PC. I am not sure whether the assessment evaluations of Deep Fritz differs much from that of Fritz 6 but at least I have an inkling what options my opponent might consider.
After the game had ended, I asked Steve Ham about how Fritz 6's assessments compares with Deep Fritz's. Here is his reply (condensed).
Dear Junior,
I confess I'm not knowledgeable about chess programs from first hand experience. However, I read and correspond with some of the nerds/geeks at the Computer Chess resource Center and have learned a lot from them. I can't answer your question very well, buddy, but I'll try nonetheless.
I'd guess that Fritz 6 should be able to predict most of Deep Fritz's moves since they are clearly related (2 generations apart on the same evolutionary scale) and share the same programmer (Franz Morsch) and distributor (ChessBase). A real improvement in computer chess programs occurred with Fritz 5.16 and then 5.32 was slightly better (smarter) due to larger hash tables. Next, Fritz 6 was considered to be a major improvement in "knowledge" without any appreciable speed loss. Then Fritz 6a not only fixed some bugs but also made the program clearly smarter with no appreciable speed loss. When Deep Fritz then appeared, there was some debate as to whether this was merely Fritz 6a rewritten for dual-computer (processor) use. However, significant testing shows that Deep Fritz is superior to Fritz 6a even on a single computer (processor). On comparable hardware, Deep Fritz is both faster and smarter than Fritz 6a. However, nearly all programs have weaknesses in technical positions (endgames and closed positions with pawn attacks upon castled Kings). Still, no program is more like Deep Fritz than Fritz 6a or 6b, so there should be closes imilarity in their general move selections.
You can get far more knowledgeable input from Uri Blass, whom I know now owns a copy of Deep Fritz. He's both very knowledgeable and objective in his statements and is a Deep Fritz fan.
So, I queried Uri about this too and his answer (condensed) was:
I do not have Fritz6 and only Deep Fritz so I do not know if Fritz6 can reproduce the same moves of Deep Fritz. I read that the main difference between Deep Fritz and Fritz6 is that Deep Fritz has better knowledge of king safety. I believe that there is not a big difference between Deep Fritz and Fritz6 in most positions.
Uri
I removed the Deep Fritz sideline from the email and sent my following moves. For the next 6 emails or so, the Deep Fritz evaluations continued with 18-21 ply deep assessments before he finally caught on. I'm not sure how my opponent could have allowed the Deep Fritz lines to emerge on the emails. Perhaps he did it on purpose or his email chess program somehow transmitted the moves.
I normally do not trust the evaluation of chess programs very much. Who cares if the computer says +0.11 or +0.19? Moreover, sometimes they value material over initiative a trifle too drastically (I think). The exception is when the evaluation suddenly turns drastically which I have probably missed a big tactic somewhere. However, Fritz 6 somehow agreed with both parties' choices of moves until I reached move 28 when I decided to hit my opponent on the Kingside with a move that was not even on Fritz 6's top ten choices! Fritz 6 also disagreed with me on move 30 when I decided to shift my Queen away from her protection of the queenside pawns. On move 31, I gambitted another pawn when I was certain that I had sufficent compensation. Moreover, Fritz 6 had not even considered the move. To my surprise, my opponent declined the pawn and allowed me to liquidate to a better endgame. I was very surprised. Could Deep Fritz have worked out that the pawn was unpalatable? Well, you can read about Han's thoughts about the move in the game notes. The endgame was difficult for Hans and fortunately for me, he missed the drawing resource which involved the sacrifice of a pawn! What really shocked me was how badly Fritz 6 misevaluated the endgame, indicating that the position was equal when it was absolutely lost!
In a TCCMB post, CC-GM Tunc Hamarat has this to say about Fritz's endgame skills:
"One big weak point of computer programs is the endgame with one exception. They are perfect in 5 piece endgames (they are also trying to finish 6 piece endgames with enormous disk space). Another point is that Fritz is a combinative program which is very weak in endgames. Just to give an example, Schredder treats endgames incomparably better.
