Can you trust a memory?

In a 2015 study, university students were interviewed three times about their adolescence, with a real event supplied by their parents and a false one, some of them criminal, described as though the parents had reported it too.

The authors classified 70 per cent of those given a crime as having a false memory of committing it. Other researchers recoded the same transcripts with schemes that separate believing an event from remembering it, and put the figure at 26 to 30 per cent: many had accepted the story and speculated about it without showing signs of remembering. How to score a false memory is still unresolved.

The interviews combined memory-retrieval techniques, including imagining the event, with what the critics called “a heavy dose of social demand”.

Even the lower figure matters where memory is evidence, as the critics said themselves. That whole false memories can be produced is not in doubt. How often is.

The method is older than that study. In 1995 a smaller experiment, with 24 adults, gave people short accounts of childhood events supplied by a family member, one of which, being lost in a shopping mall, had not happened. About a quarter were judged to remember it, fully or in part. A preregistered replication in 2023, five times the size, judged 35 per cent to remember it; 14 per cent said so themselves.

People who come to report an implanted memory hold it, on average, less firmly than their real ones, and rate it lower for clarity and confidence. Not all of them: in the two pooled studies that asked, 28 per cent of 116 people given a false event put their belief that it had happened at the top of the scale, and observers watching the accounts often cannot tell them from true ones.

Confidence at a first, fair test is another matter. In eyewitness lineups, one group of researchers finds that a witness who picks someone with high confidence at that point is likely to be right; another points to conditions in which confident first picks are wrong. Both agree that confidence expressed in a courtroom is a poor guide. Confidence can be added afterwards: repeated testing raises it, and so does telling a witness they picked the right person.

On the long-standing view, which its defenders say still holds, everyday recall preserves the general shape of the past well. The experiments above deliberately target the conditions under which it does not: distant events, repeated suggestive questioning, a trusted source supplying the story.

How far the laboratory rates transfer to a police interview, a therapy room or a courtroom is argued. One side says the studies overstate how easily false memories form; the other, that brief and relatively mild laboratory suggestion understates what longer, stronger suggestion in therapy can do. Neither has data from outside the laboratory to settle it.

Both work from the same account of how a false memory forms: a suggestion is imagined, and the imagining is taken for remembering.

Studies have looked for differences in brain activity, in language, in detail and in hesitancy between true and false memories. Differences exist on average.

None of them separates a particular memory reliably enough to be used on one person, which is the use a court would need.