Monday, April 28, 2008

Theoretical conception of an exercise

A lot has been said about how the process works. Let's see if it is possible to conceive an exercise based on the diagnosis so far. First let me summarize a few important issues. These should be of help to conduct how to exercise (my english seems to deteriorate)

It all starts with a special kind of knowledge. The knowledge describes the input (characteristics of the position) and the output (a move) along with how to get there (the narrative). This proces of input and output is visualized before the minds eye, so that the apprentice can copy it. The apprentice is the procedural part of the brain.
The attitude is active, that is to say your attitude is based on the question: how can I make that I will see the solution in the future immediately?

It is best to start with the motorskills you need every move. I have done the problems on Chess Tempo past week and I found that 3 different kinds of scans give me in almost all cases the solution:
  • Trap-mate-radiation of pieces. There are a lot of mates at Chess Tempo. It is useful to visualise the "radiation" that is emitted from your pieces and that diminish the space of the opponent's king. I consider a mate to be a special occasion of a trap. It gives you fast an idea if there is a mate possible or not.
  • Convergence of my own pieces. This is the most fruitful scan. Because if you want to attack, you usually need to move your pieces to the convergency squares. It tells immediately which pieces of you can play a role in the attack and which not.
  • Targets. Targets and the route to get to them reveal which enemy pieces are possibly subject of the attack.
A full scan of these 3 points is usually enough to trigger pattern recognition which in turn reveals the solution. After I have found the solution I formulate a narrative to distil the knowledge from the position. Then I visualise the essence of the solution so the apprentice can copy the process.

  • The 3 scans (guides of pattern recognition)
  • Transform the solution in a narrative
  • Visualise the narrative
Goal of the exercise:
The apprentice must learn to do this process automaticly. Educating your automatic pilot.

Constraint during the exercise:
Active attention. Can not be done on autopilot.

Nothing to worry about. Even those who need 70 repetitions to learn a single foreign word will find that you can't avoid far more repetitions since you do this every move.

(click to enlarge)


  1. It seems you are describing a subset of a thought process, one geared toward finding attacks on the enemy king, that should become a habit once you practice it enough.

    Have you practiced them more than 15 times? :)

    There is lots of research on the topic of habit formation that is relevant to the general topic, that provide quite useful information.

    A tiny sample:
    1. A nice review from psychology, probably more relevant than neuroscience: Habits: a repeat performance
    2. A review article with a neuro emphasis: The role of the basal ganglia in habit formation
    3. Experimental study (some very useful stuff): The Neural Correlates of Motor Skill Automaticity
    4. Another good review from the neuro perspective: Learning and memory functions of the basal ganglia

    Note they don't have the final answers, but for anyone interested in this stuff, it is usually good to read what the people that have been studying it empirically for their entire lives have to say. I haven't read the articles, but skimmed them to make sure they have some interesting stuff.

    Some may not be available unless you are at an institution with a subscription to the journals.

  2. Oh, and
    A new look at habits and the habit-goal interface looks pretty interesting, if dense.

    I think all the studies will support the basic claims we agree on:
    1. Forming a habit is slow.
    2. The key to forming a habit is repetition.
    3. Habits are procedures performed automatically, without the need for conscious deliberation.

    The question of time of habit formation will have the answer "It depends what the task is. If it is more complicated, it will take longer (e.g., a pilot of a 747), if it is simple it will happen faster (e.g., driving a car).

  3. I like your inquiry into specific kinds of scanning, the lack of which disappointed me on Hortillosa's articles. Convergence, enemy king radiation, and the generic targets. Regarding the latter, I don't consciously note en prise pieces and other weaknesses in a methodical way, perhaps why my scanning is defective. I think Chess Tempo may be a little too focused on mates and not enough on material, so I feel that convergence and mating nets are getting stronger, but skills for more subtle wins of smaller amounts of material like pawns and minor pieces are languishing.

  4. BDK,
    I'm sure the difference between our idea's is minimal. Yet I've gone to great length to convince you of my view. To no avail thus far. There is a reason for my persistence and it is not my ego.

    At school I learned that if you throw a stone away that its trajectory is a parabola. Only much later I found out that it was no parabola at all but an ellipse since the stone is orbiting around the earth (the common barycenter, to be precise:).

    For all practical purposes (killing an enemy and such) you can use the formula of a parabola. But for a theoretical understanding it is necessary to know that's an ellipse. In fact that was the great insight of Newton who realized that an apple and the moon obeyed the same law. Which was not shown by their formulas.