Of course all these programs play a big role in tournaments which are not so strong. But in strong tournaments, using computers might be even worse for your results."
Watching how Deep Tiger mauled South Americans GMs and IMs in a recent tourney and how much trouble Deep Fritz is giving German GM Robert Hubner in Dortmund, our CC days are certainly numbered. But until then, we should really brush up our endgame skills to bamboozle the computer programs (if we do survive till this phase of the game...).
I must note that I have absolutely no idea whether Hans van Unen is allowing Deep Fritz to dictate his move choice. Well, enough talk, to the game proper...
This is my 2nd game against Hans van Unen. In ICCF EM/M/A092, I essayed the Benko gambit against him which ended in a draw. He went on win the tournament with 8 out of 10 while I ended up a pathetic 4th with 6.5 points. 1.c4 e6 2.g3 d5 3.Bg2 c6 Hans is a strong Slav Defence exponent 4.Nf3 Another English gambit, reminiscient of Tay-Christov, Interzonals 2000 - See CCN 37 for more details. 4...dxc4 5.0-0 [White chose not to gambit the c pawn in the following game: 5.a4 Nf6 6.Qc2 Na6 7.Qxc4 Nb4 8.0-0 Be7 9.Nc3 0-0 10.d4 b6 11.Rd1 a5= 12.Ne5 Ba6 13.Qb3 Rc8 14.Bg5 h6 15.Bxf6 Bxf6 16.Nc4 Qc7 17.e3 ½-½ Marin,M-Jonkman,H/Tel Aviv ISR 2000] 5...Nd7 6.Qc2 Nb6 7.a4 a5 8.Na3 Qd5 [I think after 8...Bxa3 9.bxa3 (or 9.Rxa3 ) 9...Nf6 10.Bb2 White already has the better game.] 9.Rd1 Nf6 10.Ne1 Qc5 11.d3 cxd3 12.Qxd3 Nbd7 13.Nc4 Ng4 14.e3 Nge5 15.Nxe5 Qxe5 16.Nf3 Qc7 17.b3 Be7 18.Bb2
|
So far we have been following Kuznetsov,S-Fediashin,S/URS 1994. White's compensation is more evident here with open lines and free play for all his pieces. 18...0-0N [18...Bf6 19.Bxf6 Nxf6 20.Qd4! 0-0 21.Ne5!
|
It is difficult to shake off White's dominance in the central files. Indeed, Fediashin got slowly but surely snuffed out of the game by the Russian CC-IM who is rated 2442 OTB and 2453 ICCF. 21...Nd5 22.Rac1 f6 23.Nc4 Kh8 24.e4 Ne7 25.f4 Ng6 26.e5± Kg8 27.Rd2 Ne7 28.Qc5! fxe5 29.fxe5 g6 30.Rcd1 Ra6 31.Rf2 Rxf2 32.Qxf2 Nd5 33.Rf1± Black's pieces are badly hemmed in. Kuznetsov,S-Fediashin,S/URS 1994/1-0 (45)] 19.e4 Nc5 20.Qc2 Qb6 21.Nd2
|
21...Bd7
[Perhaps it's high time to return the pawn to complete development 21...e5! 22.Bxe5 Bg4 with free play for Black's pieces] 22.Ba3!? The prelude to the annexing of d6. Black's next four moves are practically forced. [Here, Kuznetzov recommends 22.Rac1 Na6 23.Nc4© The game may continue 23...Qc7 24.Nxa5 Nb4 25.Qd2 Rfd8 26.Nc4 Be8² With hindsight, I'd say that Kuznetzov's idea is still a simpler way to keep the advantage.] 22...Rfe8 23.Rac1 Na6 24.Nc4 Qc7 25.Nd6 Reb8 26.e5 Be8
|
Now all the strategic aims have been realised and the Black pieces have been tied down. Thus, White should aim at the Black K now. However, it only takes Nb4, Rd8 and Bxd6 for Black to free his pieces so White has to proceed swiftly. [26...Nb4 27.Bxb4 axb4 28.Qc4 Ra5 29.Qd4²] 27.Be4! [The computer try 27.Nb5? only leads to a worse endgame after 27...cxb5 28.Qxc7 Nxc7 29.Bxe7 Nd5 30.Bxd5 exd5 31.axb5 Bxb5 32.Rxd5 Bc6³ Black's eventual creation of a passed a pawn gives him the edge though the opposite colour bishops indicates that White can hold the game. ] 27...h6 At this point, I felt that White is ready to launch a Kingside attack given his mobility and Black's passive pieces so... 28.f4!?
