Cognitive Load and Chess Training

I put together a post today on my blog that I think is actually pretty good for once :)

http://kingpatzer.wordpress.com/2013/02/13/cognitive-load-and-a-new-chess-training-paradigm/

reposted here:

There is an idea in cognitive psychology (the study of how people think) that says humans are only conscious of the information currently being held and processed in working memory and are essentially oblivious to all the information stored in their long term memory. When given new information, working memory is quite limited in capacity and duration. Experiments show that working memory can handle about seven chunks of information. When processing information, that is doing something other than just trying to remember it, the number drops to 2 or 3 chunks of information that we can hold in working memory at any one time.

Long term memory is different. It is essentially limitless. Information in long term memory is stored in domain-specific structures that researchers call schemas. Schemas categorize information according to how it will be used, so it becomes available later for related tasks. Researchers who work in cognitive load theory would contend that “human expertise comes from knowledge stored in theseschemata, not from an ability to engage in reasoning with many elements that have not been organized in long-term memory.”

This coincides very neatly with research done specifically with chess masters. Researchers have long known that masters store board positions in their memory in chunks that correspond to a collection of pieces in relationship on the baord. Proponents of intensive tactics training argue that it is “pattern recognition” that allows for players to quickly spot tactics on the board. These patterns are perfect examples of the sort of domain specific schemata CLT researchers like Merrienboer and Sweller are talking about.

Where things become interesting is that many researchers contend that the relationship between working memory and schemas stored in long term memory matters more than limitations of working memory. The reason for this is simple: schemas increase working memory capacity by giving us domain specific methods of storing information. While working memory can only hold a limited number of items at a time, the size and complexity of each item is limited only by the schemas we have access to within our long term memory.

There’s yet another trick our brains use to help us handle information effectively. This is called ‘automation.’ Schemas are not just domain specific content, they are also domain specfic functional processes. This means that information stored in schemas can be processed automatically and without conscious effort. Automation is a result of extensive practice.

One important predicted result of this understanding of the architecture of our mind is that some entirely new tasks may be impossible to complete until pre-requisiste skills have been automated because there simply is not enough working memory capacity available for learning. The impact of this is that the goal of training for expertise should be schema constrution and automation.

Cognitive Load Theory suggests that learning happens best when instructional materials align with our cognitivie architecture. “By simultaneously considering the structure of information and the cognitive architecture that allows learners to process that information, cognitive load theorists have been able to generate a unique variety of new and sometimes counterintuitive instructional designs and procedures.”  This finding is in many ways an extension of earlier instructional design theories which held that if learning is to occur, instructors must deliberately arrange the external and internal conditions of learning.

So where is all of this going?

Well, the point is that non-expert chess players need instructional guidance to help them build the schemas necessary for self-instruction. With tools like the now famous CT-ART or CHESSIMO or other tactics training tools that need is being well met for tactics. But in other areas of chess information processing, it seems quite lacking. Books can be good surrogates, but only if they have sufficient examples and explainations to allow for the student to develop their own schemata. Moreover, few books have the level of repetiion necessary for students to become intimately familiar with an idea and to internalize it.

Consider how we learn simple mathematics. From kindergarden through about 5th or 6th grade, math instruction is about working with simple numbers  doing thousands of exercises over and over again to learn basic addition, subtraction, multiplication and division. Very few really new concepts are introduced, and when they are, they are presented as simple extensions of those basic functions. This allows students to develop their own schema for understanding and internalizing these basic concepts as schemas, and doing so to the point where they are automated. What’s 2+2? See, automated!

Chess instruction really should be done in the same way. Using smaller boards, fewer pieces, and clearly showing the interations of simiple ideas thousands of times combined with clear, well written instructional material that is focused directly on the most important factors would go a long way to helping students improve.

If you open a beginning math book, you’ll see hundreds of increadibly simple problems, presented in different ways. You’ll see a section on adding 1 + 2. Then you’ll see word problems asking the student to add one and two. Then you’ll see problems adding 2 and one. And so on.

