In a comment on my most recent post, Hal Bogner used the delightful sobriquet “high-class abacus” to describe chess computers. Quoth Hal: “What’s with this worship of high-class abacus evaluations of positions in which their evaluations are in hundredths of a pawn?”
Good question, Hal! In general I agree that you should not use a computer to evaluate a grandmaster game or any of your own games played under tournament conditions, until you have studied the game at least once without a computer. Occasionally I break this rule (for example, if I want to write about a game quickly for a blog post), but in general I stick to it. The reason is that to get the maximum educational benefit from a position, you should try to put yourself in exactly the same position again and try to do better. (Ideally, you would even recreate the tournament conditions, with a clock running as y0u analyze. However, even I am not that fanatical.)
However, there is one case where I think it’s fine to use a computer for analysis: going over blitz games. Blitz games are a dime a dozen. I never invest the kind of emotional or intellectual effort into them as a tournament game. For that reason, I do not think they demand the same effort in analysis. Blitz games are all about snap judgements, and so what I really want to know is: Are my snap judgements reasonably accurate, or is there something big that I’m overlooking? A computer can give you a quick and dirty answer to this question, which is appropriate for a quick and dirty form of chess.
Let me give two recent examples where the computer pointed out some misguided thinking in my blitz games. The first game was played at Mike Splane’s chess party last week. I was White against Austen Green, a master who moved to the Bay Area from Texas last year.
FEN: 1r2k1nr/p3ppbp/2p3p1/q1p1P3/2P1QP2/1P3N2/1B4PP/1R2K2R w Kk – 0 16
Position after 15. … Qa5+. White to move.
Austen has just gone pawn-hunting on the queenside with … Qxa2 followed by … Qa5+. I was pretty skeptical of this idea, because the queenside pawn majority is rather crippled, and meanwhile his development is lagging. Nevertheless, it’s time to get concrete here. How do I get out of check?
The move I played was 16. Kf2, with the idea of connecting rooks and hopefully getting an attack on Black’s king before he can castle. However, it doesn’t really work. Austen played 16. … Qb6, defending the c-pawns, and after 17. Rhd1 Nh6 18. h3 (I’m not sure whether this was necessary) O-O he was out of danger. After many adventures, including a positional piece sacrifice, I ended up losing on time in a won endgame.
Because it was such an interesting game, we had three other people joining in the postmortem, all masters. And yet not a single one of us even had an inkling of the best move in the above position. Do you see what it is?
The high-class abacus sees it: 16. b4!! Black cannot afford to take with the pawn or rook, because White would get to play 17. Qxc6+ trapping Black’s king in the center. And if Black takes with the queen, 16. … Qxb4+, now is the time to play 17. Kf2! connecting the rooks. It’s a whole lot stronger now than in the game, because there is an immediate discovered attack threatened on Black’s queen, and even if the queen steps aside from that, White is going to take over the b-file before Black can get his king to safety.
Ironically, in the game continuation 16. Kf2 Qb6 17. Rhd1, I would have been thrilled if Black took the b-pawn. But of course he wasn’t that stupid. By playing 16. b4!! I could have taken matters into my own hands, getting rid of the b-pawn and opening the b-file by force.
What do we learn from this? Five masters had the same blind spot — they wouldn’t consider a move that puts a pawn en prise in three different ways. I think there is a legitimate psychological issue here. I have no problem sacrificing a pawn if there is only one way that it can be taken, but somehow it’s twice as hard if it can be taken in two ways, three times as hard if it can be taken three ways, etc. It makes no sense, but I see this tendency in myself. So the high-class abacus in this case has succeeded in drawing my attention to a recurring flaw in my thought process.
On to example number two. This is from a blitz game that I played against Gjon Feinstein yesterday afternoon. Gjon was playing White, and I was Black.
FEN: r2q1k1r/pp3p1p/2n2B2/1Nbp3N/4p3/4P3/2PP1PPP/R2bKB1R w KQ – 0 13
Position after 12. … Bxd1. White to move.
Black has just taken White’s queen, and now has three pieces en prise: the bishop on d1, the rook at h8, and the queen. Which one should White take?
We spent 15 or 20 minutes after the game looking at the position just before this. White had played 12. Bb2xf6, allowing me to play 12. … Bg4xd1. Gjon had another option, to play 12. Bf1-e2, and that was what we spent all of our time analyzing in the post mortem. (The conclusion is that after 12. Be2 White has a slight advantage.)