|
This natural move is even considered in Fritz 6's top 10 choice of moves! OK, I'll crank it up for 5 minutes to show you what it considered. [Once again, the liquidating 28.Nb5 doesn't achieve much after 28...cxb5 29.Qxc7 Nxc7 30.Bxe7 Nd5 31.Bxd5 exd5 32.axb5 Bxb5 33.Rxd5 Bc6³; Fritz 6: 28.Rd4 Nb4 29.Bxb4 axb4 30.Rcd1 Qa5 31.f4 Bxd6 32.Rxd6 Rd8; Fritz 6: 28.Rd3 Rd8 29.Rcd1 Nb4 30.Bxb4 axb4 31.Rd4 Qb6 32.Bh7+ Kh8; Fritz 6: 28.Rd2 Rd8 29.Rcd1 Nb4 30.Bxb4 axb4 31.Rd4; Fritz 6: 28.Bg2 Rd8 29.Qc3 Rd7 30.Rd2 Rad8 31.Nxe8 Rxe8 32.Rxd7 Qxd7; Fritz 6: 28.Qc3 Rd8 29.Bg2; Fritz 6: 28.Bh7+ Kh8 29.Be4 Rd8 30.Rd4 Qb6 31.Rd3 Nb4 32.Bxb4 Qxb4; Fritz 6: 28.Qc4 Rd8 29.Rd3 Qb6 30.Rcd1 Rd7 31.Bg2 Nb4 32.Bxb4 Qxb4; Fritz 6: 28.h3 Rd8 29.Rd4 Rd7 30.Qc3 Rad8 31.Nxe8 Rxe8 32.Rxd7 Qxd7; Fritz 6: 28.Qd3 Rd8 29.Rc4 Rd7 30.Rd4 c5 31.Nxe8 Rxd4 32.Nxc7 Rxd3] A you can see, Fritz's choices are mainly geared towards strengthening the centre by doubling rooks. It also feels that Black has a slight edge ranging from 0.31 to 0.47 of a pawn. However, perhaps due to the feeble processing power of my laptop, it can only run at 10 ply (5 moves aside) deep after 5 minutes of analysis. Anyway, the fact that it rejects 28.f4 offers food for thought 28...Qb6+ Attempting to clamp down on the Ba3 by controlling b4 29.Kh1 Nb4 30.Qe2 Fritz 6's 1st, 2nd and 4th choices are as follows. [Fritz 6: 30.Qc4 Rd8 31.Bxb4 axb4 32.Rd4 Ra7 33.Bg2 Rd7 34.Rcd1 c5; Fritz 6: 30.Bxb4 Qxb4 31.Qc4 Qb6 32.Rd3 Rd8 33.Rcd1 Rd7 34.f5 Qb4; Fritz 6: 30.Qb2 ] 30...Rd8 31.Bb2! Fritz 6 fails to consider this alternative, perhaps because it loses more material. I suppose Fritz 6 is programmed to bolster up the centre rather than to ditch material for the Kingside attack. [And the 3rd best alternative worked out by Fritz 6 on move 30 runs: 31.Bxb4 Qxb4 32.Rc4 Qb6 33.Qc2 Ra7 34.Rcd4 c5] 31...c5!
