The Chess Mind

A few days ago, I took a look at the correspondence game Umansky-Weber, focusing on Junior Tay's informal claim that Umansky's 16.Nd5 was an (intuitive) idea most likely beyond the ability of the silicon monster. What I found was that it was well within the range of current software, as Shredder 9 on my machine preferred the move and recognized its strength within 20 minutes.

GM Alexander Baburin of Chess Today fame wrote to ask how well Shredder 9 would fare on the prototypical speculative Nd5 sac, that from the famous 10th game of the 1965 Candidates match between Mikhail Tal and Bent Larsen. (Tal won this match 5.5-4.5, thanks to his win in this game.)

Let's take a look. After the introductory moves

Tal Mikhail (LAT) - Larsen Bent (DEN) [B82]
Ch World match (1/2) Bled (Yugoslavia), 10.03.1965

1.e4 c5 2.Nf3 Nc6 3.d4 cxd4 4.Nxd4 e6 5.Nc3 d6 6.Be3 Nf6 7.f4 Be7 8.Qf3 0-0 9.0-0-0 Qc7 10.Ndb5 Qb8 11.g4 a6 12.Nd4 Nxd4 13.Bxd4 b5 14.g5 Nd7 15.Bd3 b4

we reach this position:



Here Tal played the speculative 16.Nd5 and won beautifully, though objectively - at least according to the latest analysis I've seen (especially Kasparov's in My Great Predecssors, vol. 2) - the sacrifice was not fully sound.

Interestingly, Shredder 9 starts out quite optimistically, evaluating both 16.Ne2 and 16.Nd5 as equal through 13-14 ply (within about 30 seconds to a minute on my machine), but as the depth increases, so too does its pessimism about 16.Nd5. At depth 18, Shredder thinks Black is clearly better (-.88), and it prefer the other three knight retreats, with 16.Ne2 still leading the pack.

A quick replaying of Kasparov's main line (with is also in the Chess Stars series on Tal) looks in keeping with Shredder's evaluation, so it seems that in this case, the computer won't play like Tal - it sees too much!

Thus, we have found something the computer can't do, or at least doesn't do: bluff!

[Here, for the curious, is the main line of the (at least apparent) refutation of Tal's sac, as well as the rest of the game:]

16.Nd5 exd5 17.exd5 f5

[Correct was 17...g6! - the first of a series of only moves which together convey a near-miraculous impression: 18.Rde1 Bd8! 19.Qh3 Ne5! 20.Qh6 Bb6! 21.fxe5 Bxd4 22.Re4 Bf2 23.e6 fxe6 24.dxe6 d5 (24...Bb7 25.e7 Bxe4 26.exf8Q+ Qxf8 27.Qxf8+ Rxf8 28.Bxe4 Be3+ 29.Kd1 Bxg5 "with an extra pawn" and winning chances, according to Kasparov, but Chess Stars (Khalifman?) evaluates this as equal.) 25.Re2 Qa7 (Here Kasparov offers a new move, at least with respect to the Chess Stars analysis) 26.Bxg6 hxg6 27.Qxg6+ Qg7 28.Qxg7+ Kxg7 29.e7 Re8 30.Rxf2 Rxe7 "with a technically won game", according to Kasparov in My Great Predecessors vol. 2, page 451]

18.Rde1 Rf7 19.h4 Bb7 20.Bxf5 Rxf5 21.Rxe7 Ne5 22.Qe4 Qf8 23.fxe5 Rf4 24.Qe3 Rf3 25.Qe2 Qxe7 26.Qxf3 dxe5 27.Re1 Rd8 28.Rxe5 Qd6 29.Qf4 Rf8 30.Qe4 b3 31.axb3 Rf1+ 32.Kd2 Qb4+ 33.c3 Qd6 34.Bc5 Qxc5 35.Re8+ Rf8 36.Qe6+ Kh8 37.Qf7 1-0

While I'm as happy as anyone to see humanity continue the battle with chess software, I have for some time now grown weary of pronouncements that computers can't or won't find this or that move.

