Monday, October 12, 2009

Magnus Carlsen is not that good. Is he?

For a long time now, the internet coronation of Magnus Carlsen as the new King of Chess has been annoying me. Whether it's kibitzers who declare him to be Karpov + Kasparov + Rybka, his mystical anointment into the К-Club, or instant annotations that presume flawless play by journalists who should know better, I've just not been buying it. Carlsen is, of course, very talented, already one of the world's best players, and looks to remain so for a reasonable length of time. But such deification as a chess god? Already? Really?

And it's not only knee-jerk doubt. I've been telling myself, there are logical reasons to be suspicious that Carlsen is overrated (in the historical sense). The first is that, psychologically, we chess fans are used to living in the era of a Super Champ: that is, a World Champion who topped the rating list, was expected to win every tournament they entered, and retained their crown for a more than just a handful of years. Both Karpov and Kasparov were Super Champs, and so it's natural to expect that a new Super Champ will come along sooner rather than later. Natural, but forgetful too of chess history. There have been many eras of chess without a Super Champ, after all. The most conspicuous was the late fifties and 1960s, which saw no fewer than five World Champions.

And why shouldn't top-level professional chess nowadays be similar to that era? Topalov tops the rating list and his combative style, quick recovery from defeats, and excellent preparation make him a likely front-runner in any tournament he enters. Meanwhile, there is no question that Kramnik and Anand both deserved their World Champion match victories. And plausible future rivals aren't hard to name: along with Carlsen, there's Aronian, Radjabov and Karjakin at the very least, not even to begin to name the Grandmaster children scaring me even further away from competitive chess. Add to that the ever-increasing role of affordable computers in the production of novelties - well, this era could surely produce a new, legitimate World Champion every other year, without a single Super Champ to be found.

And then there are the institutional arrangements, or rather their lack. No stable, revered World Championship to aim at; no gruelling, comprehensive qualifying structure. The consequential cultural devaluation of the ultimate crown. Qualifying for the World Championship once represented climbing Mount Everest, the ascension to the highest peak. But nowadays, who's to say that FIDE just won't parachute its latest favourite to the top for the next match? Everest is being worn down into just another modest chess hillock, the crown losing its shine, shedding its jewels. Why should chess professionals such as Carlsen not just content themselves with a pleasant, predictable life lived along a circuit of plush hotels, shrugging off the world-conquering dreams of their greater predecessors in favour of nights in watching Monty Python?

And so I had been telling myself. But that was before Kasparov revealed himself as Carlsen's second. Before Carlsen obliterated the world-class field in the Nanjing Pearl Spring tournament. Before Carlsen started talking up his intentions. Perhaps the uncrowned Super Champ he is after all. On the other hand, Karpov was a Super Champ for a whole decade, Kasparov for a full fifteen years: if Carlsen is to measure up to the hype, I at least have the consolation that the earliest I will be proven comprehensively wrong is 2019. One thing for sure is that there'll be a lot of interesting chess between now and then.

22 comments:

Ryan said...

I've tried to resist lavishing superlatives on Magnus, but his Nanjing performance certainly deserves high praise.

I think it's easy to forget just how young he still is. Given that he's only 18, it's remarkable how mature he seems it his attitude and his play.

My guess is that the next 2-4 years will be crucial for him, if he is to become a 'Super Champ'.

Jonathan B said...

Does anybody know how Carlsen's tournament compares to others in the past - e.g. Topalov at San Luis.

I suppose one thing about Carlsen and his longevity is that it's rare for a victorious challenger to be older than the champ - so if he does become Champion then there's likely to be a bit of a gap for the next one down to come through.

Being young gives Carlsen the potential to be the best for years for sure. On the other hand being so young also means he hasn't had time for things to go wrong yet ... or become bored of beating his seniors over and over ... or for somebody to come up with a machine that solves chess...

Tom Chivers said...

Geoff Sonas performed the kind of analysis you're asking about Jonathan in a chessbase article. It's the best ever tournament performance by a teenager, and something like the 15th best tournament performance ever. It's also the best performance ever by someone who has never been a World Champion.

