Wednesday, July 04, 2018

Deep understanding

My mind is easily overwhelmed. When my mind is overwhelmed, it stalls.

That is the feeling I have when I play chess.

The reason that my mind is overwhelmed, is that it can not handle complexity. Probably no mind can. The only way to prevent the mind from being overwhelmed, is to get rid of complexity.

There are quite a few areas in life where I managed to get rid of complexity. I studied the book Progress and poverty of Henry George for about six years. In the beginning it was impossibly to apply his ideas to modern economic events. But when it finally dawned on me, I could see through complicated economic events in seconds. Because the events where no longer complex.

One of the characteristics of complexity is that you see things as different while they actually belong to the same category.

The mind is overwhelmed by numbers (vast amounts). When you see that things are different that are actually the same, you add to the numbers. By adding numbers, you add complexity. The conscious mind can handle only a few numbers at the same time. That is why it is overwhelmed so easily.

When I started this blog, it had the subtitle "chess improvement by effort (hatsjoe)". It was a joke, but since nobody knew the dutch word "hatsjoe", nobody recognized the joke. Only when I explained that the dutch hatsjoe was the same as achoo in English, the allergy to effort became apparent.

I learned six languages at school, and to me all languages were essentially different dialects of the same language.

Once I had to translate a brochure from Swedish. I never had been in contact with the language before. But by pronouncing it as old English with a German accent, it became suddenly clear what the words meant. All of a sudden I recognized about 80% of the words.

Only when you think that all languages are different, you  manage to introduce complexity.

groot, groß, grand, grande, great, gran, grut, gwo, are all the same word in different languages. You add to the amount of numbers by seeing them as different.

We talked a lot about the subconscious working miracles. I always wondered in what area these miracles would be when it comes to chess. When complexity disappears, miracles are being worked. To recognize that groot, groß, grand, grande, great, gran, grut and gwo are all the same, is such miracle from the subconscious. But you must first realize that the words are actually the same.

I have gone at great length in the past  in understanding chess problems. But what I failed to do, is to consolidate that knowledge. I analyzed, but I didn't systemize the knowledge that emerged from analysis. The analysis rendered the problem as simple, in the end. But I didn't simplify the analysis.

I have wandered every path of chess improvement under the sun. I have fallen in every pitfall one can imagine. I even didn't miss a single pitfall you cannot imagine, the past 18 years. Which makes matters very simple nowadays. Only one thing I haven't tried. And that is what I'm up to now. Deep understanding. Even so deep, that matters become simple. Time to put in some effort again (hatsjoe).


  1. Robert said:

    I'm looking forward to your explanation of "deep understanding" AND how to achieve it in chess.

    I was recently reading (and rereading) Nassim Nicholas Taleb's excellent book Skin in the Game - Hidden Asymmetries in Daily Life. As you may know, he is famous for The Black Swan.

    At the beginning of Chapter 2, pg 78, he writes (my emphasis added):

    "The main idea behind complex systems is that the ensemble behaves in ways not predicted by its components. The interactions matter more than the nature of the units. Studying individual ants will almost never give us a clear indication of how the ant colony operates. For that, one needs to understand and ant colony as an ant colony, no less, no more, NOT a collection of ants. This is called an "emergent" property of the whole, by which parts and whole differ because WHAT MATTERS ARE THE INTERACTIONS BETWEEN SUCH PARTS. And interactions can obey very simple rules."

    That somehow brought to my mind this excerpt from Lasker's Manual of Chess (pp118-119):

    "The motifs of a combination, in themselves SIMPLE, are often interwoven with each other. What is it that unites the multiplicity of motifs? We call it the "idea." Motifs, as for instance, a simultaneous attack against several pieces or the encircling of the hostile King, are tricks of the trade, technicalities. The idea which links the motifs is artistic, it creates something that had never before been there. Motifs can be taught, ideas must be discovered by original effort. Ideas come from nowhere, they are sudden inspirations; the place of motifs is definite: the memory."

