Appearance
2.4 — Memory, and Why Remembering Is Rebuilding
In 1974 Elizabeth Loftus showed people a film of a car accident and then asked how fast the cars were going when they hit each other. Other groups got the identical film and one different word: smashed, collided, bumped, contacted.
The estimates moved with the verb. "Smashed" produced speeds around 41 mph; "contacted" produced around 32.
A week later, with no film shown again, everyone was asked whether they had seen broken glass. There was no broken glass in the film. The people who had been asked the "smashed" question were more than twice as likely to remember seeing it.
One word, asked afterwards, changed what people remembered witnessing. Not what they said — what they sincerely recalled. This page is about why that is not an oddity but the normal operation of the system.
Memory is not storage
The intuitive model is a recording: an event happens, a copy is filed, remembering retrieves the copy. Everything about that is wrong, and the errors it produces are the reason so many arguments between people who were both there go nowhere.
What is actually stored is a pattern of connection strengths across the cells that were active during the event — a partial trace, not a copy. Remembering runs the pattern again, and the gaps are filled in with what is plausible, exactly as the blind spot is filled in 2.2. The result feels seamless because you have no access to which parts came from the trace and which from the filling.
Three stages, each of which can fail on its own.
Encoding — getting it in. Most failures called "bad memory" are encoding failures, which is to say attention failures. You did not forget the person's name; you never encoded it, because you were thinking about what to say next.
Consolidation — stabilising it over hours to years, heavily dependent on sleep. A memory is fragile for a period after encoding and gradually becomes durable.
Retrieval — getting it back, which depends enormously on cues. Almost everything you cannot recall is still there and is simply not being cued. The tip-of-the-tongue state, where you know the word has three syllables and starts with a hard sound, is retrieval partially succeeding — proof the trace is intact and access is failing.
The kinds of memory, which are genuinely different systems
Episodic memory holds events, with a time and a place, and it is the branch that feels like remembering. Semantic memory holds facts stripped of when you learned them — you know Brasília is the capital of Brazil and have no idea when that entered your head.
Implicit memory is everything you know how to do without being able to describe it. Riding a bicycle, typing, the physical act of writing your signature. It is stored elsewhere, it degrades far more slowly, and it survives damage that erases the whole explicit branch.
The patient known as H.M. had both hippocampi removed in 1953 to control severe epilepsy, and could not form new episodic memories afterwards. He was taught to trace a shape in a mirror over successive days. His performance improved steadily, exactly like a healthy learner — while he insisted each day that he had never done the task before. Two systems, one intact and one destroyed, in the same head.
This matters personally for one reason above all. Emotional conditioning is implicit. The fear response to a raised voice, the tightening at a certain tone, the dread when a particular name appears on your phone — these are stored in a system that does not know what year it is and cannot be reasoned with directly. Which is why insight does not dissolve a trigger (5.3), and why the treatments that work for phobia and trauma work by re-running the situation rather than by explanation (14.3, 14.4).
Rebuilding, and the window it opens
Here is the finding that overturns the recording model completely. When you retrieve a memory, it becomes temporarily unstable, and it has to be re-stored. During that window it can be changed, and what gets written back is the version you just experienced — including whatever context, mood and suggestion were present during the recall.
This is called reconsolidation, and its consequences are enormous.
A memory you recall often is not better preserved. It is more edited. Each retrieval is an opportunity for drift, so the memories you are most confident about, having told the story fifty times, are frequently the least faithful. You are remembering the last telling.
The story you tell yourself about your childhood, your last relationship, or the argument last month is being rewritten by the telling, in the direction of the mood you are in and the point you are making. This is not dishonesty. It is the mechanism.
And it is a therapeutic opportunity, which is why 14.4's treatment for trauma involves deliberately bringing the memory up in a safe present — the re-storing writes back a version tagged with a room where nothing bad is happening.
False memories are ordinary, not exotic
Loftus and others went well beyond changing details. In a series of studies, participants were given a few true events from their childhood, supplied by family, plus one invented one — being lost in a shopping mall at age five, spilling punch at a wedding, a hot-air balloon ride that never happened. After a couple of interviews in which they were encouraged to try to remember, a substantial minority of people — figures across such studies commonly land somewhere between a fifth and a third — came to report the false event, often adding vivid details nobody had supplied.
They were not lying. Confidence was high. Emotion was present.
Two things follow, and both matter in real life.
Confidence does not track accuracy. People report identical certainty for true and implanted memories. The intuition that a vivid, detailed, emotionally strong memory must be accurate is simply false, and courts have convicted people on it. Even flashbulb memories — where you were when you heard shocking news — have been shown to drift substantially within a year while confidence stays high.
Repeated suggestive questioning creates memories. The mechanism is the same one that makes recall useful: imagining something in detail lays down a trace that later cannot be distinguished from a perceptual one. This is why the way children are interviewed after an alleged incident is a serious technical matter, and why leading questions are restricted in evidence.
Nothing here means memory is generally unreliable. Most of what you remember happened. It means memory is a reconstruction with no internal marker for which parts are reconstructed, and that a disagreement about the past between two honest people is the expected outcome rather than a sign that one is lying — a fact worth holding onto in Part 12.
Why you forget, which is mostly useful
Decay — unused traces weaken. Ebbinghaus mapped the curve in 1885 by memorising nonsense syllables on himself: steep loss in the first hours and days, then a long flattening. 3.2 turns that curve into the most valuable technique in learning.
Interference — similar memories compete. Where you parked today is hard to recall precisely because you have parked in that car park two hundred times. Nothing was lost; the cue points at too many things.
Retrieval failure — the commonest kind. The trace exists, the cue is missing. This is why memory is better in the room you learned in, why a smell can unlock something a photograph cannot, and why "tell me about the rest of the evening" recovers what "what was his name?" cannot.
And motivated forgetting is real but modest. Repression in the strong Freudian sense — a whole traumatic event sealed away and later recovered intact — has poor evidence and is one of the most contested claims in the field. What does have evidence is that people can suppress retrieval of specific unwanted material with practice, and that it partially works. Note also that trauma more commonly produces the opposite problem: intrusive memory that will not stay away (14.4).
What to do with this
To remember something, do something with it. Depth of processing beats repetition by a wide margin. Reading a definition four times is nearly worthless; explaining it in your own words once is worth far more, because the explaining is a different operation that builds retrieval routes. This is the whole basis of Part 3.
Give it a cue you will actually have later. Memory is cue-dependent, so store things where the cue will be present: the keys by the door, the medicine by the toothbrush, the note in the file you will open. This beats intending to remember by an enormous margin, because intending to remember does not create a cue.
Sleep after learning anything that matters. Consolidation happens then, and the effect is large enough to be visible in a single night.
Stop treating your memory as evidence in disputes. Two useful sentences, and they will save you real damage over a lifetime: "My memory is that…" and "I might be wrong about the detail — what do you remember?" You are not conceding the argument. You are describing the instrument accurately. 12.7 shows how much conflict this defuses, and 13.3 shows the one situation where it is dangerous — someone deliberately exploiting the fact that memories can be rewritten in order to overwrite yours.
And write things down at the time, not later. A contemporaneous note beats a confident recollection every single time, and the gap grows with each retelling. This is the whole reason for the prediction diary in 1.4 and the trigger log in 5.3: the version of you writing today is a better witness than the version of you remembering in March.
Next: 2.5 moves from the contents to the processing — the two speeds your mind runs at, why the fast one is right most of the time, and how to tell when you are in a situation where it will be confidently wrong.