Anonymous said...

I'll think I'll kindly ask Geoff Sonas not to analyse where my London League performances for Streatham over the last 5 years or so rank me. Even today's powerful computers probably don't have enough zeros!
Andrew

Chris Morgan said...

If Carlsen does challenge for the World title soon, perhaps it will get into the mainstream news because of his age and be good for the popularity of chess.

Morgan Daniels said...

True, Chris. What's exciting is that if Carlsen challenges for the world championship soon it may well be in London, what with Pein & co's plans for 2012.

Jonathan B said...

If seen Sonas' list of the greatest tournament performances of all time (prior to Carlsen's recent Chinese effort) somewhere or other. Some of them seemed a little doubtful to me e.g. if memory serves he rated Karpov in his first match v Kasparov as one of the best perforances ever.

Robinson said...

It will be interesting to see if Magnus is able to repeat this kind of Nanjing performance throughout the coming season of tournaments. How many TNs are there in Kasparov's laptop? (And are they really the edge that has pushed Carlsen upwards of late?)

One top player who was not in Nanjing, but who has been very hot in 2009, is Aronian. I hope there is a Carlsen/Aronian World Championship Match in a couple of more cycles, followed by a Nakamura run (for the same reasons as Chris Morgan).

Because of the computer databases, the energy required to stay on top, and the parity of the top players, I doubt that we'll see the supreme domination by any one player for any length of time. That Topalov has remained at the top of the ratings list so long is a surprise to me, but is his actual playing strength really superior to anyone in the top 5?

With respect for SonofPearl, I don't think the next 2-4 years matter so much -- Carlsen will surely shine during this time. He has obviously committed to playing chess professionally and it seems he's found someone who understands play at the very top as a coach. What matters, to become a Super Champ -- someone who stays at the top for a decade -- is how he reacts when he hits a stall point, for instance: he decides Kasparov has nothing more to teach him, or he becomes bored of chess (Fischer syndrome), or becomes interested in other things (starting a family), etc. The real test isn't the next 2-4 years, but 5 years from now, when he is at the top and asks "Is this all there is?"

Tom Chivers said...

There apparently are 16,000 TNs on Kasparov's laptop, Robinson, at least according to Dylan McClain of the New York Times.

Jonathan B said...

That 16,000 figure is bullshit ... I mean I'm sure Gazza still got quite a few moves up his sleeve but who would keep a precise count?

Tom Chivers said...

Well, judging from his comment Dylan seems confident, and he is clearly a strong amateur player. Indeed I believe that 2265 USCF is over 200 ECF.

ejh said...

Is it?

Robinson said...

That 16,000 number is nonsense. What McClain says is that he saw that there were 16,000 distinct files on a thumbdrive -- that hardly means that there would be a good, usable TN in each file.

On the other hand: You could probably farm potential TNs pretty easily. Chessbase has a function to find novelties that were already played (not unplayed ones) -- comparing repertoire games to other games. So, you could set up some chess engines to play a tournament in a given repertoire -- the repertoire of some future opponent -- and then have Chessbase go looking for the novelties from the set of engine games compared to games played OTB by your opponent or other GMs. Or perhaps someone has created a Chessbase 'TN farming module' that runs some engines against a repertoire database to look for reasonable moves that were never played (in the mega database, against your opponent or in games that would be familiar to him). Then you just have to learn the new lines, get your opponent into the new (to him) position and gain some time on the clock if not a winning position. If it's really this easy, I'm sure all the high-level professional GMs are doing it and they can all generate TNs faster than they could learn them. So, it may not be the number of TNs that are important, but the quality and the method of generating quality TNs that is more important.

So instead of asking 'how many TNs are there on Kasparov's laptop?' (which was a rhetorical question, by the way), I should have asked -- do these TNs really mean anything? Aren't these things any GM has the tools to find?

Maybe it's Carlsen's ability to memorize these TNs that is his real advantage: See http://www.chessbase.com/newsdetail.asp?newsid=5044.