    While thinking about these ideas, I was struck by how we focus on the "units" as the basis for pattern recognition. Regardless of whether we "look" at at units at the "contact" level (Averbakh) or more typically at the device/theme level (pin, forks, etc.), or even at a higher conceptual level of motifs, we are still down in the "weeds" at the unit level. We work very hard to "recognize" (System 1) "patterns" which constitute these low-level units - and as a result, we fail to "see" the patterns of interaction of the various units, thereby missing the "idea" (the whole) of the potentialities for combination(s) in a specific position. We can logically work at "combining" the various low-level units into something that "works" but we often can't figure out (using System 2) what should be done on this basis - and it eats up an enormous amount of time in the process. This is nothing more than applying the "chunking" hypothesis. The "chunks" that we need to "see" are the INTERACTIONS of the various low-level units in a specific position, not just recognizing the PRESENCE of the low-level units.

    I think this may be an area where we "miss the forest for the trees" with regard to training pattern recognition.

  2. Robert said:

    Here are two examples of the need to "see" the interactions of the "units" rather than just "seeing" the individual units and trying to logically combine them into a "solution." The positions (but not the entire games) were given in GM Valeri Beim's excellent book How to Calculate Chess Tactics - A revealing look at the nuts and bolts of chess thought".


    Black to move after 22. R(f)c2

    FEN: "1r4k1/p1R3pp/4pn2/8/3pP3/5P2/2R3PP/r1B3K1 b - - 4 22"
    Loek van Wely vs Viswanathan Anand (1999)
    Viewable chess game Loek van Wely vs Viswanathan Anand, 1999, with discussion forum and chess analysis features.

    White to play after 30. ... Qxd4

    FEN: "8/2pk2p1/Q1pb4/3p1P2/3q4/1KP1r3/PP1N1RP1/8 w - - 0 31"

  3. PART I:

    RE: the complexity of numbers (and our inability to handle that complexity):

    Let's examine something given in Lasker's Manual of Chess, Second Book, The Theory of The Openings, pp 106-107. I think it accurately illustrates the complexity problem with numbers when applied to chess.

    Fourthly: To fix the exchange value of the Pawns and pieces and the move, in order to decide whether we may sacrifice a Pawn for so many moves gained in development and similar questions, the following table will be found to be a fairly accurate guide:

    The Value

    1 = the first move
    4/5 = the second move
    3/4 = the third move
    2/3 = the fourth move
    1/2 = the fifth move
    2 = King Pawn or Queen Pawn
    1 1/2 = King Bishop Pawn or Queen Bishop Pawn
    5/4 = King Knight Pawn or Queen Knight Pawn
    1/2 = King Rook Pawn or Queen Rook Pawn
    4 1/2 = Knight
    5 = King Bishop
    4 1/2 = Queen Bishop
    7 = King Rook
    6 = Queen Rook
    11 = Queen
    Undefined = King
    [although traditionally, it has been valued at approximately a minor piece plus a Pawn]

    [or possibly statistical analyses; see below for a reference]. The numbers given are obviously only approximations. But even though the position is complicated, they can still be of service in helping to determine whether one may venture upon sacrificing a piece for two Pawns, and in elucidating similar thorny problems.

    Such a table, if it were possible, ought to include the values of dominating squares also, but I am not sure whether the table thus extended would retain its original form, WHETHER IT WOULD NOT NECESSARILY BE MORE COMPLEX, whether it would not have to work with pairs of numbers and groups of squares. THE VIRTUE OF THE ORIGINAL TABLE, HOWEVER, IS JUST ITS LACK OF COMPLICATION. It is better, therefore, to go no farther but to stop right here and use only arithmetic. It is better to breed within yourself a sound judgment concerning the value of the squares. Practice and intelligent criticism will help you to develop it. THE ACQUISITION OF SUCH JUDGMENT IS A HUNDRED TIMES MORE PRECIOUS THAN THE MERE LEARNING OF A TABLE OF VALUES just as in life it is more important to breed than to learn. THIS PRACTICE YOU WILL FIND TO BE SPLENDID DISCIPLINE FOR YOU.

