Chapter 1 · A Narrow Working Memory and a Broad Long-Term Memory

Picture a moment when you are given a phone number for the first time and have no pen to write it down. In the few seconds it takes to roll the number around in your mouth and pull out your phone, if someone speaks to you, the number scatters at once. You have to ask again. Yet the home phone number you memorized as a child does not vanish from your head even after years. Even if it does not come to you sharply this instant, the number is still there somewhere. One number cannot survive a few seconds; the other lasts for years. That two things happening in the same head should meet such different fates is because the places they reside in are different.

This book begins by dividing the head into two stores. One is where whatever is being processed at this moment lingers briefly; the other is where what has once settled in piles up for a long time. A just-heard phone number floats briefly in the first and evaporates; an old number is lodged in the second and stays. The two places are opposite in their characteristics—in size, in speed, and in what they hold and for how long.

A Narrow, Volatile Working Memory

The front place, where what is being processed resides, is called working memory. It is worth likening to a workbench where the actual work of the head is carried out. Everything—calculating, comparing, connecting—happens on top of it. Working memory has limits of two kinds. One is how much it can hold at once; the other is how much it can process at once on top of what it holds. Taking the holding side first: working memory is narrow, and it is volatile.

First, it is narrow. What you can lay on it at once is a handful at most. Try multiplying 17 by 24 in your head and you feel it right away. You multiply 17 by 4 to get 68, then 17 by 20 to get 340, and then you have to add the two—but while you compute 340 holding onto the 68 you got first, the 68 blurs, and you stall at the final addition. A calculation trivial on paper is taxing in the head because you must hold several intermediate results at once, and those few run straight into working memory's limit. Exactly how many it can hold depends on what you are holding, but that the number is small enough to count on your fingers is clear.

On top of that, it is volatile. What you place in working memory disappears within seconds unless you actively hold onto it. The earlier phone number scattering at a single spoken word is this. It stays while you roll the number around in your mouth, and the moment you stop rolling it and your attention goes elsewhere, it empties out. What is in working memory is not maintained by being left alone; it is maintained only by being held continuously. Stop holding it, and it is gone.

When these two faces of the holding side combine, working memory becomes a place that is narrow and also leaks. The amount it can hold is small, and what it holds keeps slipping away.

The limit on the processing side has a different grain. Working memory has a ceiling not only on how much it holds but on how much it can turn over at once on top of what it holds. The holding limit and the processing limit are separate.

From these limits comes the bottleneck of nearly everything we do in our heads. And yet the head's computation happens only in this narrow working memory. Understanding, reasoning, building something new—these occur nowhere else but on top of it.

A Broad, Long-Lasting Long-Term Memory

The back place is long-term memory. It is worth likening to a library's stacks where what has been encoded is shelved, and it is the opposite of working memory in every respect.

First, it is broad. We recognize thousands of faces, know tens of thousands of words, and can name a tune from just two notes of its melody. All of this is held in long-term memory, and it never happens that an old memory must be erased to make room for a new one. Unlike working memory, which fills up with a handful, long-term memory never fills and overflows. Its capacity limit can be taken as effectively none.

Next, it lasts. Just as a childhood phone number remains after years, what has once truly settled in lasts a long time. Compared with what is in working memory, which cannot survive a few seconds, the very unit of time is different.

One thing, though, resembles working memory: being held in long-term memory does not mean you can always pull it out at once. Everyone has had the experience of a name they clearly know merely hovering on the tip of the tongue without coming. That name has not vanished from long-term memory—it surfaces a little later, or all at once at a single small cue. It is held, but the path to reach it does not open well. How well you can reach what is in long-term memory at this moment grows dim as time passes and similar things intrude. What remains in long-term memory and what you can reach right now are separate questions.

The Same Unit the Two Places Handle

Though the two stores are so different in character, the goods they pass between them are of the same kind. Those goods are called representations. A representation is one chunk, one unit, of the information handled in the head. What rises into working memory, what piles up in long-term memory, and what passes between them are all representations.

A unit does not mean a fixed amount of information. A single letter such as c is one representation; the single word cat is one representation; and a single scene—the persimmon tree in the yard of a country house in late autumn—can be one representation too. The amounts of information the three carry differ enormously, yet in working memory they each take one place. This is why a string of unfamiliar letters is taxing to memorize while a single familiar word is held with ease. How much information one compresses into the same one place differs from person to person and with skill. How this compression happens, and why an expert handles far more than a novice with the same working memory, is dug into later.

The Division of Labor Between Computation and Storage

The relationship between the two stores comes down to a division of labor. Computation happens in working memory, and its results remain in long-term memory. While you solve 17 times 24, the multiplying and adding take place in working memory, but if the answer 408 you obtained, or the method of solving, lasts, that is what has been recorded in long-term memory.

So what passes between the two becomes the skeleton of what happens in the head. What is built in working memory crosses over to be recorded in long-term memory (writing), and what is held in long-term memory is summoned back into working memory (reading). To learn something is, in the end, for computation carried out in working memory to change the state of long-term memory. For a just-heard phone number to settle into long-term memory like a childhood number rather than scatter, exactly what must happen between working memory and long-term memory?