Appearance
2.3 — Attention, and the Real Price of Switching
Put on headphones. Different speech plays into each ear, and you are told to repeat aloud what comes into the right ear, word for word, as it arrives. This is called shadowing, and it is hard enough to occupy you completely.
Afterwards, you are asked what was said in the left ear. People cannot report a single word. They cannot say what language it was in, or usually whether it changed from words to nonsense partway through. They can report whether the voice was male or female, and whether it stopped.
Then the experimenter says your name into the ignored ear. About a third of people notice immediately, and the shadowing stumbles.
Both halves of that result are the shape of attention. Almost everything unattended is discarded before it becomes experience — but something is still scanning it, and a small number of things get through the gate. This page is what that gate is, what it costs to move it, and how to stop losing hours to it.
What attention is for
You take in far more sensory information per second than you could possibly act on. Every neural system downstream of the senses is a bottleneck, and the bottleneck is not a design flaw — it is the point. Attention is the mechanism that decides which small fraction of available information gets full processing.
It has three distinguishable jobs, and it helps enormously to know which one is failing when something goes wrong.
Alerting — being generally awake and ready. This one is mostly physiological. Sleep, caffeine, time of day, illness. If this is what is broken, no technique helps and you should stop reading about focus and fix your sleep.
Orienting — pointing the spotlight at a particular place or thing. This can be pulled from outside, by a bang or a movement or a notification, which is fast and automatic. Or it can be pushed from inside, by your own intention, which is slower and effortful and is what "concentrating" feels like.
Executive control — resolving conflict, keeping the goal alive when something else is more attractive, noticing that you have drifted and coming back. This is the prefrontal cortex from 2.1, and it is the part that degrades first when you are tired, stressed, or three hours into something hard.
The two ways your attention gets moved
Bottom-up capture is what happens to you. A loud noise, a sudden movement in peripheral vision, your name, a face, anything moving towards you. It is fast, it is involuntary, and it evolved because the things that grabbed our ancestors' attention involuntarily were disproportionately the things that could eat them.
Top-down control is what you do. Holding a goal in mind, keeping the spotlight where it is useful, actively suppressing what is not.
Every notification on your devices is engineered to trigger the first system, because the first system does not have to negotiate with your intentions. Motion, sound, a red badge with a number, a face — these are not arbitrary design choices. They are the specific stimulus categories that capture attention automatically, and they are aimed at a mechanism you cannot switch off by deciding to.
This is why willpower is the wrong tool. You are not weak-willed for looking at a phone that buzzed. You are a normally functioning organism responding to a stimulus category selected precisely because it defeats intention, and the correct response is to change the environment rather than to try harder. 8.3 makes this a method.
Multitasking, and what it actually costs
The plain finding, replicated many times: for two tasks that both need conscious attention, you are not doing them at once. You are switching, and the switch is not free.
The cost has three components.
Switch cost. Every time you change task, performance on the new one is slower and more error-prone for a period, because the goal, the rules and the relevant information have to be reloaded. In laboratory tasks this is a few hundred milliseconds. In real work, where the task is "write this analysis", the reload is far longer, because what has to be rebuilt is a whole mental model.
Attention residue. Part of your mind stays on the previous task, especially if it was unfinished. Research on this effect finds measurably worse performance on task B when task A was interrupted rather than completed — which is why a two-minute glance at an unresolved email costs far more than two minutes.
The interruption tail. Observational studies of office work have found that after an interruption people frequently do not return directly to the original task at all, taking a detour through two or three others first, with total recovery times often measured in tens of minutes. The commonly cited figure of around twenty-three minutes comes from this line of work; treat the exact number as approximate and the phenomenon as solid.
What can genuinely be combined: one task that needs attention plus one that is fully automatic. Walking and talking. Driving a familiar route and holding a conversation — though even that degrades under difficulty, which is why people stop talking when they are looking for a turning. Listening to instrumental music while doing routine work is usually fine. Listening to a podcast while reading is not doing either.
And the group that is worst at it are the ones most confident about it. People who habitually run several media streams at once perform worse on tests of filtering distraction and switching. Whether heavy switching causes the deficit or people with the deficit are drawn to switching is not settled. Either way, the self-report is not to be trusted.
The other bottleneck: working memory
Attention selects. Working memory is where the selected material is held while you do something with it — a mental workbench, not a store.
The famous number is Miller's seven plus or minus two, from 1956. Later work with cleaner methods puts the real capacity at about four items for unrelated material, and it is worse than that sounds, because those four slots must also hold your goal and the intermediate results of whatever you are doing.
Two properties matter practically.
It is destroyed by interruption. Not degraded — emptied. This is why being interrupted mid-calculation means starting the calculation again, and why the phrase "where was I?" exists in every language.
Chunking expands it enormously. Four items, but an item can be anything you already know as a unit. The sequence 1 9 4 7 1 9 9 1 2 0 0 8 is twelve items for a stranger and three for you. Expertise is largely the accumulation of larger chunks — a chess master looks at a board and sees six meaningful structures where a beginner sees twenty-five pieces. This is the mechanism behind why learning gets easier the more you know, and it is why Part 3 puts so much weight on building organised knowledge rather than collecting facts.
Mind-wandering, which is not the enemy
Sampling studies that ping people at random through the day find minds off-task somewhere in the region of a third to a half of waking hours. That is not a modern affliction and it is not a defect.
The default mode network — a set of regions that becomes more active when you are not focused on the outside world — is doing real work: planning, rehearsing conversations, connecting things, working on problems you set down hours ago. The insight in the shower is this system delivering.
The distinction worth drawing is between wandering and rumination. Wandering is loose, moves around, and sometimes returns something. Rumination is a locked loop on the same distressing content with no progress, and it is one of the strongest predictors of depression there is. 5.5 is entirely about the second one and how to break it.
So: unstructured time with no input is productive, and it is the first thing eliminated by a phone. The walk without headphones is not wasted time.
What to do with this
Remove the trigger, do not resist it. Phone in another room, not face down on the desk — the effect of mere presence is documented and being able to reach it is what matters. Notifications off by default, with a short whitelist for humans who genuinely need you. Every technique that relies on noticing an urge and overriding it is competing with a system optimised to defeat exactly that, and losing a hundred times a day is expensive even when you win most of them.
Work in protected blocks and defend the start. Sixty to ninety minutes with nothing arriving is the unit that produces work. The first ten minutes are the most fragile, because that is when the model is being loaded and there is nothing yet to lose by abandoning it.
Park the state before you break. Write one line saying exactly where you are and what the next action is. Working memory will be empty when you return; this line is the restore point, and it converts a twenty-minute climb into a two-minute one. This is the single cheapest technique on the page.
Close the loops that are pulling. An unfinished small thing occupies attention out of all proportion to its size. Either do it, or write it down somewhere you trust — the writing genuinely releases the attention, which is most of why external task lists work at all.
Separate the three failures before you fix anything. Cannot stay awake is alerting: sleep, light, movement, and no technique will substitute. Keep getting pulled away is orienting: environment. Keep drifting off with nothing pulling you is executive control: usually the task is too vague, and the fix is to make the next action concrete enough that there is no thinking to avoid.
And take the unstructured time on purpose. One walk without headphones. The problem you cannot solve at the desk is frequently solved by the system that only runs when you stop.
Next: 2.4 covers what happens to the material that got through the gate — how memory is stored, why every act of remembering rewrites the thing remembered, and why total confidence in a vivid memory means almost nothing.