Sunday, January 08, 2012

The downside of pattern recognition

In the comments of the previous posts we discovered already one big downside of pattern recognition. Patterns fight for their life in our unconscious mind. Their very existence obeys the laws of Darwin. The fittest will survive. When we see a chess position the patterns shout to us "use me! use me!". Every time a pattern is retrieved it becomes stronger while every time it is not used it becomes weaker. Until their shouting fades away into oblivion.

When we encounter a new position, the patterns with the loudest voice manage to make it into consciousness. While the more modest patterns are overshouted by their cheeky brothers.

When a pattern that is not the best fit overshouts a pattern that is the best fit, it takes time and energy from us away. Just like in real life where the biggest mouth with an unfitted brain has the biggest wish to rule the nations. We start to spend time to make the wrong pattern work. While the clock ticks away the seconds. If we are lucky we take another look and change the context somewhat, and a more fitting soft-voiced pattern will enter our conscious mind. If we are not lucky, we miss the right pattern altogether. So one downside of pattern recognition is that it costs us time or it makes us even miss the solution alltogether. There is yet another downside though.

I have done a set of difficult problems at CT. As Munich pointed out, it probably makes no difference what range of problems you choose, as long as you fail them.
I have added intelligence to the solutions, by defining the reasons why a move in a certain position works while others do not. I formulated a little narrative with a logical reasoning in it.

In an attempt to automate this added intelligence, I did the problemset a second time. Which revealed a second downside of pattern recognition. The patterns shout so hard that the narratives can't be heard. This means, I remember the moves straight away, or the relevant patterns, but not the logical reasoning behind them. I can solve the problem without the need to remember the logic behind it. Since I want to automate the logical reasoning, this simply doesn't work. I have to invent another method.


  1. I'm not sure you can "automate the reasoning". Usually patterns help identify candidate moves in the initial position, and down the calculated lines. However, the selection of moves still requires conscious calculation (at least a blunder check), and evaluation. If you're stuck in a variation (ie. evaluation is not favourable), either you have a new pattern which gives a hint at the solution, or you have to formulate the problem ("if only...") and generate plausible solutions from here.

  2. It is not the reasoning actually, what is automated. That can't be, since logical reasoning is a conscious process. But even in logical reasoning you can recognize logical patterns. Those patterns I seek to automate.

    Think of standard scenario's which fit certain positions. Think of hunting the King. You always must lookout for escaperoutes, must save every tempo, must add pieces to the attack with tempo, be aware of counterattacks etc..

  3. It should be possible to automate a thought process ( would not need "only" failed problems ) This thought process might start with CCT. It would be like a program written in pseudocode. You step through it every time you see a problem. ( <- this is my interpretation of Smirnovs writing about thought process, i think it might work )
    Today a made a type of error twice at CT: There was a danger for me and the possibility to get something. I made a calculation and thought: the danger is not fatal, i can take. But i was calculating wrong, i should have exchanged an attacker first and then continue to "win".

    Code Example:

    If i am in danger then
    If i can win something then
    check if i can still win
    but do some
    saftyexchanges first

  4. @Aox,
    that's true. But you can't automate it by means of repetition due to the "hindrance" of "ordinary" pattern recognition.

  5. Ah yes, I understand what you mean. It's about asking yourself the right questions, depending on the context on the board.

    That's really expert knowledge : it's about emulating the thought process of strong players, who know, from experience, what to look for in a given position.

    Actually, this is linked to the evaluation of the position, because the first stage of the evaluation is to understand what kind of position you're in. For example, if you're in the opening, your evaluation will take into account control of the center, development... If one side is down material in the middlegame, you put extra emphasis on king safety and piece activity, etc.

    Kind of "strategical patterns" we're talking about here, I think. Much more difficult to classify than tactical patterns, if you ask me.

  6. The patterns shout so hard that the narratives can't be heard. This means, I remember the moves straight away, or the relevant patterns, but not the logical reasoning behind them.

    Could you give an example of the kind of position and narrative you added ? Because if the position is of a tactical nature, I'm not sure you can do better than pattern recognition + calculation + reformulating a specific problem to trigger another pattern recognition as you explained.

    I would associate narratives with strategical problems (and strategic patterns), rather than tactical ones.

  7. I remember the moves straight away or the relevant patterns, but not the logical reasoning behind them. I can solve the problem without the need to remember the logic behind it

    You already have a good memory for moves and tactical pattern but not for reasoning , you are a "beginner" in chessrelated - reasoning.

    You did not repeat "the thing you did want to learn" early enough, one of my hypotheses: no ( early ) repetition ( of any kind ), no learning. That you remember the moves might be irritating but some may remember the size of the board or some other irelevant things.

    Suggested Exercise: At every problem write down the list of logical reasonings for this move/line.
    Take the time for doing that, check list if complete, improve speed ( by spaced repetition? )

  8. @Aox,
    is your advice very different from what you did yourself when doing slow high rated problems at CT (which didn't work for you)?

  9. Big difference: I did only try to "understand" the problem. It was like reading a detectiv story, who is the killer? Sometimes i did put my analysis as comment at CT for example here But i did not repeat any problem ( not intentionaly ) and i did never try to improve the speed.