Ironically, we spent zero minutes analyzing what White should do in the above position! We both operated under the assumption that of course, White has to take the queen. But it’s not true! The high-class abacus says that 13. Rxd1! leads to a big advantage for White. After 13. … Be7 (the computer’s recommendation) 14. Bxh8, White has a rook, bishop and knight for a queen. Let’s do the arithmetic, girls and boys. That’s 11 points. Black has 9 points. White should be two pawns ahead. And that’s pretty much what the computer says: White is 1.78 pawns ahead.
Why didn’t we consider this, either in the game or the post mortem? I think that there may be two separate blind spots at work. First is the tendency to always consider taking the largest piece first — especially where queens are involved. But I think that, as masters, we both understand that you have to “look at all the checks and captures.” That’s a mantra I tell my students at the Aptos Library Chess Club every week. So I’m going to blame it on a different, somewhat more sophisticated blind spot.
I think that Gjon and I, and most human chess players, have a great unease about going into positions with multiple pieces against a queen — for either side. Queen versus two rooks, queen versus three minors. We don’t have a lot of experience with such positions, and it’s tough to evaluate the nuances. Most of us would rather just avoid them.
So in the above position, after 13. Rxd1 Be7 14. Bxh8, my initial thought was that Black probably has compensation for the “two-pawn deficit.” There’s a protected passed a-pawn for White to contend with, and also White is a little bit behind in development.
On the other hand, here are some reasons the computer might be right.
- In a queen-versus-multiple-piece situation, the queen’s great forte is preying upon weak pawns. In this position White’s pawn formation is exceedingly solid and compact. Black is not going to be able to “steal” any pawns with forks or whatever.
- Black’s pawns can only go so far. They may get to b4 and a3, but it’s very hard to go any further. If Black chooses to play that way, it takes time — plenty of time for White to castle and get all of his pieces in play.
- Once White gets his pieces in play, they will take over the board. A rook, bishop and knight really are a heck of a lot of material. White has almost twice as many fighting units as Black does. They should be able to overwhelm Black’s queenside pawns in Benko Gambit fashion.
- Finally, let’s not forget that Black’s kingside has been somewhat compromised. If White manages to castle, pry open the f-file, and put his knight on f6, it will be a tower of strength.
Even with all this, I’m still not sure I accept the computer’s evaluation that the position is +1.78 for White. I would be interested in playing this position for either side. What do my esteemed readers think? What are the plans for both sides? Is the high-level abacus correct in thinking that White has a big advantage, or is it just doing the equivalent of 11 – 9 = 2?
In any case, I do think that there is a tendency for humans to discount such lines (R + B + N vs. Q) as being too “exotic” or “hard to evaluate.” If we find them hard to evaluate, we need to work on evaluating them better.
By the way, if you’re wondering what happened in the game, after 13. Bxd8 Bxh5 14. Bf6 Rg8 15. Nc7 we got to a position where I had to find a good move (and did).
FEN: r4kr1/ppN2p1p/2n2B2/2bp3b/4p3/4P3/2PP1PPP/R3KB1R b KQ – 0 15
Position after 15. Nc7. Black to play.
It looks as if White is going to win the d-pawn, with a significant advantage. What is Black’s saving move?
I believe that this was actually where Gjon made his mistake in analysis. When planning his 11th and 12th moves, he probably got to this position and stopped there, thinking that White was simply winning. The move he overlooked was 15. … Nb4! Of course, if White takes on a8, then 16. … Nxc2+ is not only a fork, it’s checkmate. Gjon played 16. Rc1, but then after 16. … Rc8 17. c3 Bb6 18. cb Rxc7 19. Rxc7 Bxc7 I got a slightly better endgame. As in the game against Austen, we eventually got to a position where I was winning on the board (queening in two moves) but my flag fell. In case you hadn’t noticed, that’s a pretty common motif when I play speed chess.
So in the end, we have to admit that chess is a pretty hard game, especially if you have only 7 minutes for the whole game. Perhaps if Gjon had spotted 15. … Nb4 earlier, he would have taken a harder look at the 13. Rxd1 idea, but since he thought he was just winning, he didn’t think he had to. That, too, is a very common human blind spot.
