Remembering What Never Happened

AFew beliefs about the mind are held as firmly as the assumption that a recollection reported with confidence and furnished with vivid detail must be broadly accurate. The assumption survives partly because it is useful, since courts, clinics and families all need some rule for deciding whose account to accept, and confidence is the only signal available to people who cannot check the facts elsewhere. Laboratory work carried out since the closing decades of the last century has made that rule difficult to defend. Presented with a list of associated words such as bed, rest, tired and dream, and asked shortly afterwards which items they had heard, participants report the unpresented word sleep almost as often as words that were genuinely on the list, and they report it with equal conviction. What made this unsettling was not the error itself, which is trivial, but the discovery that no introspective feature distinguished the manufactured item from the genuine ones. Nothing felt wrong. If confidence tracks accuracy so poorly under conditions this simple, the argument runs, there is little reason to trust it in the far messier circumstances where it is normally relied upon.

BThe account that now dominates the field treats such errors as a by-product of something the mind does well rather than as evidence of a defective system. Ingrid Sollberger, of the Larkfield Institute of Cognitive Science, argues that what survives an experience is rarely its surface form but its gist, meaning the theme, the emotional tone and the general shape of what occurred, because gist is what remains useful once particulars have ceased to matter. A word that fits the theme of a list, or an incident that fits the theme of a childhood, is therefore liable to be produced at retrieval whether or not it was ever encountered. In a programme of studies conducted between 2009 and 2014, Sollberger's group tested 640 volunteers and found that those who scored highest on measures of thematic reasoning were also the likeliest to fall for the word-list illusion, a correlation since reproduced in four laboratories. She maintains that this is precisely what her account predicts, and that the opposite result would have embarrassed it. A memory system incapable of generating false memories, on her view, would be one that had failed to generalise, and generalisation is most of what memory is for.

CTobias Renard, a clinical psychologist at Wexbury College, is sceptical of the leap from word lists to life. He concedes that gist extraction accounts elegantly for what happens in the laboratory, but contends that the richly detailed autobiographical episodes reported in implantation studies, such as a balloon flight in childhood or a hospital visit that never took place, may not be memories at all in the sense that matters. When Renard re-interviewed participants who had accepted an invented event, asking them to separate recollecting the episode from merely believing that it had occurred, only 19 of the 42 previously classified as remembering it maintained that they could still see it. The remainder described a belief without imagery, which is nearer to accepting a plausible report from a trusted source than to remembering anything. Renard's own evidence, however, is thinner than he sometimes allows. His follow-up involved 31 undergraduates interviewed twice within a single term, all of whom were acquainted with the researchers, and he has not shown that a belief without imagery stays inert; several practitioners point out that a belief rehearsed over years may acquire imagery of its own.

DWhatever the mechanism, its reach is narrower than early commentary suggested, and the boundaries are instructive. Suggestion works mainly where the proposed event is plausible for the person concerned and where the questioner is taken to have access to information the participant lacks; remove either condition and acceptance falls sharply. A field study published in 2021, drawing on recorded interviews from eight police forces, found that suggestive questioning altered accounts in 12 per cent of cases overall, but in 42 per cent of the subset in which the interviewer had already voiced a suspicion aloud. Age matters as well, though not in the direction usually assumed: among participants over sixty, susceptibility rose for peripheral details while remaining unchanged for the central action of an event. Such qualifications rarely survive the journey into public discussion, where a demonstration that memory can be distorted is routinely reported as a demonstration that memory is unreliable in general. The two claims are not the same, and the distance between them is where most of the disagreement among specialists now lies.

EA reasonable reading of the present evidence is that both processes operate, and that the useful question is no longer which is correct but in what proportion each contributes under specified conditions. Sollberger's account handles the laboratory data better than any rival and has the further advantage of explaining why the errors are systematic rather than random. Renard's objection retains its force in precisely the settings where the stakes are highest, since a courtroom or a consulting room seldom lacks either of the two ingredients his mechanism requires. What neither side has produced is a means of telling, in an individual case and without external corroboration, which kind of memory one is dealing with. Until such a means exists, the practical advice offered by researchers on both sides is unexpectedly similar: treat the manner in which an account was elicited as at least as informative as the account itself. Confidence, they agree, was never the right place to look.