    If you read carefully what I have said about 7-15 repetitions then you will find that it is nowhere in contradiction with your idea's. Yet there might come a moment that the subtle difference in formulating matters will play a crucial role in understanding. Right now I have no idea if that is going to happen.

  5. Soap,
    the problemset of CT is rather biased indeed. I'm still experimenting with the scans and it seems to be effective at the moment, but later I will probably go on the look out for a more suitable problemset.

  6. Blue,
    I read your links, thanks for them. As I already suspected, I have used the word habit in a non scientific way.

    You start with a goal in a context. after a lot (!) of repetitions a habitual response grows. This habit is no longer goal-cued but becomes context-cued. For chess this is unwanted. A habit is that I play 1.e4 because I have played that all my chesslife, while I actually wanted to play the Polar Bear.

    Given this definition of habits the question even arises if habits are involved in plateauing.

    In stead of the word "habit" I seem to be obliged to use the expression "context-cued responding with automatic goal pursuit". Now that is going to make my blog more readable!

    Maybe BDK has links for "automatic goal pursuit".

    It is not quite clear whether habit-forming must be prevented by abstinence from massive repetition.

  7. Tempo: I think the number of repetitions required is a minor quibble, so I agree we probably disagree very little.

    I think we do want habits, just good habits. We want it to be context-cued, not goal-cued, as goal-cued responses require deliberation on how to reach a goal (the cue is the conscious internal representation of the goal, and we reason about how to reach that). What we want is the chess board (the context) itself to be the cue, so we don't even have to think about goals, or reason about how to reach them, but just react quickly and correctly like a GM in a simul.

    This confused me at first too, as I was thinking, "But in chess the goal is mate, so we can't be talking about habits." But that goal is implicitlyl incorporated into the habits, if we form good habits, even though we don't have to think about it consciously anymore.

    So I think you use the word 'habit' correctly, I just disagree about how long it takes to form one. I don't think this is a major issue, unless you are using the word 'habit' differently from me. By my use of the term, 7-15 trials will rarely be enough to be able to do something automatically, unconsciously, and well.

    Study number 3 I like, as it offers a nice operationalization of habits--X is a habit when you can do it along with something else that takes your attention away. So, I can play a beginner while talking on the phone because my habit set for chess is probably about a 1000 rating, I don't need to give conscious attention to playing the 200 rated player.

    Habits can be more or less ingrained too. Recently implanted habits are quicker to disappear, easier to fix, than long-term habits. Ones that have had a lot of time to root are much harder to extract, require constant feedback because of the ease with which we slip back into it without thinking.

    And this is why I like chess thought processes. They make conscious something we do implicitly (follow a procedure for move selection, not necessarily the same for every move, but on every move we follow some procedure), and tweak it so you don't develop bad habits. That's why I like Heisman so much, as he is the only one that emphasizes the importance of developing good thinking habits for novices. Usually the topic isn't treated until more advanced books.

    And this bears on your hypothesis--is it really that to form habits we need consciousness, or is it only to change bad habits that consciousness and feedback is required?

    Forming an initial set of habits is something you'll just do naturally anyway (I can show you generally how to swim, but then you just start doing it, build up a bunch of habits, and ten years later I come see you and you are a horrible swimmer). So I think it isn't that consciousness is required for (a good deal of) habit formation, but that it is especially important for forming good habits, such as a thought process that isn't "Impulsively make the first move that pops into your mind."

  8. Relevant link from Big Five Chess on the 'stages' of breaking a bad habit. It dovetails nicely with our discussion.

  9. Blue,
    ok, I can be wrong. Let me assume you are right. Then there are a few questions that have to be answered.

    Question 1:
    How is it named what I can learn in just 7-15 repetitions if it is not called a habit? For instance the example that I gave with the word "mayhem". After looking it up a few times I now know what it means. If I see or hear the word immediately an unconscious process starts that translates the word.

    Question 2:
    My argument for the need of consciousness is based on the following observation:
    Our choir doesn't learn something when the members have a passive attitude while we learn fast with an active attitude.

    Can you explain this without the need of consciousness?

    Question 3:
    Besides that, the argument itself was initially inspired by Baars. Or do I misunderstand Baars?