    Given that Dr. Lasker had a PhD in mathematics (hence a deep respect for the power of mathematics), a PhD in philosophy AND was the World Chess Champion for 27 years during the heyday of the chess Titans, we should consider accepting that ANY TABLE OF VALUES IS VERY IMPRECISE and only of use while a beginner. (If you doubt that assertion, try playing a game of chess using any of these tables of values to make decisions that involve material balance or tempi. Yes, it IS a blatant “appeal to authority” - of one who IS an authority on the subject.)

  4. PART II:

    The easiest table of values to remember and use are the so-called Reinfield values:

    1 = Pawn
    3 = Knight
    3 = Bishop
    5 = Rook
    9 = Queen

    I submit that SIMPLE values are more useful for practical play than more COMPLEX schemes of valuations, ceteris paribus.

    An extensive statistical analysis based on a large collection of master games (over 300,000) was conducted by IM (now GM) Larry Kaufman and published on at:


    There's a well-known joke about statisticians. "Did you hear the one about the statistician who drowned trying to wade across the river? He knew it was three feet deep...ON AVERAGE."

    (Given the general level of innumeracy, ignorance of statistical measures and over reliance on Gaussian bell curves, it is virtually guaranteed that chess players in general will make erroneous assumptions about the practical benefit of such statistical analyses. Caveat emptor: I AM NOT A STATISTICIAN!)

    There is a pernicious illusion with regard to statistical precision. It is often ASSUMED that more precise mathematical values imply more precision of the conclusions drawn from them. Statistical information about an AGGREGATE gives nothing but “noise” with regard to any specific individual unit which is part of that aggregate. If the data has been “normalized” (by throwing out apparent anomalies), it is worse than useless: the specific unit in question may have been thrown out of the data set prior to calculating the statistical values! [See IM Kaufman’s method in the article referenced.] Or, in computer terms, “GARBAGE IN, GARBAGE OUT.” (For a much more in-depth and informed discussion of this phenomenon, see the various books, such as The Black Swan and Skin in the Game by Nassim Nicholas Taleb. Refer back to my first comment on this thread above for something applicable to chess complexity.) I submit that taking statistical information out to two decimal places may be useful for computer programs that play chess, and totally worthless for human beings. Contemplate that you have enough problems with the Reinfield values; it is IMPOSSIBLE for any human being to calculate to two decimal places as a means to play a good game of chess.

    This problem with complexity in general and complex systems (like chess) plagues us all.

    And yet, as Temposchlucker has pointed out from his own experience, when one has “deep understanding” across apparently disparate entities, (somehow) the mind can “see” the commonalities and therefore the system can become “simple” (in some relevant sense).

    Which seems to put us right back where we always start AND end up:


    I think there is a broad “hint” in Temposchlucker’s post: EFFORT, properly directed.

    It is NOT going to be through "solving" hundreds (or hundreds of thousands) of tactical problems on Chess Tempo. Been there, done that, and it did NOT work.

    He who has “eyes” to “see,” let him “see” – the commonalities, the analogies, the “common threads” which bind the disparate units into the whole system, thereby reducing the complexity of the whole system.

    What?!? You thought it was going to be EFFORTLESS?!? HAHAHAHA!

  5. I am watching the improvement of a 15 year young man. He improved 200 elopoints the last 6 month almost exclusivly by playing blitz and bulletgames and reading (not studying!) some chessbooks. I did try to convince him to do some tacticpuzzles but he dont like to spend more than a few sec for a move. He is now rated near 1700

    By the way there is a replacement for the cts now :

  6. I dont know what deep understanding is. But the example with the word "great" and recognising similarities - that is somehow familiar. I looked into openings of the chesstempo-database.
    While I was searching for moves that are statistically more promising, I recognised some similarities of statistically "better" moves. It then generally helped me to find better moves in the opening stage, even if I am out of book.

    2 things easy to explain are: Symmetry favours white, while black better seeks asymmetry.
    Example? The highly asymmetry O'Kelly Sicilian 1.e4 c5 2.Nf3 a6 is much better for black that for instance the very symmetric Petrov Defense (1.e4 e5 2.Nf3 Nf6).
    And it doesnt stop there, once I discovered this rule, I saw it everywhere in the openings.
    worst black answer to 1.Nf3 is (according to the rule) 1...Nf6 (commonly played) - surprised? Not if you follow the rule.

