Chunking: How a Skill Is Built

Hand a beginner a twelve-step recipe and they'll dart between the stove and the text, forget whether they've salted it yet, and burn the onions in the end. Hand the same recipe to a chef and they'll glance at it once and start cooking without looking back.

The difference isn't memory or talent. In the chef's head that same recipe takes up not twelve items but two or three. Here's how that happened.

The prep

In a professional kitchen nobody thinks "chop the onion, chop the carrot, heat the oil, sauté the onion until translucent, add the carrot, sweat for five minutes." The chef thinks one word: the sauté. It's a single operation, one item in the head, one movement of attention — even though inside it's still a dozen actions.

Such a glued-together block is called a chunk. Once it was a set of separate steps, each demanding effort. Then repetition fused the pieces, and now the brain handles them as a single whole.

Here's where it gets interesting: the small preps stack into bigger ones. Sauté + stock + spices = a soup base. Soup base + protein + a side = a dish. Dish + dish + balanced flavours = a menu. A chef holding a twenty-item menu in mind isn't remembering a thousand actions — they're remembering a dozen blocks, each of which unfolds automatically on the inside.

Every skill works this way. A programmer doesn't think "declare a variable, set a counter, check a condition" — they think "a loop." A pianist doesn't think about eight separate fingers — they think "a chord." A doctor doesn't run down a checklist of symptoms — they see the picture.

Why it's the only way

Working memory — the memory that holds what you're juggling right now — has an embarrassingly small capacity. George Miller's classic estimate (1956) was "the magical number seven, plus or minus two," but modern research, above all the work of Nelson Cowan, gives an even more modest figure: about four items at once, if you don't help yourself with rehearsal or notes.

Four. That's everything you operate with in the moment.

Which raises a question: how, then, does a chess player hold a whole game in mind, or a surgeon an operation? William Chase and Herbert Simon answered it in 1973, in an experiment that became a classic. Masters and novices were shown a chess position for a few seconds and asked to reproduce it from memory. The masters did incomparably better — you'd think they simply had extraordinary memory.

But then the pieces were placed on the board at random, with no chess logic at all. And the masters' advantage almost evaporated: they remembered about as much as the novices.

The conclusion is ironclad: it's not about memory capacity. The master sees not thirty separate pieces but five or six familiar configurations — "castled king," "isolated pawn," "a pin." Five or six items fit comfortably into four-plus slots. A random arrangement doesn't add up to any familiar block — and the master ends up exactly where the novice is, with the same four cells.

Chunks don't expand working memory. They enlarge what you put into it.

What a chunk is built from

For pieces to fuse into a block, three things are needed. Miss one and you get not a chunk but clutter in your head.

Undivided attention. As long as attention is torn between the text, a notification, and a video in the background, the fusing physically doesn't happen: connections between neurons form only where attention is held. This, by the way, explains the familiar feeling of "read for an hour, don't remember a line" — the hour was there, the attention wasn't.

Understanding the structure. A chunk has to be not a sequence of movements but a meaningful unit: why the steps go in this exact order, what breaks if you swap them. Without that you get a "dead" block — a set of actions that falls apart at the slightest change in conditions. That's exactly why a crammed formula doesn't help on an exam where the problem is phrased a little differently.

Practice with context. Knowing how to do it is half. The other half is knowing when to apply it. A prep you can cook but can't place in a dish is useless. Context is built only by solving different problems and choosing the tool yourself, rather than being handed it in the prompt.

Why some preps take longer to build

A recipe for borscht and a recipe for the derivative aren't tasks of equal difficulty, and it isn't about the subject.

Borscht has something the derivative doesn't: it exists physically. You see the onion, you smell it, you burn yourself on the pot. For millions of years the brain trained to remember exactly that — objects you can see, pick up and taste. Every step of the recipe already has a support waiting in your experience.

An abstract concept has no such support. You can't smell a derivative, you can't shift a set with your hand, you can't point a finger at a pointer in memory. The only thing such material catches on is other abstract concepts, and then only if those are already mastered. Hence an honest conclusion worth accepting in advance: maths, physics and programming need more passes than subjects about things. The feeling "this comes harder to me, so I have no aptitude" most often means simply "there's nothing here to catch on the first time round."

Two consequences follow. First: abstract material needs concrete examples especially badly — they work as the support the concept itself lacks. Second: a gap in the foundations hits abstract subjects harder than any others. If a basic concept hasn't been assembled into a block, the next level has nothing to stand on — and in chemistry or programming there is nothing else to stand on in principle.

What happens with repetition

It's worth saying outright what has been implied so far.

The first pass through new material is the most expensive: you have to hold everything at once, and working memory fills instantly. The second pass is cheaper. By the fifth, some pieces demand no attention at all.

This isn't habituation or a "familiarity effect." Connections that carry a signal are physically strengthened by regular use, while unused ones weaken. Every repetition isn't merely reminding yourself of the material but work on the structure itself: it becomes sturdier and takes less effort to hold.

The practical meaning lies in how you read your own sensations. If a topic comes hard the first time round, that's the normal state of a first pass, not a diagnosis. If it comes hard on the fifth — that's a signal the passes were done wrong: most likely you were rereading instead of reconstructing it yourself.