I also wonder to what degree there was a bit of a 'psych out' going on in Nanjing. For example, Carlsen played something different against Jakovenko in the opening, but it wasn't anything unusual. However, Jako wasted a great deal of clock time making a rather normal reply. I think he might have been looking for a specter around the corner that wasn't really there, instead of just playing chess. At least from Nick de Fermian's analysis of the game, Carlsen didn't even play best moves at one point in the middlegame and Jako could have secured a draw with better play. But Carlsen's current mystique (and extra time on the clock) just seemed to carry the day and the game seemed to win itself. I think this may have been a typical example of what went on in Nanjing.

I don't really question McClain's playing strength, but his answer that there are 16,000 files kind of skirts the question "are there 16,000 TNs?" Maybe there are (maybe the files were in a folder named "TNs"), I just wish he'd been more clear/direct in his answer.

Jonathan -- someone who is anal enough to sort through and remember books of opening theory and become the strongest chess player of the 20th century is the kind of person who would know precisely, or be able to easily look at his computer and see exactly how many TNs he's collected. I'm sure these super GMs are hyper-organized when it comes to their computer files. So I just don't think your comment is any kind of argument against this number.

Also, I'm surprised nobody has commented on Carlsen's blindingly red silk wardrobe (From the photos I saw, everyone else chose much more subdued colors). Did he wear that red thing every day of the tournament? That may have been the real magic that pushed him over the top -- just as Tiger Woods' Sunday red and black so often works for him.

Jonathan B said...

Preparing TNs isn't being anal - it's being prepared.

Knowing you have 16,000 - not 15,000 or 17,000 but 16,000 - *is* anal.

I'm not disputing it's possible to find out, I'm just not convinced a player would bother. It is, however, exactly the sort of thing a journalist would like to write.

As for the psych out stuff - I'm sure you're correct. In fact I think Carlsen has said exactly this himself.

Tom Chivers said...

Your idea of a TN Farm is very interesting Robinson. Would it not though result in a large number of novelties *without* theoretical value, and so a very strong player would still be needed to eyeball the far fewer ideas that might actually promise a move theoretically significant?

Robinson said...

@ Tom:
At first I thought such TN farming would end up with potential TNs that would still need evaluation by the GM (I even wrote my original post that way). But I re-thought that -- with engines playing at about 3000 level or above, human checking may not be absolutely necessary. There's some things that can be put to work here: Have a group of strong engines play out the farmed TNs and use Chessbase to prepare an opening report on those games so the player can learn the ideas behind the new positions. These engines seem so strong now, that if given enough time to run they seem to find anything that would be found OTB. The idea isn't that you (or your computer) have to find great moves -- you just have to find moves that won't transpose easily back into known positions, create new problems for your opponent at the board, and give you even or better chances. Computers should be able to be put to work to evaluate these relatively few conditions, such as:
Does the play lose? Not a TN.
Does it lead to the same position in the database (transpose back into repertoire)? Not a TN.
After the potential TN, is there one candidate move that is so much stronger than any of the other candidate moves or is there a near-future move that is greatly stronger than those in the alternative lines, so as to greatly limit the number of candidates my opponent will consider? Not a TN.
Are there only a few candidate moves/lines that lead to my advantage? Not a TN.
Etc.

Another idea: It could also be interesting to see how much these engines can be made to act like particular players -- loaded up with a repertoire of their games and decision-making parameters. You could have a Kramnik-replicating engine (K2) that plays like Kramnik. Now when you're doing your TN farming to prepare for human Kramnik you start filtering for lines that win against K2.

You could also have "anti" engines. Here's the engine that always seems to find the moves that beat K2, I call it Anti-K.

I don't know if this is the kind of thing that is going on, but if it is, I would think Kasparov, who has been into using computer databases longer than anyone, would be the first one to do it. I read somewhere, perhaps on the Chessbase site, a line in an article that suggested that the super GMs have something more to their Chessbase programs (in the way of software or modules) than the rest of us and that started thinking about this.

