The Chess Mind

There are just too many events going on! Fortunately, almost all of them end today and tomorrow.

The Bosna tournament in Sarajevo is one that ends tomorrow (it's always important to at least finish before Mark Crowther finalizes TWIC), but first place was clinched today by Eljanov. With a win, Predojevic could have stayed in contention, but he had to work to draw with Eljanov. Likewise, Movsesian failed to defeat Harikrishna, so Eljanov's 1½ point lead remains intact entering tomorrow's final round.

The 17th World Computer Championship finished today, and Rybka remains the king (or queen, as its programmers fancy it a she) of the computer chess world. Rybka led Junior by half a point coming into the last round, and defeated it in a good tough game.

Final Standings:
1. Rybka 8 (of 9)
2-4. Shredder, Junior, Deep Sjeng 6½
5. Hiarcs 6
6. Jonny 4½
7. The Baron 3
8. Equinox 2
9. Pandix 2009 1½
10. Joker ½

Forget Anand and Topalov; the real champions (even if they don't actually exist as things in their own right) are playing. The 17th World Computer Chess Championship is underway in Pamplona, Spain, and after four rounds of this 10-engine round-robin, both Junior and Rybka are 4-0. Fritz isn't playing, but those familiar with the ChessBase stable will be glad to see that Shredder and Hiarcs are tied for third (with Jonny).

Fans of his opening work and readers of Chess Today will be familiar with GM Mikhail Golubev. A recent and fairly extended interview with him can be found starting on this page, but here I'll draw your attention to something he said near the very end:

[Question:] I practically never saw nice websites for those who may wish to learn how to play chess from the complete zero level...

[Golubev's Reply:] Personally, I prefer, for example, to analyse some Fischer Attack game with a novelty around the move 20. I like to annotate games. Also to annotate them in the "Informator style", without any words - in such way I annotated games for Informator and New in Chess in the pre-computer era. With some ideas and, not too often, decent quality. But I am afraid that nowadays it is possible to teach [a] monkey how to push a few buttons inside the Rybka or Fritz interface, and the result will be of [...] better quality than these old notes of mine... I see that explaining computer’s variations becomes a larger and larger part of the annotator’s work. This trend is not new, but it started to really disturb me only recently, right now I am still not sure what to do about that. Nothing dramatic, in any case. [Emphasis added.]

Ah, those darned monkeys! In truth, the monkeys won't replace the GMs. They really just exist online in the form of 1300s berating the grandmasters they're watching for missing some "obvious" move found by their quad-cores running Rybka; those of us who want to learn will read what strong analysts have to say. We might check their analysis with the computer, but whatever we come to understand almost always results from the text, not the printout.

Here's an interesting email from "Seena":

Hello Dennis,

Just a curiosity about computer programming and chess softwares; there's no expert on either field around here, so I thought it's a good idea to ask you: How does a chess software decide that in a certain position, no furthur calculation is needed, and the evaluation of the position (which is, for a computer, based on materialistic considerations) is satisfactorily precise? Grandmasters do that with their so-called "positional judgement". They can "visualise" the forthcoming position and evaluate it due to some well-known considerations (pawn structure, good/bad bishops, "compensation" for material, and so on). Hence, even though their calculating ability is limited, they can play the game masterfully.

Computers, on the other hand, have tremendous calculating powers but don't have a clue about positional play. So how do they evaluate positions? They can't calculate variations all the way down to the last move, so they have to "cut" the variations at some point. In a game with limited time, how do they decide that they've gone deep enough and the evaluation of the position will not suddenly change , in the very next move to come?

I'm trying to develop a draghts-playing computer programme, so any information on this issue would be helpful.

Thanks in advance! Seena

I'm not a programming expert either, so I'll leave a fuller discussion to my more learned readers, who are also invited to point Seena to other resources that might prove helpful. One note, though: I wouldn't say they have no clue about positional play (at least if we ignore ontological issues and assume for the sake of discussion that they have a clue about anything). While they lack concepts and the ability to wield them as we do, programmers do build positional ideas into their algorithms. The trick is just to mathematize features like space, mobility, pawn structure, bishops vs. knights vis-a-vis the pawn structure, control of open files, diagonals, king safety, etc. We make the same sorts of judgments in a more informal and intuitive way when we say things like "White is better thanks to the bishop pair", or "Black's superior pawn structure outweighs the opponent's bishop pair." (Very) broadly speaking, the computer makes the same sorts of judgments we do.

As you may all remember, Shakhriyar Mamedyarov lost, and quickly, to Igor Kurnosov in a late round of the recently finished Aeroflot Open in Moscow. After the game - at the end of it, really - Mamedyarov, bothered by the conjunction of Kurnosov's leaving the board after practically every move and his coming back to make crushing moves. Mamedyarov resigned after Kurnosov's 21st move, filed a protest over his opponent's behavior, and then withdrew from the tournament.

As I argued here (and others, including Kurnosov (here), have similarly argued), Mamedyarov's case for his opponent's cheating isn't very impressive on the face of it. (That doesn't mean that Kurnosov's behavior couldn't have aroused suspicion, only that his performance in the game (a) fell short of a Rybka standard and (b) didn't require Rybka either, but was well within the competence of a 2600-level grandmaster.) Ideally, that would be the end of the cheating allegation, and then discussion can move on to appropriate measures that allow players to roam around and smoke their lungs out while ensuring that they can't readily cheat.

Unfortunately, this was not the end, and Mamedyarov has released another open letter. You can check the link just given, but here's the gist:

In round 2 vs. Onischuk, Kurnosov responded to a novelty by making 15 Rybka moves, winning beautifully.

In round 4, vs. Moiseenko, he again responds strongly to a novelty with 14 Rybka moves, but this time the opponent's position was too solid and the game was drawn.

In round 6, we have the Mamedyarov game, again featuring Kurnosov playing Rybka's moves.

In round 8, now under heavy scrutiny thanks to Mamedyarov's allegations, he played poorly and got crushed.

The conclusion of his argument is that "this is [a] clear indication that Kurnosov used a computer program, leaving the tournament hall practically after every move".

Well, no, this doesn't follow at all. As already mentioned, his argument concerning his own game is very poor, and the inference he drew about round 8 is really preposterous. Let's suppose for the sake of argument that Kurnosov is innocent. His name has been dragged through the mud in a very public way, and now he's supposed to play at full strength, with confidence and complete concentration? Not very likely. Not only will he be emotionally affected, but now he can't even engage in his usual smoking routine because now it will continue to raise the suspicion. So what does he do? If he goes to smoke, he has to worry what others will think, and if he doesn't, his mind won't work the way he expects it to. If anything, his lousy round eight performance might be counter-evidence: if Kurnosov was such a brazen cheater (cheating every single move in all [but only?] his Black games, then even if he couldn't use Rybka, he'd at least be strongly motivated to play well to keep up appearances. And he is a 2600, for goodness' sake!

I'm not impressed by the round 6 or round 8 parts of his case, but how about rounds 2 and 4? (And why not rounds 1, 3, 5, 7 and 9?) First of all, Mamedyarov's claim that Kurnosov always went with Rybka's first choice is simply wrong, or at least I wasn't able to reproduce Mamedyarov's results. Further, many of the moves in question were fairly obvious. Some were components of a straightforward plan, some moves were obvious (e.g. recaptures) and some were elements in a tactical sequence. Out of the 29 post-novelty moves in the two games, only one - 21...Qf5 in the round 2 game - strikes me as even a candidate.

I can understand Mamedyarov's frustration in the situation, and I also get the tendency to dig in and retrench when criticized, as has happened since his initial salvo. But I don't find his argument any more plausible this time around, and I hope he'll drop the accusations (at least in the absence of far more impressive evidence than he has offered so far) as soon as possible.

I've already given and annotated Mamedyarov-Kurnosov; here are the three further games Mamedyarov refers to.

HT: Harris Nizel

Charles Sullivan of TrueChess.com has attempted to work out an answer by having his machines analyze almost 19,000 games by world champions, and you can find his method and results here. There are other ways to make assessments about the "greatest" player, but his data merit a look.

...is underway in Beijing, though with computer programs running it really doesn't matter where it happens. As of this report in ChessBase, at about the halfway point, Rybka and Hiarcs are tied for first with 4.5/5, Junior is half a point further back, and then there's a gap to the rest of the field. (Surprisingly, my old favorite program Shredder is languishing with a -1 score; hopefully this will serve as an impetus for its programmers to work hard and return their creation to the forefront of computer chess.)

No word yet on whether any of the programs are trying to contact Kramnik or Topalov during the games.

Both are interesting and worth your time. One article takes a look back at Indian GM Humpy Koneru's tough semi-final loss to Hou Yifan at the recently completed women's world championship. What went wrong? Some possible answers are floated by the author, but I think Anand's very gracious comments at the end will probably prove the best balm for Humpy's heartbreak.

The second article recaps another Rybka odds match, this time against 2705-rated Swiss GM Vadim Milov. Milov won 4.5-3.5 overall, going .5-1.5 in the two games where the only handicap was his having White; going 1.5-.5 when he had pawn and move; and scoring 2.5-1.5 with exchange odds. It was a good match, especially since Milov did a nice job of avoiding a common pitfall against the computer, to wit, accidental self-destruction. (The games are here, on the ChessBase site.)

The other day I was preparing a summary sheet on the French for one of my students, and while looking through Powerbook 2008 within ChessBase 10, I came across a shocker. After 1.e4 e6 2.d4 d5 3.exd5 exd5 4.Nf3 Nf6, I was under the impression that 5.c4 would be the main move. It was well represented, but absolutely swamped, by a 3503-272 margin, by 5.Ne5. What?!



I'm far from being aware of all the sidelines in every opening - it's unlikely that even Kasparov and Anand know every sideline in every opening. But to be unaware of such a popular move in a reasonably common sideline? It took me a few seconds, but then I realized what was going on. ChessBase is handling the opening book in a fashion that's helpful in one way but misleading in two others. The first way it's misleading is that 5.Ne5 isn't ever played there; rather, 5.Ne5 transposes to a known position. (A slight exaggeration: 5.Ne5 shows up a whopping 22 times out of 4797 games in ChessBase's online database.) But that too is misleading. It's not that 5.Ne5 transposes to a position that White normally reaches, but one that occurs from the opposite side of the board:

1.e4 e5 2.Nf3 Nf6 3.Nxe5 d6 4.Nf3 Nxe4 5.d4 d5



The Petroff is a perfectly viable opening, but it's still better to have White there, not Black. To be fair, even though ChessBase doesn't explain that, one could still gather from the scoring percentages that White should avoid 5.Ne5 in the Exchange French. Still, users might wonder why the move is so popular, and think that it must be good, despite its scoring percentage, if its frequency so overwhelmed that of other moves. I'm not sure if this sort of error/misleading info occurs often enough to be a problem, but if it is, perhaps ChessBase ought to consider adding one or two notations in such cases: one indicating that it arises with colors reversed (at least a certain portion of the time), and a second indicating that it's a transpositional move.

Two comments, by way of tidying up. First, some of you might see the first position and think 5.Ne5 looks good, and then be led to wonder how Black could possibly stand worse (or at least have the burden of proving equality) in the Petroff. The answer is that the knight's advanced location isn't an unmitigated blessing. The opponent (let's assume it's a Petroff, so that would be White) can try to show that it's overextended, playing moves like Bd3, Re1 and c4. Sometimes White will also continue with Nc3 or Nbd2, and if Black captures, as he usually does, the knight will have left its fine post and exchanged itself off with a loss of time.

The second point returns to this post's title. A few weeks ago I (following Alex Baburin in Chess Today - it's his fault!) presented a similar "backwards" position, so it's only fair that I should be similarly afflicted. And then before that there was some April Fool's business...

One of the approximately 30,000 events this August finished a few days ago, the Mercosur Cup in Villa Martelli, Argentina. It wasn't a super-GM event, but was a respectable Category VII tournament with an average rating of 2410. The winner? Pocket Fritz 3, scoring 8-2 (six wins, four draws) and achieving a 2690 TPR. I hope tournament directors of open Swisses are taking note.

This was the score in an odds match played this past Monday. GM and anti-computer expert Roman Dzindzichashvili started every game with the white pieces and Black's f-pawn, but even so he was unable to defeat Rybka 3.0 in the match or even a single game. Rybka won game 2 and drew the rest, though "Dzindzi" had an overwhelming advantage in game 3. These computer are awfully tough!

There's another article about Rybka on the ChessBase site promoing its features, and it looks really good, too. Shritzarcs is going to have its work cut out for it if it hopes to compete with Rybka.

The Rybka-Dzindzi match games can be replayed here.

Frankly, this is a good thing, and it makes sense for both sides. Rybka has been dominating computer chess the past couple of years, while ChessBase has the biggest name and distribution network. I'd pass along a link, but something seems wrong with the ChessBase site for the moment. Anyway, Rybka 3.0 (a significantly new version) is scheduled to get mailed out August 1, and they're taking pre-orders for it now.

Update:

(1) The link is good now - here it is.

(2) The ad is worth checking out, as the new Rybka has some interesting bells and whistles that make it interesting for reasons other than its strength as a playing/analysis partner.

I was browsing Bobby Fischer’s My 60 Memorable Games earlier tonight, and came across this position and comment, from his 1960 game against former world chess champion Max Euwe:



Here Fischer played 15.Rb1, writing “The innovation. Months before this game I had showed this line to Benko and he suggested this innocent-looking move. Upon looking deeper I found that, horrible as White’s Pawn structure may be, Black can’t exploit it because he’ll be unable to develop his K-side normally. It’s the little quirks like this that could make life difficult for a chess machine. (P. 135, emphasis added.)

It was this last comment that intrigued me. Did Fischer insightfully predict a problem for computer chess, or was this a perhaps understandable but mistaken assessment of what would or wouldn’t be possible for chess engines? I think there are two aspects to evaluating Fischer’s claim. First, do the engines find 15.Rb1? Second, do they evaluate the situation properly? On the first question, Fischer’s conjecture is a failure. Both Rybka (2.3.2) and Fritz (11) found it instantly. Rybka fluctuated for a little while between that move and 15.c4, in terms of absolute preference, but it settled on Fischer’s move soon enough. What about the evaluation? Euwe played 15…Rd8, which Fischer awards a question mark. Fischer recommends 15…Qxb5 instead, asserting that White has “an enduring pull” after 16.Rxb5 Kd6! 17.Rb7 f6 18.Ke2 Kc6 19.Rf7 a5 20.Be3. How do the computers fare on this score?

Here, perhaps, there's a little more difference. The engines both agree with Fischer that White has some pull, but neither seems all that impressed - both evaluate the position as (much) closer to equality than anything substantial for White. (Details here.) So maybe there's a bit of difference when it comes to evaluating long-term factors. Fischer sees that White can torture Black for a long time in the ending, while the computer thinks that as long as everything is pretty safe at the moment, Black is fundamentally okay. Of course, another possibility is that the computer is right, but it can be said that engines do sometimes tend to underestimate long-term possibilities. On the other hand, it's not at all surprising that the computer liked 15.Rb1, as it brings a piece while impeding Black's development. Why wouldn't it be attractive?

The odds match between Rybka and Dzindzichashvili ended in a 4-4 tie, and the plans to have a 4-game playoff today were cancelled, as Dzindzi had a family emergency. Hopefully all's well there, and maybe we'll see a rematch in the near future.

On day 4 of the Rybka-Dzindzichashvili odds match (Dzindzi gets White in every game plus a different Black pawn in every game), Dzindzi caught back up, winning game 7 (with the extra g-pawn) and drawing game 8. It wasn't part of the original deal, but they're going to have a 4-game playoff today (Saturday). Rybka will play without the f-pawn in every game, which is probably the biggest handicap pawn, but the time control will be a relatively speedy 20' + 5", which is surely in its favor.

Meanwhile, today's games can be replayed here.

Carbon made a small comeback today in the odds match between GM Roman Dzindzichashvili and Rybka, as both games were drawn against the silicon beast. The event finishes tomorrow, with Dzindzi first receiving Black's g-pawn in game 7 and the h-pawn in game 8. Meanwhile, you can replay today's draws here.

Monday was a good day for Roman Dzindzichashvili in his pawn-odds match with the computer program Rybka, but Wednesday, the second day of their match, was not. Giving up the c- and d-pawns, respectively, Rybka won both games against the American GM to take the lead in the eight game match.

Games here. (Note: the funny opening moves designed to set up the proper pawn-odds position are given on the site; they're not an expedient on my part to create the proper start position.)

All those years of playing computers have paid off for GM Roman Dzindzichashvili so far, as the first day against Rybka with pawn and move has given him a draw and a win. It would be nicer if he was ahead without any odds, but I guess we'll take what we can get at this point.

Games here.

Rybka, probably the strongest commercially available program out there at the moment, has been beating up on humans the last year, handily defeating GMs while offering various sorts of odds. (More about that here.) Starting tomorrow (Monday), it's time for a new challenge: an eight-game match with Roman Dzindzichashvili, once one of the world's strongest grandmasters and a real anti-computer expert. "Dzindzi" will have White and an extra pawn in every game (Rybka will play without the a-pawn in one game, the b-pawn in another, etc.) The time control will be G/45 + 10 seconds, with two games a day on Monday, Wednesday, Thursday and Friday. More info here.

While waiting for the first round of Morelia to begin, I took an occasional glance at a very long series of games between Hikaru Nakamura and an engine called "TransWarp" on ICC. As far as I know, Nakamura didn't win any games (though he should have won one, but ran out of time while trying to mate with five or six knights), but he managed a pretty fair number of draws by playing a Hippopotamus + blockade strategy with both colors.

I've linked to one of them for your entertainment and anti-computer instruction. While many of the games were pretty similar, this was perhaps the most impressive of the lot. That's because he didn't achieve a particularly good version of the Hippo this time around, and was already in serious trouble on move 16. His solution (see especially move 17) was an ingenious one, and he was just as clever on move 86. You'll get the idea - have a look.