Monday, November 28, 2011

Lousy logical reasoning and familiar patterns

Munich said:

An idea:
There are two sorts of difficult puzzles:
First, the ones with lots of calculation, and deep variations.
Second, the ones which could have been easy if you simply had spoted it.
The 2708 CT Blitz rated (your example) is a puzzle of the second category: the relativelz easy ones that for some strange reason almost nobody gets correct.

I would love to have some puzzles sorted into these two categories.
I could imagine it is worth to study the character of the "easy" puzzles, that are very difficult.

90% of the 2200 - 2300 rated puzzles at CT fall in the category "easy" imo.
This weekend I worked out about 30 of them while looking what is going on in my mind. I noticed that the patterns are well known enough usually, but I fail to concoct a flawless logical reasoning. That is strange, since I usually score very high in tests of logical reasoning.
This means that the ability is there in principle, but something is lacking.

Look at the following diagram of a 2299 rated problem
You can find it here.

White to move.

The patterns are simple enough, the problem isn't very deep, yet it is easy to go astray due to lousy reasoning. What I found is that there are basic elements of logical reasoning too. But to concoct them yourself on the fly is an error prone business. If you write those elements of logical reasoning down though, you recognize after a few problems that the same kind of reasoning happens over and over again. What I'm trying to say is that it should be possible to automate the uprise of the basic elements of reasoning so that they come to mind without effort. Just like patterns do. In that way you can add intelligence to the solution of a problem and yet make maximum use of the automated semi intelligent elements of reasoning. (Still somebody there?)

The dog above performs a semi intelligent action.


  1. Me Munich:
    I failed the puzzle, too.
    I try to explain my thoughts:

    I have seen the first move in an easy puzzle, too. So for me the correct 1st move was easy to find, and it actually was the only candidate move I could think of for quite some seconds.
    The second move was planned, too. pinning the black bishop. Simple.
    The 3rd move I finally failed, like 97 other people, too.
    Strange, because I actually saw the right move, too. Saw the move Qxg3+ which unpins the black rook. For some reason, I did not count material correctly afterwards and forgot that I dont have just a queen for 2 rooks, but additionally I am a rook up.
    How can I be so stupid?
    I guess I got irritated by the previous variation, where I take with my 2nd move the pinned rook with
    2.Rd1xRd4 BxRe3+
    And material is even. The similarity in this variation and one move later got me. It was simply bad visualization.

    But again, I can show you the simple tactic puzzle that gave me the first move.
    I might find the anti-tactic of black, too, which is an unpinning theme. I tagged most of the tag "unpinning" myself (I tagged most of them during my set "capturing defender", which is not the correct tag to be applied anymore for this kind of tactic), so I know them all. (there are not so many of them).
    My 3rd move was stupid, because the skewer is not difficult to see. I somehow thought I allready found the difficult part of the puzzle and got careless. This time I am very surpized that this was a 2771 CT Blitz rated puzzle.
    You are right, it is not so difficult, and the failure was at the end a bit of lazy thinking, assuming I solved it already. Lousy reasoning. Should not have happened, but happened to a lot of players.

    When I come across my pin set, I will show you the easy puzzle which looks striking similar. It makes me wonder how this one is so high rated while the one I have in mind is so low rated?!

    At the end, I need to admit, that looking at the hard puzzles and trying to explain why I failed the puzzle is helpful for a training, too. It guess this reasoning burns deep into my LTM, too. But then again, it helps to know a similar easy puzzle with the same pattern, too.
    We might exchange this knowledge? Sort me "easy" difficult puzzles. Are really 90% of them "easy"?! I didnt have it like this in my memory, but it is a year ago when I did puzzles around 2100-2300 (measured in standard rating).
    I had the impression that the solution was often long, with a lot quite moves in between. But back then I didn't do lots of easy puzzles, do I didnt have anything to compare upon. I might think different about it today.
    I might start the pin set the day after tomorrow, and I show you the "easy" little brother to this one.
    At least I have the feeling that the "little" brother helped me to almost get the difficult one correct. Many players failed allready move 1, while it was the move I had in mind in the first seconds.

  2. I recognised the pattern of the pin, of course. I remembered the pattern of attacking the pinned piece. But the the 3 moves to do so (Re4, Rd3 and Rd1) looked rather equal. Besides that I recognized the pattern of Rxd1+ with check, which is not good so I decided to avoid any risk by playing 1.Rd3

    So I behaved like the dog in the picture: I knew that I had to jump through a hole so I jumped.

    That is why I call pattern recognition semi intelligent. You remember a lot of related patterns but they are actually the cause that you draw the wrong conclusion. It's hard to believe that learning even MORE patterns will solve this specific problem (of drawing sloppy conclussions).

    The ONLY reason why 90% of the 2200 - 2300 rated problems are rated so high is that EVERYBODY suffers from this phenomenon.

    You can't fit the difference between Re4, Rd3 and Rd1 in a pattern, can you? Somehow we must add some intelligence to the equation.