  10. OK, assuming I am right (and frankly I don't assume I am right, but for purposes of discussion):

    Question 1:
    You can't use one example to make a general hypothesis (confirmation bias). We have a specialized, genetically entrenches processor, a built-in language module, so learning language is probably not the best example, especially not something that should be used to make a general point. Acquired skills like tennis are a counterexample suggesting things are more complicated. How many times do you practice forehand before you do it well?

    Yes, of course you can learn do to any old forehand in two trials. But that is not habit formation. Leave you alone doing forehands for a few weeks or months, and you will then have forehand habits (that are probably bad).

    Question 2:
    You are just talking about the need to correct bad habits. Your habit, the natural state, is a bad performance. To correct them consciousness is needed. This is like the swimmer who, without consciousness, develops a bunch of bad habits, naturally goes to a path of least resistance which feels "natural" but is suboptimal.

    Question 3:
    Just to review Baars a little for those who don't know him. The conscious global workspace contains information from (one or many) modules, and broadcasts it to all of the modules. Information is put in the conscious workspace if it is likely to be relevant for a good number of modules, just as you will send an email out to all the specialists at the company if it is going to be helpful for a good number of specialists (modules, remember are extreme specialists--one takes care of linguistic grammar, another object color, etc).

    His view is that 'procedural memory' is like another module that receives inputs from the global workspace.

    He discusses this in some detail here. Basically, habits (procedures) are built up via reinforcement learning, which is enhanced when the outcomes/inputs are conscious (so, you see person X, say 'Hi', they smile, and you are more likely to say 'Hi' again in the future when you are conscious of seeing the person and their response than if you didn't notice it).

    So you are right that Baars thinks consciousness is helpful for gaining procedural memories (which I have never contested).

    But this doesn't imply anything about the speed with which habits are learned, and doesn't imply that habits can't be learned without consciousness, just that it makes things move along faster.

    An interesting implication is that when procedures (habits) are ingrained, it often hurts to have the workspace send something out to it, as it interferes with its smooth operation (e.g., thinking about form of a tennis swing actually hurts the swing in an expert--they do better when not consciously thinking about how they are doing it).

    While I like Baars, clearly we can't take everything he says as the truth--it's just a model. However, I think his model is consistent with everything we are both saying.

  11. More on question 3: Just to give an example with more detail on how we can form habits without conscious knowledge.

    I can teach you how to swim with a couple of vague tips (get on your stomach, move your arms like windmills out of phase, and kick). I then walk away and come back in 20 years. You have all sorts of habits in your swimming that were never reinforced, that were never conscious. E.g., the way your hands enter the water on each stroke is now a habit, but something you literally never thought of. The way your body twists around its axis, etc. Hence, not all procedural knowledge/habits are the result of explicit learning.

    But I can make them conscious as an instructor, I can teach them to you, correct you, which will be key to change the habits. Most people don't stop habitual drug use without some conscious effort :)

    Similarly, we develop all sorts of bad chess habits without thinking about it. I learn the rules of the game from my dad, he leaves and comes back in 20 years and he's like "Wow, why did you just trade your rook for his bishop on move 3?" All sorts of subtle nonverbalized habits that we aren't conscious of are happening.

  12. Ok, fair enough.

    In an old post there is a scientific article which ends as follows:

    Lead researcher Ongjen Amidzic said that amateurs never encode in long-term memory the information they cull over the years from a chessboard. Why remains a subject for additional study, he said.

    This suggests that no matter how often an amateur repeats the same things, he will not form habits. Much unlike a grandmaster.

    Or does a grandmaster distill good habits from the same repetitions where an amateur forms bad habits?

  13. Or does a grandmaster distill good habits from the same repetitions where an amateur forms bad habits?

    Exactly, though they are by no means doing the same repetitions! E.g., the crap player has awful habits, never checking for simple tactics, making the first move that pops into his head. The GM stopped doing that early in their training likely.

    A GM isn't made just by playing a lot. Every GM has had formal instruction, almost every one of them in the form of a live coach, and all of them in the form of books, and now computers, postmorte ms, etc.. Especially going over their own games.

    I was able to become a better swimmer just by applying common sense to the stroke. Some of that common sense was wrong, but analyzing my stroke is like a patzer analyzing his or her own games. It's probably better than nothing, but they could end up with kooky ideas (e.g., bishops are worth the same as rooks). So, even a lone stranger analyzing his own games would improve more than someone simply playing a lot.

