Saturday, April 26, 2008

Phantom-knowledge and habit-forming

Let's see if it is possible to build a model of the transformation process of knowledge into procedural habits. What do we have?

When you read books, blogs, magazines, talk to other people etc. you get lots of information.

By thinking about information you transform it into knowledge. What is the nature of this knowledge? In the end chessknowledge relates the characteristics of the board (input) to a certain move (output). All other knowledge is obsolete with regard to chess improvement.
The output generating process is divided into two stages which compare as 1 : N
  • diagnosis
  • remedy
To make Soapstone happy you can say that diagnosis culminates into evaluation of the position.
Let's have a closer look at how knowledge generates a move.

Spectrum of knowledge.
On the gross end of the spectrum we find rules:
  • Characteristic: open file
  • Rule: occupy an open file with a rook
  • Output: move the rook
The advantage of a rule is that it gives a single kind of output for a whole lot of similar positions.

In the middle of the spectrum we find conflicting rules:
  • Characteristics: open file, king in danger
  • Rules: occupy the open file, bring king into safety
  • Output: move the king
You have to value the goals you achieve by both outputs against each other. This valuation can generate new meta-rules.

At the other end of the spectrum we find concrete analysis. Whether concrete analysis gives an answer to your problem is a matter of the goal you serve with the resulting move. If the goals are too subtle, the analysis is a waste of time.
Another point is the amount of possibilities you cover with concrete analysis versus the amount of possibilities on the board. Often concrete analysis is just too time consuming in comparison to the value of the goals that you can achieve.

Spectrum of goals.
There is a whole spectrum of goals too. Winning a piece is on the gross end of the spectrum and trading a piece to get some dominance over a square is on the subtle end of the spectrum.

Transformation of knowledge.
Explicit thinking about the generation of a move is an extreme slow process. Since there are so many aspects involved it is absolute necessary that the gross part of this task is done automaticly. This is done by transforming the knowledge into habits. To transfer the knowledge to the procedural part of the brain.

How does this transformation work?
The following issues play a role:
  • consciousness
  • repetition
  • different pace of the procedural part and declarative part
  • emotional reward
It looks like the procedural mind works by imitation. You have to repeate a certain thoughtprocess a few times and the produral memory starts to imitate it. So it becomes semi-intelligent. It copies the intelligence that's in the knowledge.
Let's have a look at a few examples first:
  • To learn to drive a car takes 45 hours at average.
  • Adjusting my pitch took me 22 hours. I sang 2500 tone-ladders.
  • I played Troyis for 22 hours
  • A crash-course skydiving took me 32 hours. Pun intended.
With both driving a car and skydiving you learn multiple complex motorskills simultaneously.
But are these examples really a good indicator for how long it must take to learn a new motorskill? If I seriously and consciously repeat a word of a foreign vocabulary for 7-10 times, I have learned to translate it in the future. How close is this related to a complex motorskill? Words of foreign languages tend to fade away overtime. From riding a bike and swimming is known that once you have learned it you will never unlearn it. Has that to do with the fact that you will always repeat those motions much more than 7-10 times? How many conscious repetitions are necessary to prevent a word from being forgotten during a lifetime (forget about the demency at the end of life. Pun intended.) How often must you consciously put your rook on an open file before it becomes a habit?

15 times sounds reasonable. But if 15 times is enough to form a lifetime habit, why are so many hours needed in the examples of driving, singing, Troyis and skydiving? Why know most sports so many hours of training? If you filter in with your car 15 times you know it for life. If you park your car 15 times you know it for life. If you swim 150 meters you know it for life.