But open a chess book aimed at teaching about, well, anything, aimed at the lower end of the class ladder, and you’ll see entire chess boards with all the pieces in place. There’s just too much extraneous information that isn’t part of the schema that the student is being asked to build and automate.

Which is easier to understand:

The above diagram showing how a rook and knight can be used to create a double attack, or this diagram below more typical of beginner instructional material:

These are clearly showing basically the same idea: that a knight can move from in front of a rook to attack a nearby target, and the result of the double attack will be that the opponent will lose material so long as the targets are of high enough value in relationship to the rook and knight.

But the second example, taken from a beginner’s tactics book (and the third problem in the section on double attack), contains a huge amount of extraneous information that simply doesn’t matter to the construction of the schema that matters. We’re telling a beginning math student to look at a calculus problem, find the addition problem burried inside it, and answer the addition question. No one would try teaching math in such a manner. But we try to teach chess that way all the time.

So this is my call out to the producers of chess instructional material: make it simplier. Give us more repetitions. Focus on the essential schematic elements of information that we need in order to deal with more complex problems.

Yes, we’ll have to use the material in a real game, so at some point students will need the second example. But if cognitive load theory is correct, then the students who spends more time building schemas with simplier problems will be able to progress faster and farther into more complex problems.

Moreover, if this approach includes not merely tactical work, but positional analysis, visualization, and so forth, there is a path to a real chess curriculem rather than just a collection of unrelated books. 

OK, I'm gonna break out Bain's "Chess Tactics for Students" and do another round.  Thanks for the pep talk!  It was actually a very well written piece, so I don't mean to be flippant.  It's just a little beyond my expertise to comment meaningfully on beyond "makes sense to me".

Interesting. I gave a beginners chess book from my club to a friend, and a lot of it's diagrams were small like that and just focusing on the relevant info. Unfortunately I can't remember what it's called.

This makes some very good points.  I especially like the part I quoted above.  This is when you see a beginner annotate his game and it's filled with superficial comments he's seen from instructional books "this bishop is on a long diagonal" , "I do this to control dark squares" , "I'm looking to open a file" when the key features in that position are anything but what the student is describing.  When it's so prevalent, this can only be the fault of the training material.

So in a lot of ways I think this hits the nail on the head.  Smaller boards with only the pieces and size necessary to convey the idea and nothing more?  I think that's a fantastic idea actually.

To be fair I think this is done already in some endgame exercises... and may be another reason why endgames are a good place to start.  You really can't evaluate a middlegame position without seeing the entire board... but this:

and

Seems to be correct once again.  Usually it just takes experiance and continued dialogue with stronger players to sort out when you were keying in on the important features of a position and when you were being distracted by something unimportant.  You see this all the time when the beginner says something like "but that doubles my pawns" and the master essentially say "the doubled pawns don't matter here."

I actually thought about putting together something like this for pawns (as I felt clueless for a long time about which pawn to recapture with, when to capture, when to allow them to capture, etc).  And just have a bunch of examples that pounded the fairly simply underlying ideas over and over.

Anyway, again, I think this has some great points, and if instructors followed up on it could make some very useful training material.

Kingpatzer, thank you for these very interesting and instructive thoughts.

I've always been a big proponent of the teaching by elements method (I'm sure there is a better way to say this in English )

You find this idea of automating basic skills in many other fields like putting in golf, serve/forehand/backhand in tennis, doing scales in music, basic steps when learning various dances, etc. It makes sense, and it certainly works.

However, when it comes to teaching material, I believe repetition is somewhat more important than isolation for three reasons :

- seeing many times a similar pattern (say your discovered attack example) helps retention by repetition

- our (adult) brain seems to be able to isolate the repeated elements by itself when exposed to them in various guises - it "spots" analogies and differences between examples - this is not true for younger children and older beginners, who may benefit more from the isolated elements, at least as a first stage in learning

- learning to isolate the pattern among the "noise" of other elements is actually an important skill to develop as well, as it's required in a real game, as you've noted

So although I agree that being exposed to simple patterns is essential to chess learning, I don't think mixing them with other elements is that bad, as long as the pattern is clearly identified, immediately present (1-move exercise !), and naturally prevalent (you don't want to mix it with other more important patterns in your examples).

