The Narrow Gate
New information reaches long-term memory only through working memory — about four chunks at a time — so attention and chunking set the ceiling on encoding. · 11 min
Last folio split learning into two acts. This folio is about the doorway to the first act. Before anything reaches long-term memory, it must pass through working memory — the small holding space for whatever you are thinking about right now. That space is astonishingly small. It empties within seconds unless you keep refreshing it. And nothing you fail to attend to enters it at all. The size of that space, and the one honest trick for packing more through it, set the ceiling on what you can encode.
Guess before you learn
Twelve random letters appear for a few seconds — K, Q, V, H, Z, R, T, L, B, M, X, D — then vanish. How many can a typical adult write back in the correct order?
About six — and that is with single, familiar letters. George Miller's famous 1956 estimate was seven plus or minus two items; Nelson Cowan's later review put the truer limit for new, ungrouped material at about four chunks. If you guessed ten or twelve, you are in good company: working memory feels roomier than it is, because with familiar material it quietly packs several items into one. That packing has a name — chunking — and it is the second half of this folio.
9–12
3–5
Your memory has two rooms. A small front room holds what you are thinking about right now — only about four things fit, and they fade in seconds. A giant back room holds everything you know, for years.
Everything must pass through the small room first. But you can beat the limit by grouping: 1, 9, 4, 5 is four things — the year 1945 is one thing. Grouping four things into one leaves room for three more.
6–8
Working memory is where thinking happens: the handful of items you are holding and using right now. Its limits are strict. Unrehearsed, items fade within about twenty seconds, and capacity is roughly four chunks — a chunk being any pattern so familiar it counts as a single unit. F, B, I is three chunks to a young child and one chunk — FBI — to you.
Two consequences follow. First, attention decides everything: what you do not attend to never enters working memory, so it cannot be encoded at all. Second, chunking is the only honest way past the limit. You cannot widen the gate — you can only pack more into each thing that passes through it.
9–12
George Miller's 1956 estimate — seven plus or minus two — measured items like digits, which adults already chunk. Nelson Cowan's 2001 review, controlling for rehearsal and grouping, put the purer capacity near four chunks. The unit is the chunk, not the item: a chess master glancing at a real mid-game position for five seconds can replace about sixteen pieces; a novice manages about four.
Show both players a random scatter of pieces, though, and the master falls to the novice's level. The advantage was never raw capacity but a library of familiar patterns, each worth one chunk. Expertise does not enlarge working memory; it enlarges what one chunk can hold. That is Chase and Simon's 1973 finding, and it repeats in every skilled domain tested.
K–2
Your hands can carry only a few blocks at a time. Your toy box holds hundreds. Your mind works the same way: a tiny carrying space, and a huge keeping space.
Snap four small blocks into one toy, and your hands can carry more. Grouping things makes them easier to carry. Your mind can group too.
Undergrad
Baddeley and Hitch's multicomponent model divides working memory into a phonological loop, a visuospatial sketchpad, and a central executive allocating attention between them — later joined by an episodic buffer that binds the pieces. Measured capacity depends on which component carries the load and how much refreshing the executive can spare.
Cowan's embedded-processes alternative treats working memory as the activated portion of long-term memory, with a focus of attention limited to about four chunks. Sweller's cognitive load theory draws the instructional consequence: learning fails when intrinsic load — how many elements must interact at once — plus extraneous load — how badly the material is presented — exceeds capacity. Good teaching manages the load it cannot remove.
Postgrad
The measurement itself is contested. Cowan's estimate of about four comes from paradigms that block rehearsal and grouping; visual work in the Luck and Vogel tradition supports discrete slots, while Bays and colleagues model a continuous resource spread thinner as array size grows. What survives every model is a severe, stable central limit.
Ericsson and Kintsch's long-term working memory explains how experts appear to evade it: retrieval structures in long-term memory let a chunk act as a pointer to arbitrarily rich contents — their runner S.F. pushed digit span from seven to seventy-nine by mapping digit groups onto race times. The educational corollary: element interactivity, not information volume, is what overwhelms a novice.
working memory
The small holding space for whatever you are thinking about right now — roughly four chunks, fading within seconds unless refreshed.
chunk
Any pattern familiar enough to count as one unit in working memory. FBI is one chunk to you and three to a young child.
Now watch the trick work. Try to hold these fourteen letters: N, A, S, A, F, B, I, Y, M, C, A, D, N, A. As letters, they overflow the gate three times over. Regroup them — NASA, FBI, YMCA, DNA — and they pass in one glance. Nothing about the letters changed; the unit changed. This is why experts seem to have supernatural memory inside their own field and perfectly ordinary memory outside it: years of practice built large chunks, and each chunk still costs one slot.
So attention and chunking set the ceiling on encoding — act one from last folio. Nothing unattended enters. Nothing enters faster than about four chunks at a time. And the only lever you own is the richness of the chunks: every subject you study is, in part, the slow manufacture of bigger units. Next folio we follow a memory that made it through the gate — and measure, hour by hour, how it fades.
Note
When a page here feels crowded, turn the Depth Dial — the same idea, younger or deeper — down one register. Fewer new elements at once is not remedial; it is load management.
Practice — new ink and old, interleaved
1.Used properly — answer produced before the card is flipped — flashcards mainly exercise:
2.Thirty minutes remain and you have already read the chapter twice. Which plan does the evidence favor?
3.Match each term to its meaning.
4.You skim a page while texting, close the book, and can recall nothing. In the language of the two acts, which act failed?
5.Without looking back: how large is working memory for new material, and what is the one honest way past the limit?
About four chunks, fading within seconds; the way past is chunking — packing familiar patterns into single units so each costs only one slot.
How close were you? Grade yourself honestly — it sets your review date.
6.Which redesign most helps a beginner follow a twelve-step recipe?
7.Order the journey of one new fact, first to last.
- Attention selects it out of everything around you
- It is held among a few chunks in working memory
- It is encoded into long-term memory
- A cue retrieves it back into working memory