|
After the game, I asked Hans about this move. His reply was "I played 31...c5 because I was afraid of Ba3 diagonal in most of the variations. I figured if I could block the diagonal and move my knight to d5 I could then capture the white knight. The plan was good, but I never considered entering the endgame..so soon!" [Lashing out with 31...f5 is blatantly foolhardy as 32.exf6 (The flashy 32.Qc4 only draws by perpetual check after 32...fxe4 33.Qxe6+ Kf8 34.Qf5+ Kg8 35.Qe6+ Kf8 36.Qf5+ Kg8 37.Qe6+=) 32...Bxd6 33.Qg4 Qc7 34.fxg7 Bf7 (34...Qf7 35.Qh4 h5 36.Qg5 e5 37.Qh6 Qxg7 38.Rxd6 Qxh6 39.Rxh6+-) 35.Qh4 h5 36.Qf6+-; 31...Nd5 32.Bd4! What Fritz 6 missed! 32...Qxb3 33.Rb1 (33.Bc2 Qb4 34.Bc5 Qxc5 35.Bh7+ Kxh7 36.Rxc5±) 33...Qa3 (33...Qxa4 34.Bc2 Qa3 35.Nc4+-) 34.Nc4 Nc3 35.Qf3 Nxb1 36.Nxa3 Nxa3 37.f5‚; 31...Qc7 32.Qg4! A) 32...f5 33.exf6 Bxd6 34.fxg7 Qf7 (34...Bf7 35.Qh4 h5 36.Qf6+- intending Qh6) 35.Qh4 Qh5 (35...h5 36.Qg5+-) 36.Qf6; B) 32...Qb6 33.Bd4 Qc7 34.f5‚; 31...Bxd6 32.exd6 c5 (32...Nd5 33.Qg4 f6 34.Bxd5 exd5 35.Bxf6 Rd7 36.Rd3±; 32...Rd7 33.f5‚) 33.Qg4 g6 34.Bxg6 fxg6 35.Qxe6+ Bf7 36.Qf6+-] 32.Nxe8 Rxd1+ [Keeping the 2 Rs would require a huge defensive task after 32...Rxe8 33.Qf3 (33.Qb5 Qxb5 34.axb5 Nd5 35.Bxd5 exd5 36.Rxd5 a4 37.bxa4 Rxa4 38.Bc3 b6 39.Rd7²) 33...Nc6 34.Rd7 Rac8 35.f5! with a good attack.] 33.Rxd1 Rxe8 34.Rd7 Nd5 35.Qb5!
|
Forcing Black into an ending where he has to tread with extreme caution 35...Qxb5 36.axb5 b6 37.Bxd5 exd5 38.Rxd5 Kf8?
|
Hans offered a draw here but his position is already difficult. Fritz 6 actually claims that the position is equal! I don't know how Han's Deep Fritz evaluates this position... [Black must not remain passive. It is imperative to sacrifice a pawn with 38...c4! 39.bxc4 Bc5 and the passed a pawn needs watching A) 40.Rd7 a4 41.f5 (41.Ra7?? Rd8-+) 41...a3 42.Bc3 Re7! 43.Rxe7?? (43.Rd8+ Kh7 44.Kg2 Ra7 45.e6 fxe6 46.fxe6 Re7 47.Rd7 Kg6 48.g4 a2 49.Ba1² So far, I am unable to find a winning plan for White. 49...Bb4 50.Kf3 h5 51.h3 hxg4+ 52.hxg4 Kh6²) 43...Bxe7 44.Kg2 f6! Creating a strong point at e5 ends the game... 45.Kf3 (45.e6 Bd6-+ followed by Be5) 45...fxe5 46.Ke4 a2 47.g4 Bc5 48.Bxe5 Be7-+ and Bf6 ensures that the a pawn promotes...; B) 40.Kg2 a4 B1) 41.Kf3 a3 42.Ba1 Kf8 43.Ke4 (43.f5 Ke7 44.g4 a2 45.h4 Ra8 46.f6+ gxf6 47.exf6+ Ke6 48.Re5+ Kd7 49.g5 hxg5 50.h5 Ra4 51.Re4) 43...a2 44.Rd7 Re7 45.Rd8+ Re8 46.Rd7=; B2) 41.Rd3 41...Rc8 42.f5 Bb4 43.e6 Rxc4 44.Rd8+ Kh7 45.Bd4 fxe6 46.fxe6 Kg6 47.Bxb6 Kf5 48.Rd4 Rxd4 49.Bxd4 Kxe6 50.Bxg7 Kd5=; C) 40.Bd4 40...Bxd4 41.Rxd4 a4 (41...Ra8 42.c5! bxc5 43.Ra4±) 42.c5 bxc5 43.Rxa4 Rb8 44.Ra5 c4 45.Ra4 c3 46.Rc4 Rxb5 47.Rxc3 h5 with excellent drawing chances.] The interesting part about this ending is that the mandatory 38...c4 is not even among Fritz 6's top three choice of moves. 