It seems that lots of repetitions are needed when precision plays a main role.
For singing it isn't sufficient that you can hit a note at the right pitch by copying a note produced by a keyboard:
  • You must to do it from 32 different kinds of previous notes. Since the stand of the muscles of the larinx of the previous note influences the next note. So you must exercise 15 x 32 seperate habits.
  • If you maintain every note for 5 seconds, it is simple, because you have time to adjust your larinx. But the shorter the note, the more difficult it gets and the more control is needed.
  • If you have learned to copy a note from a keyboard you have learned to produce the harmony do-do. But during a polyfone piece you must be able to produce any harmony with precision, so you have to repeat the two points above for all kinds of intervals.
  • Once you can hit any interval fast, you must concentrate on the overtones. Those have to harmonize with the rest too. So the whole training process starts over but no for the overtones.
  • Once all this is mastered you sound like a dead midi-file from a computer and you have to learn to make music from the notes by adding emotion to it. But how do you do that? etc. etc..
  • Not only must you know all this but you must train your ears too. You must learn how a harmony must sound, how overtones sound. There are altered sounds reflecting from the walls, transmitted through your skull etc.. All this you must learn to distinguish.
If every little subtask needs 15 repetitions then it becomes clear why it takes so much time to master a complex composed skill. If your consciousness starts to dwindle the process will stall immediately.

The same is true for chess. You must learn an awful lot of habits to master chess. The resulting semi-intelligence is as good as the quality of the initial knowledge. If your consciousness starts to dwindle the process of transformation of knowledge into habits will stall immediately.

Knowledge that isn't consciously processed into procedural memory becomes impotent phantom knowledge. It is assimilated into the stinky pool of scholarship. It will sink into the marsh like a hippopatamus!

There is no limit in the amount of knowledge that can be transformed into procedural knowledge. Nor will the amount of repetitions necessary to store the information differ much.
The limitations of the process seem to be due to the following factors:
  • the quality of the knowledge
  • the quality of the consciousness
  • the sheer amount of subtasks

There are some big advantages when knowledge is ingrained in procedural memory.

Procedural memory can produce output at a much higher pace than the slow explicit thinking brain. Besides that it can process information parallel.

Visualisation of the output is something you seem to get for free when knowledge is transformed.

In contrast to explicit thinking there is virtual no effort needed in the procedural processing of data.

Freeing resources.
Procedural tasks don't need the short term memory (STM) while explicit thinking can't do without STM.

A problem is though that the output is only semi-intelligent. If you have learned impropriate habits the outcome will be faulty moves.

The output of your procedural "chessmodule" must be constantly monitored by your conscious thinking. You have to identify faulty moves and to define what is necessary to correct them. The result of the feedback re-enters the process as new knowledge.

According to my interpretation of Baars, consciousness is paramount in the feedback.
  • The lowest form of consciousness is automatic. Here you will find the blitz-players with a rating of 1200 after zillion games.
  • The next level is trial and error. This works well when the result of a move is gross and well-defined, so as the win of material. But if the result of a move is subtle the correction will escape you since more effort is needed to develop a strategy.
  • The next level are narratives. Those help you to identify what has gone wrong and to develop a strategy to do it better next time. Every new strategy needs its repetitions to become a new habit.
I realize this post isn't an easy read. But any feedback is appreciated!


  1. i always knew that you were very understated temposchlucker. :) bravo man!

    the issue is NOT what is discussed but what is not discussed, of which there is little that can be found here! you are a real gem sir.

  2. DK,
    I have planned a post on a subject that is never been discussed here: creativity in chess:)

  3. This all seems pretty reasonable, with some lacunae of course.

    A little technical point about Baars--most people assume the majority of modules are innate. Some parameters are set via feedback, but in general the innate modules need little effort to build (e.g., the language module needs inputs about the specific language to get the specifics right). A kid just picks up language naturally, very very quickly, for instance. Humans don't have to work to see things, our visual system takes care of a multitude of tasks, of which we see just the barest outputs (e.g., object segmentation, depth, color, etc, are effortless and unlearned).

    Chess, of course, is more like swimming, tennis, or riding a bike. It taps into our ability to acquire skills more generally, skills that are not genetically entrenched. So it is likely that the chess "module" like the swimming or tennis "modules" indeed depends much on conscious feedback as you describe. (Here 'module' is used more generally to refer to any specialized processor for which we don't typically have conscious experience of its fine-grained operations, but its output (e.g., a back-rank mate just pops out at you--beneath the surface a lot of neuronal computation just happened that you are not aware of)).