    But is that deep understanding? Nah, I'd say it is more about recognising that grande and great are somehow similar. Of course it may not always work, see: to see, sehen, regarder, mirar, widywac - not always so similar, is it.

  7. deep knowledge: I never really could understand why my Blitz and Bullet ratings are always much worse that my rapid and longer time ratings.
    Aox said, I am relative weak in tactics, but in longer games I have a high "K" factor. a high K factor - that means I can make more out of thinking time than others. Like a chess engine gains around 50 elo in strength if it has doubles its thinking time. Most humans seem to have 100 elo per doubling of thinking time, whereas I have 200.

    However, we really dont know how to train a K factor, nor am I sure it can explain why more time for me means much more than for the average player.
    The other hypothesis is: I have "deep knowledge" - again, it is not clear what that is.
    I believe I know more. You say it makes chess more complex. Yes, it probably does. But more time will allow me to think of more rules, more angles to look at, more ideas due to better knowledge.

    And I do have some rules where I am sure I use them if I have enough time.
    Example: In a rook endgame the stronger side who is 2 pawns up can win if he has the rook pawn (h/a -file) and a knight or centre pawn (pawns g/b, e/d). The weaker side, however, has drawing chances if white has rook-pawn and a bishop-pawn (f/c) - because of stalemate possibilities.

    And I think early about such a rule, and my thoughts circle about how to make such an endgame happen.
    But in Blitz I just do not have the time to think about something like that. Instead, it is highly likely I dont have much time left. Then I only think: how to create a passed pawn, and calculate promotion tactics (if my pawn runs - does it get lost or can he block it?)

    Other deep knowledge examples: "how are my chances for an attack? How many pieces can I realistically get there before my centre crumbles. Which piece has no job to do, or can one of my pieces take over the job of one of my pieces, so it is free to go?"
    And when I freed a piece of mine from a job, I think hard what its future job could be. Where are my opponents weaknesses, and do I have a realist chance to ever get there?

    Such thoughts are absent in a blitz game. But I do have them in longer games.
    Thus I just about an A-class player in Blitz, but are at the brink of master level in longer games.
    My lichess classic rating is better than 99.9% of all players. My rapid is 99.6% better than all players, but my super fast ultra bullet rating is just 45.5% (!) better than that of all players.
    Well, I guess the rating pools are tougher at ultra bullet, but still, not even being above every 2nd player is not what you would expect if you see my classic rating, which places me about place 20 of 32000 players.
    Online ratings dont count? Well my english ecf rating made me place 368 of 9000++ active players 3 years ago, but I am not active anymore.
    So my anomaly is proven. I do have either deep knowledge or a high K-factor. I believe it is deep knowledge. Relative knowledge though.
    Uri Blass said, he is amazed about how accurate my moves are if I have sufficient time. So my fine evaluation is high, but then again, why is it higher than others? I'd say again: deep knowledge. Knowing relevant rules - and make use of them during my games.

    (see link to my lichess profile for ratings across various time controls and chess variants)

  8. Another thought:

    Many people learn a language at school. But not all (if not most of them) are not becomming fluent, even after having a foreign language for many week, months, years.
    When they see a text, they can understand what it is all about, but they can not watch TV or talk with people in this learned language.
    Surely, in school they had exams, tests, homework - and they were able to do all this.

    I believe that you need to learn the vocabulary inside out. You need a lot of repetition, and probably it is impossible to learn a foreign language without visiting the foreign country and try to engage in active conversation in this foreign language.

    And translated to chess?
    Hard to say. Just learning theory by heart is not enough. I guess you would need to play against a friend what you have learnt. If you learnt about rook endgames, you would need to play it several times against your friend and do spaced repetition with him for some weeks? Then you would be able to know the theory so well that you will still remember it even if you see it only in a blitz game 3 months later.