    I developed my ideas about learning much later, inspirated from you and others.
    Now i know: No repetition in time -> no learning. You reach a plateau as soon as your "quantity*quality of forgetting" = "quantity*quality of learning ( new )".
    With spaced repetition you can minimise the "quantity of forgetting" ( ~ 0 ) and you increase the "quantity of learning ( new )". The quality... hm.

    I think Empirical rabbits ( And Chessimo's and a little de la Mazza's and,and,and ) Idea is the ultimate tool; it just has to be finetuned to the present purpose.

    I will continue with tactics. If i cant improve in tactics (tactics pattern) another hmm minimum 200 points i can forget getting, where i want to be. But after tactics... i wait for your report ;-)

  10. @Laurent,
    I probably will devote a post to an example, but not right away.

  11. In your reasoning you just do not make a list of candidate moves. You just start analysing the first idea/move that pops up in your head.

    To avoid this make a list of candidate moves first and only then begin to analyse.

  12. I think candidatemoves are step 2. First you have to understand whats going on, elsewise your candidatemoves are "random".
    Step 3 is calculation

    1) analysis of current situation
    2) Target state-analysis

    Thats my thinkingmethod

  13. At one tournament, I used Pocket Fritz to help generate a list of candidate moves.

    I've never heard such whining.

  14. Yesterday I had a look at GM Smirnovs chess lessons. Again.
    I watched "breaking stereotypes" part 1, 2 & 3.
    Very good lessons from Smirnov. I admit I forgot a lot of it. So repeating is important here, too.
    His videos are somehow guidlines, so especially tempo might find them interesting, so I am writing this message into his blog, too.
    (Edit: search in google with keywords "GM Smirnov breaking stereotypes")

  15. Seems to me that the problem is not that the pattern is shouting. It is the position that is drowning out the narrative.
    A pattern is the essence of a position. I can understand why you would want to define it with a narrative, but language is not or hardly suited to define chess patterns. But the stored pattern should be a visual image that can be used in a different position.
    You will see a back rank mate in many position, because this is a solid image that you can use in completely different positions. But for other patterns that are less obvious for you mind, you may remember the image of the position, but not the image of the pattern, so you won't be able to see it in different circumstances (positions).
    This is one of the flaws of the 7 circles. Pattern recognition is replaced by recognition of positions.

  16. @Phaedrus,
    the function of the narrative/logical reasoning is to guide the pattern recognition. That rescues the almost drowned patterns.

    You are right about the difference between recognizing a pattern and recognizing a position.

  17. for the pattern recognition: it does not matter if you recall the pattern or the whole position - as long as you understand the solution move.
    If you know that Rxe8+ IS the winning move in a position, because you remember the position, and at the same time you also know WHY Rxe8+ works --> it does not matter, really.

    I doubt you will be able to remember a move (Rxe8+) if you dont understand it.

    Try to memorize this:

    And now this:
    "Humpty Dumpty sat on the wall."

    The second example (though containing more letters) is by far more easy to memorize - because it makes sense.

    So for a move Rxe8+ --> you will memorize the move because you will have understood the move by the time you remember the position.
    It will be like this: you wont look at the rook when you see the position. You simply know that this position requires a rook check. So first thing you will recall is a check giving rook taking on e8. Only then you will look for the rook. It is the square you will remember, not the piece.

    "This position requires that move".

    So dont worry about the difference of pattern and memorizing full positions. It wont matter, because by the time you memorized the position you very well understood the pattern behind it.

  18. Munich, I have to disagree. If you couldn't solve a position and then look up the solution, it is very well possible that you memorize the moves, but not fully get the underlying pattern.

    In that case, if you solve the puzzle again, you might remember the moves of the solution, but not the tagged narrative and/or the pattern. So if the same pattern occurs in a (slightly) different position, chances are that you will again fail to find the winning move.

    I notice this when I train my students (young children. their succesrate in solving puzzles they have done before is higher if I give them the same position, then if I give them the same position with colors reversed.

  19. Phaedrus, my son and me, we both notice during our games tactics we memorized from CT. What do say to that? we see the difference, too.

  20. 1) If you cant understand something then its a good alternative to learn it simply by heart.
    2) Things you cant understand are almost impossible to learn ( try to learn a poem in Japanese )

  21. Aox: you explained it very well, thanks!

    I wanted to add about the worry, that you mistake a puzzle for another due to similarity.
    Even if so, then you get a "fail" in CT. And since you repeat your fails, you will learn about the difference between similar puzzles.

  22. - - - -
    phaedrus wrote: their succesrate in solving puzzles they have done before is higher if I give them the same position, then if I give them the same position with colors reversed.
    - - - -

    Well, yes, true. I found out that if I turn the board by 180° (an option in CT), then it takes me much longer to find the solution.
    I dont know if this is comparable to inverting the colors, but I assume it is to some extend.
    Nevertheless, after looking some seconds on the puzzle, then I can remember the puzzle as "known".
    I simply need some extra seconds to access my memory (my "database").
    Turning the board slows my access to my "database" down. But I am also suffering from the effect, that I like to move foreward in action and not so much to look downward. Sometimes I need some seconds to find out that I need to move black instead of white.

  23. Where pattern recognition hurts - look at this puzzle with a CT Blitz rating of 1830! Can you even solve it within a minute?!