Memory Is Not a Recording. It Is a Reconstruction.

Imagine that you are on a street one evening when a violent confrontation breaks out a few metres in front of you. It lasts perhaps ten seconds. Someone shouts, a man in a dark jacket runs off, a bang rings out, and for a moment you catch sight of his face.

An hour later, you are sitting in a quiet room describing what you saw. You are convinced that you remember the event. You recall the jacket. His height. The direction in which he ran. Perhaps even the expression on his face. If someone asked whether you were telling the truth, the answer would be simple: of course.

And yet there is a much larger gap between the statements “I am not lying” and “I remember it accurately” than we intuitively feel.

Modern research on memory and eyewitness identification has long shown that human perception and subsequent recall have systematic limits. In its extensive report on eyewitnesses, the National Academies of the United States summarises that accurate identification depends on several steps — a person must first perceive, encode, retain and later retrieve the event correctly — and an error can arise at any of these stages.¹

I saw it. So I know it.

Our intuition works with a simple model: the eyes are a lens, the brain is storage, and a memory is a file. An event happens, it is saved, and later we “open” it.

It is a useful metaphor. But it is wrong.

A camera does not need to understand the meaning of a recording when it plays it back. Memory does. A person does not store every detail of a scene at the same quality. Attention highlights some information and barely processes other information. What we already know affects what we notice. Emotion changes the priority of stimuli. And when we later recall an event, the brain has to build a coherent whole from the traces that are available.

That does not mean memories are worthless or that “nothing can be trusted.” It means something less dramatic but more important: the reliability of memory depends on the conditions under which information was formed and later recalled.

The brain dislikes empty spaces

When we recall an event, we do not work only with what we saw at the time. The meaning we assigned to the scene, later information, conversations, expectations and the very way a question is phrased all enter the picture.

Psychology uses the term misinformation effect for one of these phenomena: a person may later mistakenly recall misleading information received after the original event instead of the correct detail from the original experience.²

This matters especially because a memory may not feel uncertain from the inside. A new detail can fit into the story so smoothly that a person feels no difference between what they actually saw and what entered their later recollection by another route.

One word, a different crash

One of the most famous experiments in this area was published in 1974 by Elizabeth Loftus and John Palmer. Participants watched film footage of traffic accidents and then answered a question about the vehicles’ speed. The critical part of the question differed by a single verb.³

One group heard wording equivalent to “how fast were the cars going when they hit each other,” while another heard a word evoking a much more violent collision. Average estimates differed systematically according to the verb used. In the original English wording, the average ranged from 31,8 mph with the mildest verb, contacted, to 40,8 mph with smashed.³

The second experiment was even more interesting. Participants watched the accident again. Some were asked a question containing smashed, some one containing hit, and the control group was not asked about speed at all. A week later, they were asked whether they had seen broken glass — even though there was no broken glass in the footage.

“Yes” was answered by 16 of 50 people in the smashed group, 7 of 50 in the hit group and 6 of 50 in the control group.³ The authors interpreted this result as support for the idea that subsequent information can influence a later memory — not merely an immediate numerical estimate.

Why it matters in criminal investigations

In everyday life, a false memory may end in an argument over who left the keys on the table. In criminal proceedings, an error may be tied to the identification of a particular person.

That is precisely why the intuitive belief in “camera memory” is dangerous here. A witness may appear convincing, speak consistently and be emotionally authentic. None of that, by itself, proves that the original perception was accurate.

In a 2014 consensus report, the National Academies drew attention to the inherent limits of human vision and memory and recommended standardising identification procedures to minimise the unintentional influence of witnesses.¹ This is an important shift in thinking: the problem is not a “bad witness.” The problem may arise from the interaction between the person, the environment and the procedure used to elicit the memory.

These figures need to be read correctly. They are not the share of all wrongful convictions in the United States, and certainly not a statistic for the Czech Republic. They come from a specific dataset of cases maintained by the Innocence Project, but they clearly show that eyewitness-identification errors are not merely a laboratory curiosity.⁴

“I’m sure.” What exactly does that prove?

A witness’s confidence has a powerful effect on us. When someone says “maybe it was him,” we automatically pay attention. When they say “I recognise him one hundred percent,” we tend to regard the matter as much more solid.

But confidence is a psychological state. Accuracy is a property of an answer. The two can be related, but they are not identical.

What is more, confidence is not necessarily stable. After an identification, it may be influenced by the reactions of others, information about the investigation or repeated recall. From a forensic perspective, it is therefore important not only how confident the witness is, but also when that confidence was recorded and what happened between the original event and the testimony.

Contradictions are not simple either

It is easy to make the opposite mistake and say that a witness is unreliable when they get a detail wrong. That does not automatically follow either.

Memories are not binary — correct or false. A person may remember the central part of an event accurately and confuse a peripheral detail. They may know what happened but misjudge how long it lasted. They may describe the sequence correctly while getting the colour of an object wrong.

The important forensic question is therefore not merely “does the testimony contain an error?” A much better question is: “what type of information is this, under what conditions did it arise, and what alternative sources for its content exist?”

That is the difference between mechanically judging a witness and analysing evidence.

We use the same mechanism every day

The reconstructive nature of memory is not a disorder reserved for courtrooms. It is how the normal human mind works.

After years, two partners may sincerely remember the same argument differently. Siblings may have different memories of a holiday they took together. Colleagues may be certain that a sentence was spoken at a meeting, when each of them reconstructed it from the general meaning of the conversation.

Our knowledge helps the brain fill in the world. Without this ability, we would be overwhelmed by details. Yet the same mechanism that lets us quickly understand an incomplete situation also means that “it makes sense” is not the same as “that is exactly how it happened.”

What this means for a good investigation

If we accept that memory is sensitive to context, the logical goal is not to listen to witnesses less. Quite the opposite. We need to gather their information in a way that leaves as little room as possible for unintentional distortion.

That means separating the witness’s own spontaneous memory from information introduced by the question. Do not put a detail into the question that the witness did not provide. Record the first account as faithfully as possible. Limit prompting during identification — including unconscious prompting. And always track the timeline of what the witness could have learned after the event.

This is not distrust of the person. It is respect for how human cognition works.

The most dangerous witness may not be a liar

It is much easier to defend ourselves against someone we know is lying. We look for a motive, a contradiction, evidence that disproves their claim.

The more difficult situation is when someone decent, convincing and completely certain stands before us. Their voice does not tremble with uncertainty. They are not making anything up. In their own experience, the memory is real.

That is precisely why it is important not to treat sincerity and accuracy as synonyms.

In a courtroom, during an investigation or at the kitchen table at home, we might benefit from one small change in the question we ask. Instead of “do you believe them?”, sometimes ask:

Because memory is not a camera. But neither is it an enemy. It is an extraordinarily capable system that continually builds a meaningful world from incomplete traces.

And sometimes its greatest strength is also its greatest weakness.