39.f5! Rd8??+- This leads to a lost ending without any saving chances. White's King has a clear path to c6 and he can also create a passed e6 P. Fritz 6 even rates this position as equal. [Now 39...c4!
|
still works40.bxc4 A) 40...a4 41.Bd4 (41.Kg2 a3 42.Ba1 Bb4©) 41...Rc8 42.f6 gxf6 43.exf6 Bd8 44.c5 bxc5 45.Rxc5+-; B) 40...Bc5 41.Rd7 Re7 42.Rxe7 Kxe7 43.Kg2 f6 44.Kf3 a4 45.Ke4 a3 46.Ba1 Bg1! (46...a2 47.Kd3 Bg1 48.exf6+ gxf6 49.h3 Bh2 50.Kc2 Bxg3 51.Kb3+-) 47.h3 Bf2 48.g4 a2 49.Kf3 Bg1 50.Kg2 Be3=] 40.Rxd8+ Bxd8 41.Kg2+-
|
Fritz 6 still regards the position as equal but Junior Tay (not Deep Junior / Junior 7.0) evaluates this as completely won for White. 41...g6 [Now the drawing idea doesn't work anymore because the White K centralises swiftly 41...Be7 42.Kf3 c4 43.bxc4 Bc5 44.Ke4 a4 45.Bd4] 42.g4 h5 43.h3 Bg5 44.Kf3 gxf5 45.gxf5 Ke7
|
Only here does Fritz 6 give White a += ! 46.Ke4 Kd7 47.Kd5
|
The Black K can only look on as White creates an e6 P which the former has to guard. Meanwhile, the White K mops up the Black Queenside. 47...Bf4 48.e6+ fxe6+ 49.fxe6+ Ke7 50.Bg7
|
Circling the B around to eventually control the e7 square. Now Fritz 6 sees White's idea and it evaluates White as having a clear advantage. 50...Bg3 51.Bh6 Ke8 52.Bg5 Bh2 53.Kc6 Bg1 54.Kxb6
|
And finally, Fritz 6 acknowledges this position is lost for Black. 54...c4+ 55.Kxa5 cxb3 56.Bf6 b2 57.Bxb2 Ke7 58.Be5 Kxe6 59.Bg3 Kf5 60.b6 Bxb6+ 61.Kxb6 Ke4 62.h4
|
When Hans finally resigned, he said. "Well, the only consolation for me is that I also play this opening as White!". 1-0
Although I cannot verify how much Hans depended on Deep Fritz for move assessment and selection, I think there is still a lot of room for innovation and anti-computer strategy. I did not intentionally go into lines which I felt Deep Fritz might not be assessing accurately but I was quite certain it would not see how that the sacrifice of the c pawn was necessary to hold the position in the endgame. Gambits and pawn sacrifices, especially positional ones might mess up the program's assessment but do beware, a pawn is a pawn! You really got to be certain!