Friday, May 18, 2007

Feedback.

Let us, for the sake of reasoning, assume that the method as described in my previous post is the best method to learn something. What are the essences of the method?

Baars makes it plausible that 99% of the learning takes place in an unconscious way, while 1% or so of the process happens conscious. On the other hand, that 1% conscious work is absolute necessary. Without that, no unconscious organisation of the brain will take place whatsoever.

Since this 1% conscious work has an immediate effect on the 99% unconscious maintenance of the "chessmodule" in your brains, we have to know which rules govern the conscious part of the work and how it effects the quality of the chessmodule. The importance of this can't be overestimated.

The core of the process is based on feedback.
Which elements do we have here?
  • Consciousness. The feedback must take place with full focus of attention.
  • Produced errors. Flaws in the chessmodule results in errors and suboptimal moves during play.
  • Identification of the errors.
  • Correction of the errors.
Consciousness.
Everthing we do on the automatic pilot is not conscious. This might explain why solving 100k+ problems isn't as effective as one would expect. Only the moments when we work with full attention it will have an optimal effect on the unconscious part of the work. What is necessary here is discipline, alertness for distraction and energy. Passivity and laziness are your enemies here.
Conclusion: consciousness is everything, being active is everything, automatic pilot is tabu, passivity is tabu.

Produced errors.
If you study your own games, the errors and suboptimal moves will for 100% be related with the flaws in your chessmodule.
If you use a standard problemset like CT-art, it will only be effective if a presented problem is a lookalike from a situation you would do wrong when you encounter it in a game AND there is a chance that you will encounter it. It is not unlikely that that is the case for only 50% of the problems (figure is arbitrary). If so, 50% of your efforts are wasted beforehand and will not lead to better play but only to better solving CT-art.
When you study grandmaster games, you try to invent a move yourself before you look at the move that was actually played in the game. In that case mistakes originate for 100% in the chessmodule in your brain. On the other hand the positions you will find yourself in can bear very little resemblance with what you probably encounter in your own games.
Conclusion: study your own games. Second best: study mastergames.

Identification of errors.
It is undoable to find all the errors yourself. If that was easy, you would not have made the errors in the first place. Trying to do this yourself must be a waste of time logically.
There are two good methods to identify your errors: a chessengine or a coach.
Chessengines are so good by now that this would work. A coach will work fine too, as long as he is prepared to analyse your games and he don't let you train an K+B+N vs K endgame (which you can do well without a coach).

Correction of errors.
This is the most tricky part. What exactly makes that you will not make the same mistake again? Can it be done WITHOUT repetition?
Since the chessgame isn't "solved", there are usually no alternative moves from which you can know that they are 100% correct. Basically you replace a bad move of your own by a better move suggested by a coach. Hence you try to adopt the playing style of your coach.
A computer is not a good alternative here since it only tells you WHICH move to correct and HOW to correct it, but not WHY the correction is better. This WHY seems to be necessary contextual information for a good maintenance of your chessmodule.
This part of the story needs more research.
Conclusion: you need a coach to analyse your games.

3 comments:

  1. Interesting. At work can't say much now, but one think I'd add is that this 'module' does more than just pattern recognition, but can be trained to take care of time management, thought process (e.g., look for all checks, captures and threats first). So those CT-art problems that aren't likely to come up in real games can still be used to train the module on the thought process (though I think real games are required for time management module building).

    The problems are also good for practicing analysis, which is hard partly because it is the one stage of the game that is explicitly conscious, effortful, attention-demanding. But still, practice improves analysis skills.

    Fritz vs coach is interesting. Obviously, Fritz is going to be better at chess than almost any human player, so a coach is good for reasons other than showing you the best move. I think you've hit on one reason why (another is the slow build of a rappport with a coach: he knows your play, and tailors his instruction to your module specifically especially as you get to know one another better).

    Plus, there is no overestimating the 'nagging internal mother' effect that you will develop over time. 'Don't play passively when winning.' 'Open up lines, don't attack just yet, you don't have enough material developed.' Etc..

    ReplyDelete
  2. I would guess that the chess module, like other parts of the brain, will fade with time if not maintained. I think the solving of simple problems is a kind of maintenance for the chess module. As you point out, the chess module won't be upgraded by automatic pilot solving, but perhaps this kind of exercise is useful in the maintenance of the foundation of the module.

    ReplyDelete