When Memory Lies

In 2013, a team at MIT gave a mouse a memory of being shocked in a room where nothing had ever happened to it. Susumu Tonegawa’s lab let mice explore a box and tagged the hippocampal cells that fired during the visit with a light-sensitive protein, one that only switched on in active neurons. Days later the mice were put in a different box and given a mild foot shock, but only while a laser reactivated the cells tagged in the first box. Put back in that first room, where nothing bad had ever happened, the mice froze in fear (Ramirez et al., 2013). Nothing outside their heads had fooled them. Their own neurons, encoding a real visit, had been stitched into an event that never took place.

People do a slower, cruder version of the same thing. Since Frederic Bartlett’s 1932 studies of how people retell stories (the primer on the reliability of memory covers the background), researchers have known that we don’t replay the past so much as rebuild it from fragments, filling the gaps with whatever fits. Usually that’s harmless. Sometimes it isn’t. Confident, mistaken eyewitness memories have helped convict people of crimes they didn’t commit (Loftus, 2005). A mega-analysis of eight memory-implantation studies classed about 30 percent of participants as having formed a false memory of a childhood event the researchers had made up, through suggestive interviewing alone (Scoboria et al., 2017). Stricter definitions give lower figures, but none of them gives zero. In the lab it takes almost nothing. Show someone a list like bed, rest, awake, tired, all related to a word that never appears, sleep, and later they will recognize sleep as confidently as words that really were on the list (Roediger and McDermott, 1995).

The brain leaves a trail while it does this. Diffusion imaging links accurate recall to the integrity of the inferior longitudinal fasciculus, a white-matter tract that carries visual and semantic detail, and links susceptibility to false memory to a different tract, the superior longitudinal fasciculus, which connects frontal and parietal regions involved in attention and top-down integration (Fuentemilla et al., 2009; Janelle et al., 2022). Thicker gray matter in parts of the parietal and occipital lobes goes with more false memories; thicker middle frontal gyri and occipital poles go the other way.

Scans taken during recall add more. False recognition raises activity in the dorsolateral prefrontal cortex and the anterior cingulate, regions tied to monitoring and resolving conflict, as if the brain were working harder to rule on a memory it isn’t sure of (Abe et al., 2008; Kurkela and Dennis, 2016). That reading is reasonable but not the only one. Both regions also respond to plain difficulty, so the extra activity may only mean false items are harder to sort, not that a special checking process has switched on (Poldrack, 2006). What is clearer is what’s missing. A true memory reactivates early visual cortex, a faint echo of the original seeing. A false one mostly doesn’t, even when the person feels just as sure (Slotnick and Schacter, 2004).

The body seems to notice the difference even when the mind doesn’t say so. Skin conductance and heart-rate responses are weaker for falsely recognized items than for real ones (Baioui et al., 2012), an involuntary signal that never reaches conscious report.

Underneath all of it is the machinery that lets memories form at all: long-term potentiation, NMDA receptor signaling, CREB, and plasticity proteins like BDNF and PKMζ. The processes that bind the pieces of a real memory together can bind pieces that never belonged together. Stress makes this worse in a specific way. Under high stress, cortisol acting on the hippocampus pushes encoding toward the gist and away from precise detail (Finsterwald and Alberini, 2014; Kim et al., 2015), which is part of why a traumatic memory can be vivid and, in its particulars, wrong.

None of this makes the mice’s fear less real to them, and none of it makes a mistaken witness less certain on the stand. Courtrooms still tend to treat a confident memory as something close to a recording, even after decades of research showing that remembering is reconstruction, and that the process which lets anyone “recognize” a word they never read is, at bottom, the same kind of process a laser exploited to convince a mouse it had been hurt in a room where nothing had ever touched it.

Editor’s note: the word-list effect described above isn’t just something to read about: you can run it on yourself. Take the false memory test.

Editor’s note: this essay is based on a research poster from a summer research program at Vanderbilt University.