    I'd say it is also very important that you learn relevant ideas in chess. Like: the best defender you could wish for (in case your king is on the run) is: the knight.
    It is bloody tough to win against your naked king if there is a knight around, and maybe here and there some obstacle, which block the remaining checks that the knight does not cover.
    It is an issue I frequently consider when I am thinking of sacrificing a piece in order to rip up his casteling position (the 3 pawns in front of him) so I get a strong attack. When there are two knights around, I wont sacrifice - unless I see a clear material win. A tactical combination. But if I only see: "I have then a strong attack and he has 2 knights nearby" - I forget about the sac-idea.
    This simple rule serves well in 90% of all games. You really have hardly a good attack if your opponent has 2 knights around. Often, one knight is enough to spoil all your attacking hopes.
    That is why most exchange-sacrifice in chesstempo are probably "a rook for a knight" instead of "a rook for a bishop". Plus knowing this rule usually helps to save time (on average) and find the chesstempo tactic faster: always try first if getting rid of his knight could be the winning combination?

    Anyway, deep knowledge - knowing it very well is maybe still not enough. You can have very good marks in school in the foreign language and still you dont talk the foreign language. And probably only - ONLY - traveling to the foreign country and try to get engage in active talking can work the miracle that you start to speak the foreign language. Once you reached the stage where you can have conversations well enough to have friends (not just asking for the way) - only then you might be able to improve further with more study of the language (with a book at home), without engaging into active talks. When you reached the B1-level (see "Common European framework of Reference for a language"), only then you can become even better without further exposure, but just with learning at home. Simply add more vocabulary will then work wonders. Dont know if "simply" is exaggerated here, though.

  9. Tempo

    Are you going to list all (or most) elements you want to test and check out... while trying to learn "deep understanding" (in your perspective). If you can write a bit I will be very grateful. I am really interested at "the last method (way to improvement) I have not tried yet".

    Thanks in advance!

  10. Sorry, not to the subject, but maybe a thought for your next post?

    There is this known fact, that the difference of the ratings in long games and blitz games is not hugely different among grand masters.

    However, there are some exceptions to that rule.
    At lichess, there had been a bullet-tournament, where Magnus Carlsen took part, too. Even though Carlsen won previous such tournaments, the most recent winner wasnt Carlsen but GM Andrew Tang. He is a 2500++ player.
    Here his profile.

    Or GM Bogdanovic: currently rated 2525 Fide elo, has an official Fide Blitz rating of 2697. But his best Blitz rating ever was 2815 Fide elo, placing him among the best players in the world.

    So there are some players out there, who are very good in fast stuff, and not so good (relatively seen) in longer time controlls.
    If there are such players on one side of the gauss-bell-curve distribution, there must be for sure players on the other side of the bell-distribution curve? Players like me, who are good in long time controlls, but much weaker in faster games. Here my profile:

    I have not seen any player similar to me, though. But I believe they exist, and probably among grandmasters, too?
    My ultra-bullet rank places me at "better than 47% of all players".
    My bullet rank places me at "better than 70%"
    My blitz rank places me at "better than 95%"
    My rapid rank places me at "better than 99.6%"
    My Classic rank places me at "better than 99.9%"

    I played many games against "our" Tomasz, who writes here at temposchlucker blog, too.
    Result: He beats me easily in bullet, but I faire much better in Blitz. We have not played any really long games, but I suppose I would win most of the games in such a case?

    Aox said my K-factor is very high. K means, that for instance if the thinking time is doubled, the player gains on average 100 elo rating points. But this is individually different. My K is rather a 200 elo jump per doubling the time, and some other players had only 50 elo jumps for doubling the thinking time.

    I have no idea, who K can be trained, and I guess, the trouble is that we dont really know what is the cause (the driver) of a high K.
    But to make an educated guess, it is like that: Blitz games and shorter games are more tactically driven. And I am weak in tactics.
    But in longer time controlls, fine evaluation is more important, and I can move very accurately, finding more often than others the best best computer move.
    Why is my fine evaluation better? I can only guess again: I know a lot. I have many, many little small rules, and probably think about them consciously, but also (most often I suppose) unsciously. My "Filter-System" for deciding what to move must be quite complex. I did "guess the move" games at chesstempo, and did very well there. Again, maybe not so well in deep tactics, but rather in positional games.