Wednesday, April 23, 2008

Knowledge, calculation and evaluation

While developing the theory about motorskills I was well aware that I deliberately left out a few topics. Reading this post from Soapstone I realized that it is now the time to address them in the light of the new findings. I don't know where this post is going since I use the writing as a means to think about it. So please bear with me. Before I start to talk about evaluation, calculation and knowledge I summarize the previous in an attempt to get some focus.

Experienced (7+ years) chessplayers don't differ all that much in amount of patterns they recognize nor do they differ all that much in the skill of recognition they have. The real difference is made by the automatic scan-habits they have.
When a task is performed with the aid of procedural memory you get these for free:
  • speed
  • visualisation
It works the other way around too: when there is no speed and you can't visualise it you can be sure that you lack the motorskill to perform the task.

Knowledge is a strange commodity in chess.

Acquisition of knowledge.
Sometimes it is very difficult to obtain useful knowledge, but once acquired it is usually short lived. Within days or weeks the knowledge is converted into procedural memory.
For the acquisition of new knowledge the following elements are indicated as being helpful:
  • a coach
  • analysis of your own games
  • books
  • blogs
  • etc.
Transfer of knowledge.
The transfer of knowledge into procedural memory is assisted by narratives. Just like the knowledge of the rules inevitaby will transform into procedural knowledge the same is true with almost any chess knowledge. You can't avoid it. Narratives form a bridge from acquisition to transfer. New knowledge that isn't transformed into procedural memory will usually be forgotten soon.

Opening knowledge.
The only edge that you can have with knowledge while it is still in the state of declarative knowledge, is the advantage you have with opening knowledge. Actually this isn't quite true since opening knowledge is constantly renewed. If a certain amount of opening knowledge isn't renewed within the last decade or so then usually everybody knows it. Even in a chesscafe almost every tourist knows how to play the first 5-8 moves of the Najdorf with white. Automaticly and without even knowing why.

The word calculation is used in a rather unusual way in chess. It has nothing to do with arithmatic. How shall I define it? Maybe this "Soapstone-inspired-definition" is useful:

Calculation is everything what happens in the mind before you evaluate the endpositions of the tree of analysis.


After the calculation you have to assess the endpositions.
We like to belief that our decisionmaking is conscious and rational. Usually there are too many factors involved so we make a sort of emotional assessment.

Right now I realize I first have to think alot more about this subject, so I will shift that to a future post. There is clearly a lot that has to be taken into account. Sorry if I have awoken expectations that I can't fulfill at this very moment.


  1. Nice. It is refining nicely.

    I like soapstone's definition of calculation. It highlites something I've thought but not articulated--calculation is a messy business. It involves a combination of pattern recognition and highly focused and concentration-taxing visualization. E.g., visualizing a line in a concentration taxing way, I might see an end note of one branche leaves a back-rank mate. The mate pattern I notice with no calculation or thought it is so engrained in me at this point. The more such 'chunks' we have at our disposal (pattern-outcome associations), the less I have to perform the taxing visualization.

    It's because it is so messy that I find some descriptions of calculation superficial, anemic, unrealistic, as they don't stress he pattern aspect. I'm thinking of Kotov here. On the other hand, one must be careful...we can see a pattern but not notice a defense (e.g, I've visualized 4 moves ahead, see a back-rank mate pattern, but his bishop is actually guarding the mating square!).

    None of this really gets at board evaluation (I wrote a lot about this in Chessplanner), something that is also ultimately a messy mix of conscious evaluation and pattern recognition.

    I think that the framework you are developing for scanning can also apply to anything worth learning, such as board evaluation. It starts out conscious, with narratives, and becomes an unconcsious automated procedure, unconscious processing that we are only consciously aware of the outputs, not the steps anymore.

  2. Blue,
    I think that the framework you are developing for scanning can also apply to anything worth learning, such as board evaluation.


    There seem to be two problems with procedural knowledge though.

    First, it is semi-intelligent. If you have learned a flawed procedure, a faulty solution comes up.

    Second, the question arises why your rating actually plateaus after a certain amount of years. This is not because your brain is "full" at a certain moment. Must the cause of plateauing be sought in the knowledge stage or the procedural stage? The answer of this question is important because it hints to where you can improve.

    knowledge phase
    Is it because it becomes harder overtime to find "winning knowledge"? That is certainly true. But this is a problem that is equal for everybody and it tends to level out overtime automaticly.

    procedural phase
    An amateur with 50 year experience has far more knowledge gathered overtime than a 13 year old Magnus Carlsen. Yet the amateur will be blown away in a game. To me this indicates that the procedural stage of knowledge is paramount.

    This seems to indicate that the automatic conversion from declared knowledge into procedural knowledge is of low quality in amateurs. Hence procedural training is the way to correct that.

  3. Tempo: excellent! I think this is great stuff--those of us advocating experience, intuition, volume training always need to remember those at ICC with a million games still rated 1200. That's some important data.

    Note I just revised Chessplanner, polishing off the final revision, as discussed here, so ignore the above link and use this one.

  4. It's funny, as I train for the triathlon I got a swimming instructional DVD. They talk about two phases of learning--the 'cognitive' phase when we have to think about what we are doing, when we make lots of mistakes that are best met with quick corrective feedback. Second, the 'skill' phase, when the stuff we have been practicing has become unconscious, automatic, and effortless.

    This is such a ubiquitous thing, so clearly applicable to chess, it points to the overarching usefulness of feedback (the less feedback you get, the more likely you are to end up with crappy habits, like that guy with a million games rated 1200). This is why I prefer literature with expert narratives rather than just variations or game/position dumps. and is also why I harp so endlessly and annoyingly about chess coaches. :)

    On the other hand, even watching a good swimmer can be very very helpful if you know what to look for, so once you hit a certain level in chess, you can just see a good chess game and learn from it.

    Blitz, I'm sure, cultivates awful habits. The bad thing is, it is so easy to not see that fact, to feel like you are getting better just because you are moving chess pieces around. It does have feedback, though, in the form of major fluctuations in the evaluation of the position through tactics and mates, so there is something about the form of the feedback that must explain why it has diminishing returns.

  5. Blue,
    so there is something about the form of the feedback that must explain why it has diminishing returns.

    That sounds very logical. Just as you can only "fill" procedural memory initially by means of conscious declared knowledge, procedural memory can only be corrected with conscious feedback. We tend to apply automatic feedback, though.
    Since automatic means unconscious, this doesn't work.

  6. Blue,
    I read your new version of the Chessplanner. It reminds me of the metaphor of a doctor and an ill patient. There are two main stages in that process:

    Applying remedy

    In this cognitive paper about the 5 steps from novice to grandmaster it is said that a master has fully automated the stage of diagnosis while he has to think about the remedies. That is logical since diagnose stands to remedy as 1:N.
    The grandmaster has fully automated the remedy phase too.

    We can know that this is true because Susan Polgar has scored 96% in a simul against 1100 opponents while she used 2.6 seconds per move at average.

  7. This gives in fact an answer on my question regarding evaluation: it must become fully automated too.

  8. That's an interesting article I had forgotten about.

    I guess everyone and his mom has an N-step process described for what it is like to go from beginner to expert. I wonder if there is a 'standard model' in psychology yet. At least most of them agree on the general picture.

  9. I wish going from beginner to expert could be fully automated. :)