    What I try to accomplish is to guide my thoughts, somehow. By asking myself questions.

    In the questions and the answers I try to discriminate patterns. Since patterns can be automated.

    A sort of "pattern assisted reasoning". Somehow.

  3. Some nuance:

    2200 - 2300 rated problems:
    3% 1-movers
    15% 2-movers
    24% 3-movers
    32% 4-movers
    10% 5-movers
    12% 6-movers
    2% 7-movers
    2% 8-movers

    The problem from this post wasregistered as a 4-mover.

  4. So I restrict my problemset to 1-4 movers. Have a look at the 2-mover 47215

  5. Today I did not come across your "dog" puzzle from your post. I only came across similar puzzles, but not really that convincing similar. I might find the puzzle I have in mind tomorrow.
    Here my 3 alternative easy puzzles:

    For the 2-mover: I am with the b6-b5 crowd.
    A 2589 CT Blitz rated puzzle.
    (I always take Blitz rating, since it is a more reliable measure of the difficulty of a puzzle).
    Hm. Like everybody else I had the right idea but believing the exchange would be enough. 1...b5 misses the puzzle by only a little margin, though: +1.59 instead of the required +1.75 points.

    Dan Heisman adviced: If you found a good move, put it in your pocket and look for a better one. :-)

    I guess you are right about pattern knowledge that actually can hurt when solving puzzles. And it might be somehow the reason, why these puzzles are so high rated.
    Then again, even though this can happen, it is as you said in one of your previous posts: You can only visualize in a cloud if you know about it.
    The very first hard puzzle you gave - I believe a very good pattern knowledge would have given me the distraction move (Bg3+) instead of just thinking about Rf8+. This is so, because deep in our variations we dont stay thinking at a future position for more than a few seconds. If you dont get it in a few seconds, chances are high you wont find the move in a real game.

    So there are different kind of difficult "easy" puzzles:
    The ones where you did not have the knowledge of a pattern quickly available (= not stored well in you LTM), and the ones where pattern recognition stipulates you.

    I learned a lot of new patterns. But mainly I reinforced the knowledge of patterns I allready had --> I see them quicker now. In a later stage, I will move up the ladder, and learn more difficult patterns, where there are more puzzles mixed with other patterns, and/or have exceptions to the rule.
    I am not ready for the step to move on to these kind of puzzles. And failing 2589 rated puzzles as well as 2700++ puzzles just seems to be natural. Even Magnus would fail some of these.
    Looking at such high rated puzzles is sometimes like these optical illusions:
    The picture where there are only circles, but you think you see spirals.
    Or many other illusions, that are hard to believe even if you are told the solution.
    My aim is modest: I want to become an CM. (However, once I reach this, I might try to become an FM, then an IM, then GM and then WM...).

  6. The point is, we need the patterns which have no names. And something that has no name can't be remembered. Or visualized.

  7. Do we need them? Maybe to become a GM. I mean, these puzzles are expected to let most IM's fail (according to their CT Blitz ratings).

    Your point is true: most patterns we actually do know.
    It is not the case that we did not see the patterns. We often did.
    Nevertheless we fail some problems. But I think, this is because we know the patterns with dangerous "half" knowledge. Jerimy Silman says in his books, that the typical A-Class players has covered everything that is there to be covered. The trouble is, that some parts of what he used to know is just losely in place. Well, yes, we know about the lucena brigdge builing in an engame K+R+B vs K+R. Can we actually still win it against Rybka?
    Or we heard about the far opposition in a King/pawn endgame. I found out I could find a win in such an simple endgame, because even though "I heard about it" - I did not know it as well as I assumed I would.

    Our fundaments are so rotten, that further progress is not possible, because further knowledge-building just wont hold on the rotten fundaments.
    The difference with an A-class player and a master is, that a master has the information quickly available. Therefore, as aoxomoxoa points out, a master can cover much more information in his working memory. The information must be accessible within 20 seconds and also beeing processed within this time. The more you can "zip" the information, the more you can see.
    Beeing quick is very important, especially in those puzzles you showed. Because, as you said, we had a short view on each candidate move. We saw the right move, but somehow didnt chose it. Even in the last puzzle with trapping the white queen with 1...b6-b5 - didnt we see the move RxN and simply thought "well it is covered by the knight. Anything else white could do?" and then we moved on, checking other possibilities. Then we decided b6-b5 is the right move.
    What we failed to see was, that after the exchange sac with RxN, the other knight didnt cover the pawn a7 anymore. So the knight was overloaded.
    In the very short time, where we had a look a RxN --> if we had the easy overloading puzzle ready within 1-2 seconds - we would not have failed to see the danger.
    And hence we would have concluded to move the knight away to g6 and attacking the wB.