I'd be interested to hear anyone's opinions (or actual knowledge) or find sources for what these things might be or how very strong players go about organizing their information and study. (This is out of shear curiosity, as my club opponents don't often publish their games, and I don't spend enough time even looking over my past games with them).

Tom Chivers said...

You could test your method by removing genuine big TNs from your database, and then seeing if engines find them. Topalov's 12.Nxf7 say, or Kasparov's 14.Bc2. In fact Topalov's 12.Nxf7 is an interesting case, because it's not clear it's objectively the best move in the position, just a very hard one for a human to face.

Robinson said...

I wish I had the programming/database smarts of my brother -- he would someone who could really set this up, but has no interest in chess.

That's exactly the kind of TN I am talking about (let's call these type b TNs) -- one that may not be objectively the best move -- just one that is difficult for an opponent to deal with OTB because of its many options and uniqueness.

My reasoning is: because the criteria to find these type b TNs can be logically defined, computers can be made to find them.

Objectively best moves that haven't been played? (type a TNs) Computers should very easily be able to find these. The problem is, these moves probably don't exist anywhere near the opening.

Another thing to farm for would be moves that lead to lines that have been dismissed
because they lead to inferior positions -- but the computer is able to find improvements that lead to equality and enough options to lead to surprises. Let's call these type c TNs. These might be a little trickier to cultivate, but could lead to the reopening of some lines that were played out or discounted. Dzindi and Alburt advocated a Sicilian line that allows the Maroczy bind, claiming that it wasn't all that bad for black. If true, this is the kind of re-consideration I'm talking about.

Sorry to drift so far from the original post, but I find this fascinating to think about. Maybe you need to post something on this subject?

Jonathan B said...

In fact Topalov's 12.Nxf7 is an interesting case, because it's not clear it's objectively the best move in the position, just a very hard one for a human to face.

That's the point for most TNs isn't it? Giving your opponent something new to face otb that you've had the chance to research deeply (or at least imply that you've done so)

Tom Chivers said...

No need to apologise Robsinson, I'm enjoying your comments & thinking about the topic. I just wish I had some time to even open my database up...

Robinson said...

I see on liverating.org that Carlsen has actually added nearly 29 points to his rating to go just above that magical 2800 (I'd realized he'd moved into second on the ratings list, I think even before his win in the last round of Nanjing, but not that he'd added so many points). Even though 2800 might not be what it was (ratings inflation) when Kasparov broke it, this is still a very impressive landmark.

Of course that opens up a whole 'nother can of worms: at the very top levels, hasn't the quality of chess risen in the last 20 years, so that this whole idea of ratings inflation at the super elite level a little silly? In 1982, Kasparov hit 2799 (and No. 1 ranking) at age 19 years 5 months. (http://db.chessmetrics.com/CM2/PeakList.asp?Params=)
Carlsen has hit 2800, No. 2 ranking, and turns 19 next month. He's just 8 points shy of No. 1. Topalov. I don't know how you could even match up 1982 Kasparov with 2009 Carlsen (do they get to use 1982 or 2009 databases and technology, for instance?)

I don't intend to start a storm about who is better, I'm just wondering why there is any stink at all about ratings inflation? Ratings are merely meant to be a tool to measure the relative strength of current players, for seeding in tournaments and be predictive of outcomes of games between them. Until we invent a time machine, we'll never have to worry about 1982 Kasparov playing 2009 Carlsen, and far more than 99.999% of people on earth are in no good position to even have an informed opinion about their relative chess strength.

I'll just touch on one more thing about this ratings inflation "problem." Yes, you can mathematically prove that the ratings system itself will cause an inflation of ratings over time. But, knowledge/skills do usually also increase over time (especially if they are based on information which isn't lost), because of our own experience and because we stand of the shoulders of giants who have come before us. So an inflating system could be a more accurate tool for comparing playing strengths from players of different periods than a non-inflating system. I don't think the fact that we have rating inflation is the real problem, so much as the lack of a way to determine at what rate the system should inflate.

Tom Chivers said...

By curious coincidence Robinson, I am currently writing a blog post about rating inflation, with a proposed solution to the problem. It should be published a week on Monday...