Sometimes, of course, it's true: to take one standard case, computers do have a tough time recognizing that certain blockades are airtight. But I think that if I had a quarter for every time I've heard a false claim about what chess software can't do, I'd be able to afford a trip to Europe this summer.

Enter tonight's issue of Chess Today (#1638). Junior Tay writes about the role and increasing strength of chess computers vis-a-vis correspondence chess, but then presents a game under the header "Exclusion Clause" (that is, this game is an [alleged] exception to the encroachment of chess software on correspondence chess), introducing the game as follows:

This is not to say that the human element in normal CC [DM: Correspondence Chess] has been quashed. I once wrote an article called "Umansky - the wave of the future" to pay tribute to the 13th [DM: correspondence] World Champion's enterprising style. It's almost incredible to believe that Shirovian-type chess can be played in CC today when computer engines have more or less mastered tactical calculations. Don't believe me? Watch this game:

Umansky,Mikhail (2524) - Weber,Daniel (2176) [E04]
AC-2003-S-00002 Chessfriend.com (1), 01.12.2004

1.d4 Nf6 2.c4 e6 3.g3 d5 4.Bg2 dxc4 5.Nf3 Nc6 6.0-0 Rb8 7.Nc3 a6 8.e4 b5 9.d5 Nb4 10.b3 cxb3 11.Qxb3 c5 12.dxc6 Nxc6 13.Bf4 Rb7 14.Rad1 Nd7 15.Qc2 Na5



Tay writes the following: "Now Umansky comes up with a remarkable (intuitive!?) idea which I doubt computer programs can spit out."

It is a beautiful move, well-calculated and though reminiscent of other Nd5 sacs (as in the Sicilian, for instance), the position is sufficiently different that one cannot unthinkingly make the sacrifice based on the Sicilian analogy.

Despite its many virtues, however, inaccessibility to computer programs is not one of them. Shredder 9, running on my above average but not state of the art system, ranked the move in the middle of the pack through depth 15, by depth 16 (around 10-15 minutes in) put it into a tie for the second-best move, and by the 20-minute mark if not sooner, had it as a clear #1 choice at depth 17. At that point, once I let it go forward, it found all but two of Umansky's next 10 moves instantly, and those other two moves required 5-10 seconds to reach the top of the chart.

This doesn't diminish Umansky's achievement, but it does underscore Tay's worry elsewhere in the article, that correspondence chess is under severe pressure from computer users (sometimes cheaters, sometimes not - different correspondence events have different rules). To my mind, it's a shame that such a noble form of the game is being undermined in this way - I don't mind if the players all agree to use computers, but if player X wants a mano-a-mano battle but player Y surreptitiously "consults" with Mr. Chips, then that seems to me unfair.

Readers, thoughts?

16.Nd5!! exd5 17.exd5 Be7 18.Nd4! 0-0 19.Nc6 Nxc6 20.dxc6 Ra7 21.c7 Qe8 22.Rfe1 Nc5 23.Bd6 Bxd6 24.Rxe8 Rxe8 25.Bc6 Rf8 26.Rxd6




The smoke has cleared, the tactics are over, and all that's left is a technical task Umansky is more than able to solve.

Ne6 27.f4 Rxc7 28.Qd2 g6 29.g4 Ng7 30.Qa5 Ne8 31.Bxe8 Rc1+ 32.Kf2 Rxe8 33.Rd8 Rxd8 34.Qxd8+ Kg7 35.f5 Rc6 36.Qd4+ Kf8 37.Qh8+ Ke7 38.Qxh7 gxf5 39.g5 Rc2+ 40.Ke3 Rg2 41.h4 Be6 42.Qh8 a5 43.Qf6+ Kd7 44.h5 Rxa2 45.h6 Rh2 46.Qg7 Kd6 47.h7 Rh3+ 48.Kf4 Rh4+ 49.Kg3 Rxh7 50.Qxh7 b4 51.Qh8 Kd5 52.Qa8+ Kc4 53.g6 fxg6 54.Qc6+ Kd4 55.Qxe6 Kc5 56.Kf4 Kd4 57.Qd6+ Kc3 58.Ke3 1-0