    Incidentally, I am training to swim well--it is very, very hard. There are tons of aspects (breathing, body rolling, correct micro-placement of hand as it goes into the water, the arm as you pull during the stroke, the kick, the position of the head in the water). It is impossible to practice all at the same time and end up doing it well--people's attention can't divide that way--typically attention can only focus on one thing effectively.

    So there are swimming drills, such as the 'spear the fish' drill in which you practice only one thing, putting your hand into the water so as to minimize resistance in the water. This must be done many many times over many months to really get it right. At first with much feedback, thinking about little mistakes that are possible, a good swimming coach is ideal (of course). Ultimately it becomes a habit.

    This is why I like MDLMs micro-drills, and general drills that have you repeat things over and over for chess.

    In swimming, also, you don't just practice the water entry drill. You typically spend about half your swim sessions doing drills, half just doing your freestyle stroke trying your best to integrate everything you've learned into a coherent hole. I image this is much like singing practice--you don't just do drills, but also like to actually sing songs.

    One other thing, there is some learning by osmosis in swimming (pun intended). Just by watching others swim well, an integrated whole, you pick up a lot of subtle cues that you might not pick up by looking at the drills. For instance, the way the breathing and body turn work together. It might be hard to see this if you just work on breathing in one drill, the body turn in another drill ('body turn' in swimming is when you twist your whole body so it is almost on its side as you pull your arm through the water, as that generates much more power than if you just used your shoulders).

    As I mentioned in a previous post, this is probably a good motivation for studying master games.

    Some questions that this spawns:
    1. What is the optimal ratio of simple drills to playing? In swimming most agree you should spend about half your time on each.
    2. What is the best way to split up the chess drills? E.g., focus on one thing for two months? Focus on three things, a little each day, or a different one each day? Should the study of master games count as simple drills, or is it more like playing an actual game?
    3. What are the best drills to improve at chess? A corollary of the above is 'simple drills.' Of course this depends on your skill level, so it translates to 'simple for you.' So, simple (for you) endgame, tactics, strategy problems. To make the drills complex (for you) drills is a mistake. There are tons of simple tactical drills out there, but very few strategy drills (PCT seems to be the exception, as most strategy books and software are fairly sophisticated--this may be because tactical skills are the hardest to acquire, while positional skills are relatively easier as they tend to rely on general principles rather than the concrete).
    4. Where does master game study fit into this? It seems qualitatively different from both doing drills and playing, but many instructors think it is a key to success.

    It is kind of funny that chess is considered an 'intelligence' game, when in fact it seems to me much more like swimming. Nobody would say you were stupid if you couldn't learn to swim well. On the other hand, swimming just involves bodily movement, while superficially chess involves intense thought. But as you wrote a while ago, it is really only that extra 200 rating points or so that is really about this kind of stuff--the core rating is based not on thought but on how much you have acquired by working your ass off. This is what the studies support (well, admittedly there is an IQ component, but IQ is never as good a predictor of chess ability as is time spent studying the game).

    People who don't play chess see it, and think "Wow, if I were to do that, I would have to be thinking so much, way more than I could ever think in the amount of time you are taking to make the moves, so you must be really smart." They don't realize that misses the point--people with experience actually aren't thinking as much as they would have to in the same position.

  4. On my above point about not just doing drills, but actually playing.

    There were many times during the Circles when I was kicking ass at tactics problems, but was so caught up in the Circles that I wasn't playing. Invariably, when I'd come back to playing I would play horribly, I would even miss simple tactics that I would have recognized immediately in the context of a puzzle. This suggests that the coordination of drills and real games is not something to do just because a drill-only approach is boring. Integrating the two is required, there is a co-evolution of highly specific drill-dependent skills and ability to tie things together and actually play well.

    Just working on the entry of the hand into the water, and then entering a swimming competition, would be a huge mistake. This is because the hand entry actually works synergistically with other proper movements (like your pull through the water, body turn, and all that). While drills can teach you compontents, it takes actual play to get the synergies down.

    But this is where master games might be good, as they typically give you a picture of synergy par excellence. Then again, you gotta play too.

    Perhaps with singing it is similar--you don't prepare for a performance by practicing just C sharp, or even doublets of notes. Real songs involve hundreds of notes for which no amount of one or two note drills could fully prepare you.