How to use other people's solutions properly

Reading someone else's recipe and looking at a photo of the finished dish is pleasant. A clear feeling arises: I get it, I could do this. That feeling deceives almost everyone.

But the conclusion isn't "worked examples are useless" — it's subtler than that.

When you have no blocks of your own yet, solving a problem from scratch overloads the workbench: you have to hold the problem, an unfamiliar method, and your own floundering all at once. John Sweller and colleagues showed that at this stage a novice advances faster by studying worked solutions than by unsuccessfully storming problems alone. The example works as a first borrowed block, on loan until you've built your own.

The trap isn't in the examples but in two things.

The first — watching the steps rather than the links. While working through a solution it's easy to follow what is being done and miss why this, and why here. But the steps are exactly the part you won't need later: in a new problem they'll be different. Only the logic of the transition transfers. A useful question for every line of someone else's solution: why is the next step this one, and what suggested choosing it.

The second — staying on examples. As experience grows, a worked solution turns from a support into a brake: what helped the novice now hinders the expert — this is the well-documented expertise reversal effect. So the support has to be withdrawn gradually: first a complete worked example, then an example with steps left out for you to fill in, then a problem with no support at all.

A chunk appears only at the last stage — when you've walked the path yourself, without prompts, and preferably several times, stumbling. Mistakes here aren't a side effect but part of the assembly: they show where the block hasn't fused yet.

The formula worth taking from this article:

You've understood the recipe — you're not a chef yet.

The gap between "I understand the explanation" and "I can do it myself" is the most expensive illusion in learning. An entire next article is devoted to it.

The two directions of assembly

Blocks are built from both ends at once, and you have to work in both directions.

Bottom-up — through practice and repetition. An individual prep is strengthened to automaticity: you stop thinking about how to make it, and it frees space on the workbench.

Top-down — through the whole picture. You see where this prep lives among the others, which dishes it goes into, how it differs from its neighbour. That's what gives you context — knowing when the tool is used and when it isn't.

One direction alone is never enough. A polished block with no picture of the whole is a skill with nowhere to apply it: you can make the sauté but don't know which soup to put it in. A picture of the whole with no polished blocks is a beautiful menu with nobody to cook it: you understand how it all fits together, but every step takes effort.

The practical translation is simple: a five-minute overview of the structure before you start is the top-down direction, working through specific problems is the bottom-up one. Skipping either doubles the time the other takes.

In practice

Table of contents first, text second. Before diving into a new topic, skim the whole thing: headings, diagrams, conclusions — five minutes. That gives you a frame to hang the details on later. The idea isn't new: back in the 1960s David Ausubel showed that a preliminary overview — an "anchor" — markedly improves how material is absorbed. Without a frame, details pile up rather than form a structure.

Close the solution and redo it yourself. Worked through an example — close it and do it from scratch. Didn't manage — peek, then close again. Repeat until you get through without a single prompt. That's the only moment a chunk actually assembles.

Compress what you've covered into one sentence. After each block of material, state its essence in one sentence in your own words. Can't? Then there's no chunk yet — just a scatter of facts.

Vary the problems, don't repeat one. Ten same-type examples in a row create the illusion of mastery: you already know which tool to grab, because you're grabbing it for the tenth time. Mixed problems force you to choose — and that's exactly what builds context.

Check yourself

Close the article and answer in your own words:

Why did chess masters lose their advantage on a random arrangement of pieces, and what does that prove?
Do chunks increase working-memory capacity, or do they do something else?
The three conditions for building a chunk — what are they, and what breaks if you remove any one?
Why doesn't studying someone else's solution create a skill?
Why do abstract subjects need more passes than subjects about physical things?
How does the difficulty of a first pass differ from the difficulty of a fifth, and what does the latter tell you?
Why is studying a worked solution more useful to a novice than storming the problem alone — and at what point does that stop being true?
What should you be looking at in someone else's solution besides the steps themselves?
What does the bottom-up direction give you, what does the top-down one give you, and why isn't one enough?

In short

A chunk is a block fused by practice that the brain handles as a single whole: a dozen actions turn into one "word."
Working memory holds about four items. Chunks don't expand it — they enlarge the contents of its cells.
The Chase and Simon experiment: mastery is a library of familiar blocks, not extraordinary memory.
Building one needs three things: whole attention, understanding of structure, and practice with varied context.
Abstract material has nothing in your experience to catch on, so it needs more passes and more concrete examples. Difficulty here isn't a sign of inability.
Repetition physically strengthens the connection: the first pass is the most expensive, after which the structure holds itself. A hard fifth pass means the passes were done wrong.
Worked solutions are the right start while you have no blocks of your own: they take the load off working memory. The mistake is watching the steps instead of the links between them, and not withdrawing the support as you grow.
A block is built from two sides: practice from below and the whole picture from above. Without the first there's nothing to work with; without the second it's unclear when to apply it.
You've understood the recipe — you're not a chef yet. Skill appears where you walk the path yourself and make mistakes.