    Since the true patzer never gets feedback about his bad habits, they stay bad. Like the swimmer that has an awful stroke after 20 years.

  14. From the article of my previous comment:

    He however explained that with each new move, amateurs must focus on analysing what essentially is new information. While experts are able to retrieve "chunks" of long-term memory while relying on brain circuitry outside the medial temporal lobe.

    The brainscan revealed that an amateur does something different than a grandmaster. If they both worked from habit then the brainscans would show activity in the same areas. According to this the amateur doesn't create habits. He treats the positions as if they are new to him. Why doesn't he build habits from the repetitions while the grandmaster does?

  15. A good question (incidentally, do you have a link to the study?). I would need to see the details to answer. E.g., where they doing games or puzzles? Were they matched for time spent on chess, or did the masters have a significant amount more of time spent on chess in their life? What, specifically, were the instructions given to the subjects?

    Clearly a master has tons of patterns/intuition/chunks of chess knowledge tucked away to use as procedure rather than something that must be thought through consciously. So if the positions shown in the study simply accessed such habit procedures, then they didn't need to consciously think about the position as much. What if we gave them much harder positions, as hard for them as the positions were for the amateur?

    Or, what if the problems were extremely simple for both. E.g., a series of simple back-rank mates would probably show similar results for club players and masters.

    Like pro versus amateur singers comparing 'Mary had a little lamb' versus some really hard opera.

    Again, I'd need to see more details on the study to see if these comments apply.

    The amateur has tons of habits (at the most basic, he knows without thinking how the pieces move, but also habits in the way he thinks about the position (e.g., he looks at whether he can give check, and gives it, that's a really bad habit), habits in the way he scans the board, pieces he prefers, depth of opening knowledge that is habitual rather than declarative in form), simple tactics. The master has much better habits (at every level more breadth and depth of knowledge encoded in procedural memory).

    Another thing that might be a habit is to attack each position as if it required a great deal of explicit, conscious, time-consuming, analysis-heavy thought. Masters may be much more comfortable, given that they know how much they implicitly know, to simply relax and let their habits take over for them, using conscious deliberation only in tough, strange positions.

  16. PS That smiling goat is freaky. You have the greatest pics.

  17. Tempo: The hippocampal structures activated in amateurs are important for memory consolidation (declarative, not procedural). Such structures are less important for conscious thought and procedural memory. The frontal cortex, where GMs surprisingly had more activity, is more important for active conscious attentive thinking which surprisingly, is where the GMs had more activity.

    The function of gamma-frequency oscillations is a subject of some controversy. Some like to link it to consciousness, but that is speculative.

    It suggests the GMs were actually thinking more though in a way that didn't involve the structures for forming new memories (i.e., they were relying on the procedures).

    The amateurs, on the other hand, likely also used procedural memories (no difference in basal ganglia, suggesting similar for both groups), had less conscious thought (becaues of lower frontal cortex activation, perhaps signalling they were relying too much on procedural memory because of a bad thought process?). Amateur brain structures that make them more likely to form long-term declarative memories of the games were also activated.

    At any rate, this neural study is very hard to extrapolate to more interesting psychological/functional issues that we've been discussing (as is usually the case with such coarse-grained neuronal studies).

    There are controls that might be important: perhaps GMs were more relaxed than amateurs. GMs probably have much more experience with chess, and don't treat random games like this as learning experiences, so their hippocampal structures are less likely to be activated. In general very hard to interpret unequivocally especially as bears on the stuff we've been discussing.

  18. So it is probably best not to lean too heavily on theories about this subject (nor scientific nor homebrewed) but to use my own experiments as a guide. Which is my approach anyway.

  19. I agree, prefering experiments over hypotheses. But I prefer to find real experiments (i.e., N>1, with statistics) to cut through the problem of the anecdote.

    In particular I like experiments specifically geared toward chess psychology. So, things like de Groot, that study of the help of explanations for learning, the recent interesting (and important, I think) study of confirmation bias in chess amateurs.

    Psychology looks directly at the stuff we are most interested in. Neuroscience looks indirectly at what we are interested in by looking at the brain, and then speculating about what this means for psychology.

  20. I wouldn't mind to be the subject of a successful anecdote which is scientificly unclear, a statistical aberration and not reproducible:)

  21. Yes, like MDLM! But I also enjoy the study of the science of chess improvement. Perhaps once I am a professor somewhere, I will do some studies.