CT-Art and Chessimo are more or less build that way, though CT-Art includes isolation of patterns as an extra learning tool, which may be useful as the patterns are not "basic", and Chessimo suffers from a lack of systematization (examples are mixed).

IMO, a excellent learning tool for a adult beginner, would be something like Peschk@ beginners software, where you repeateadly solve very simple exercises to drill the patterns and hard-wire them.

Also, your idea that we learn better what is linked to existing schemas prompts another question : are theses schemas rather like images, or can they be a mix of mental images and language ? In the 2nd case, the creation of "meta-rules" could accelerate learning as well (ie. "castle early", associated with specific examples of early castling)

hicetnunc, while it is true that repetition is vitally important, as it is repeated exposure to the elements that allow us to build our own internal schemata, isolation can't be ignored. In every field where it has been studied, isolation + repetition yields higher performance than moving on to integrated activities + repetition early. The reason for this, if CLT is accurate, is actually contained in the theory: the germane cognitive load is the balance of working memory left over after  intrinsic ( material relevant to the schemata being constructed) and extraneous cognitive load (material that is not relevant to what is being learned) are considered. For us to learn, that is for us to create schemata that meaningfully abstract and systamitize the material for later use, we need to have left resources in our working memory for schema construction. 

Now, if we recall that research shows that only 2 to 3 elements can be held in working memory effectively when processing new material, it becomes pretty clear that significant repetitions of the most basic building block elements of a task will give us the greatest opportunity to create functional schemata that can be further abstracted and automated. 

 The idea of meta-rules for beginners is very useful indeed. Though they are in some ways extrinsic cognitive load, they can act as a surrogate for missing schemas, thereby promoting schema construction. 

However, the danger is that they will be taken as intrinsic rather than extrinsic, and become part of a schema. So such meta-rules must always be presented carefully and with enough qualifiers that the learner can recognize the need to look beyond the rules.

And it is true that instructional techniques that are highly effective for novices lose their effectiveness, and can even have negative consequences for learning, when used with more experienced learners. This is actually referenced in the literature as "expertise reversal effect."

However, the research around CLT has real impacts on instructional design research. Discovery learning, anchored instruction, problem-based learning, and constructivist learning, to name just a few, all assume that such pedagogical techniques help learners  integrate the knowledge, attitutdes, and behaviors necessary for effective performance. And ultimately enable students to transfer what is learned to novel problems encountered eslewhere. 

However, when considered from a CLT perspective, all of these approaches suffer from at least one significant flaw: working memory resources of learners, particularly novices, can quickly become overwhelmed by task complexity, and learning will suffer as a result.

Direct guidance, isolated elements, and scaffolding of material has been shown in to simply be superior. As Kirschner argued "In so far as there is evidence from controlled studies, it almost uniformly supports [CLT techniques] rather than constructivist-based minimal guidance during the instruction of novice to intermediate learners."

Isolated elements, to progressively more complex interactive elements, with significant repetition and instructional guidance simply works best. 

Now, what's interesting is what CLT researchers look at as novice to intermediate learners. The would look at the performance spread from true beginner to mastery over quintiles. The top quintile would be mastery, the next intermediate, and then beginner-intermediate and beginner would be the bottom three quintiles.  For USCF ratings, this means that anyone below about 1400 who is training using traditional techniques instead of being given designed isolated training elements is almost certainly suffering negative impacts from cognitive load during their training. 

Please dumb it down a bit guys, my sleep deprived brain can't take this. 

But seriously, thanks this is very useful stuff.

Out of interest, at what level do you guys think it becomes ok or even useful to learn with minimal instruction?

Good stuff...