    But in the very short time that we took a look we failed to see it.
    Also in the very first puzzle some days ago, we only might have considered Rf8+. But the 1-5 seconds that we had a look at this variation, we did not see the distraction pattern. Which is on its own easy. But at move 4 we cant spend much time in our search tree. Instead we had only a short look, and then continued to check a different branch in the tree. We might come back several times to a variation, and after half an hour or so, we will evtl. find the easy pattern.

  8. To "know" ( recognise )something ( a pattern for example ) is not enough. GM-Smirnov says at one lesson:
    There are different levels of knowledge.
    My short version:
    1) To know - have heard and understood
    2) understand it deeply, the limitations, the applications,...
    3) to automate it - to make it a skill ( wich needs 3-6 weeks of training )

    If you know, how to drive a car
    then there is still a lot of practicing necessary till you can shift a gear f.e. without thinking...

    I think with tactics step 2 is not necessary for to do step 3. At kate-kid the boy had to wipe the floor in a strange way. At his first fight he used this movement to defend an attack without thinking.

  9. Since an automated skill is semi-intelligent, its level of intelligence depends on the second stage.

    In the case of the karate-kid, that intelligence was added by mister Myagi.

    When I did Troyis I became better by just practicing. Yet it was suboptimal. Had I devised an intelligent strategy first, my results would have been better.

  10. Me Munich:
    Trois is a good training, to some extend.
    Like these boardvision trainings with the Fritz GUI (attack, check, defence training. For instance: "how many attacked pieces can you find per minute?").
    It is possible to spead up the skill to have a quick board overview and dont overlook "hidden" pieces at the other side of the board (such as a black bishop an a7 hidden between pawns a6/b7 - and this bishop is attacking the white pawn on f2).

    But you must not expect wonders. It is a little extra skill, and every little bit will help.
    Still, in the very fist puzzle (with the distraction theme) I only considered the check ...Rf8+, but not the check ...Bg3+. Nevertheless it is good to be able to find 35 checks/ minute instead of 6 or 7 checks per minute (the value I started with when I did the check-training).

    Actually, you (Temposchlucker) found out what is the hypothesis of aoxomoxoa: difficult puzzles are made of several easy patterns. While an easy-easy puzzle contain often just 1 simple tactic, an "easy"-difficult puzzle often contain 2 or more easy-easy puzzle tactic ideas (remember my 2+2 = 4 equation?).
    Here the hypothesis of aox again, see point 5 & 9:

    I might add, that once you covered all easy patterns, it could be a good training to train hard puzzles to learn the combination of the patterns. But this implies, that you know the underlying patterns very, very well.
    At least I found out, that I can solve most of the easy puzzles, but many of them I solve slower than the average player who is rated 400-700 elo below me!
    Up to one third of the easy puzzles I cant solve quicker than players 400-700 elo below my CT Blitz rating (again, I would not measure in CT Standard rating, since I believe it is more inaccurate as a gauge for difficulty).
    Just seeing the right candidate move is not enough. Most of the moves I saw in a difficult puzzle - but did judge them wrong, because I did not see the simple tactic. So at least I looked at the right moves, but didnt find the danger behind them, because I looked not long enough (such as in the last example of yours with the queen trap puzzle with the wrong move 1... b6-b5).
    It would have helped if I had seen the true meaning of RxN tactic. It is easy, but I did not spot it when I had a look at it. And I only looked very few seconds on it.

    So looking now at all these difficult puzzles make me believe, that most of them are easy, too - if you know the underlying simple tactics within a few seconds. (Not just beeing able to solve it, but beeing able to solve it when I have a look at the move for a few seconds.)

  11. When I did look at this puzzle, I remebered your difficult pin puzzle. There is some similarity somehow. Again it is not the puzzle I had in mind, but of all puzzles I presented so far, this one is probably the closest little brother of the "easy-difficult" puzzle you gave here. Still I am not 100% happy even with this one. I know that there is an easy one out there, giving you exactly 1 pattern which is striking similar.

    Anyway, the one I found today is rated 1590 CT Blitz rating, so it isnt "very easy" but "rather easy" for an A-class player:

    If you failed the difficult puzzle allready in the first move, chances are, you might have failed the "rather easy" puzzle with CT Blitz rating 1590, too.

  12. P.S. I looked into the comments of the "rather easy" puzzle with CT Blitz rating 1590.
    I like dusterov's comment (a CT-Blitz 1670 rated player), which is a good point:
    "The pin on the black queen kinda made this one obvious! no thought required here for those that see pins without thinking."
    Nevertheless the average solving time is 30 seconds, which tells me that a lot of players did not see the idea for quite a while. So either you see it instantly (within ~5 seconds for the idea, plus some seconds for blunder check, plus time for clicking), or you dont really knew the pattern.

    I saw the pin idea instantly in the difficult puzzle, it was the very first move that came to my mind.
    If it had been a hidden side variation, I think I would have spotted it.

  13. Could not help solving it in 9 seconds (which includes clicking 3 moves) AND at the same time thinking about the difficult hard puzzle when I saw this one, which is CT Blitz rated at 1402: