Julia Shaw on False Memory, Eyewitness Testimony, and the Psychology of Evil
Julia Shaw, a criminal psychologist at UCL known for her research on false memory, argues that human memory does not record experience — it reconstructs it on demand, so reliably that a person can be led, in three short interviews, to believe in vivid, confident detail that they committed a crime that never happened. The conversation ranges across the psychology of evil, murder, and relationships, but its scientific core is memory: how false memories form, why eyewitness testimony is far less reliable than confidence suggests, and why she calls generative AI ‘the ultimate false memory machine’.
Key ideas
- Every autobiographical memory is false to some degree — the question is only how false. Shaw distinguishes ‘gist’ memory (the general shape of an event, which humans retain well) from ‘verbatim’ memory (specific sensory detail, which humans retain poorly). Courts and interrogations demand verbatim precision from a system built only to deliver gist, which is where memory most reliably breaks.
- Shaw’s own doctoral research implanted complete false memories of committing a crime. Using a structured protocol — rapport-building with a true memory, then guided imagination and social validation applied to a fabricated one — roughly 70% of one study’s participants came to report detailed, confident false memories of assaulting or stealing from someone, a finding now among the most cited in forensic psychology.
- Generative AI reproduces the exact structure that makes memory implantation possible. Because chat-based AI is a tailored, socially validating conversational partner that tells people what they want to hear, Shaw argues it recreates — at scale, and without the safeguards used in her own ethics-reviewed protocol — the same conditions under which she implants false memories experimentally.
- ‘Evil’ is a continuum of ordinary traits, not a category of monstrous people. The dark tetrad (psychopathy, sadism, narcissism, Machiavellianism) are dimensional traits everyone carries at some level; Shaw argues the label ‘evil’ functions socially to end inquiry rather than start it, and that dehumanisation and de-individuation — not innate monstrousness — are what let ordinary people commit atrocities.
- The single most protective habit against memory distortion is contemporaneous, individual record-keeping. Shaw’s practical advice — for witnesses, investigators, and intelligence officers alike — is never to trust memory to hold a detail: write an account down immediately and alone, before any group discussion, because collective recall can just as easily manufacture a shared false memory as recover a true one.
Content
Evil as a continuum: the dark tetrad
Shaw’s book Evil: The Science Behind Humanity’s Dark Side opens the conversation. She describes the dark tetrad — psychopathy, sadism, narcissism, and Machiavellianism — as four dimensional traits, not diagnostic categories: everyone scores somewhere on each, and only a person who scores high across all four is genuinely at elevated risk of harming others. She resists the word ‘evil’ itself, arguing that it functions as a conversation-ending label — once someone is called evil, the listener is licensed to stop asking why they act as they do, because ‘I would never do such things; I am good.’ Dismantling that artificial line between good and evil, she argues, is the precondition for actually reducing harm, because prevention requires understanding the psychological and social levers that produced the behaviour.
Two mechanisms recur through her account of how ordinary people commit atrocities: dehumanisation (treating the other side as less than human) and de-individuation (submerging personal identity into a group, so that ‘us versus them’ displaces individual judgement). She cites survey research suggesting that murder fantasies are common — roughly 70% of men and over half of women in the studies she references — and argues, following researchers like Philip Zimbardo, that rehearsing such fantasies in imagination is protective rather than dangerous: fully imagining the consequences of an act is what lets deliberate reasoning override the impulse. Zimbardo’s Stanford Prison Experiment (now heavily criticised methodologically, but historically influential) and his concept of the ‘heroic imagination’ — consciously rehearsing intervention rather than bystanding — anchor her broader claim that most people are capable of both extremes, and that which one shows up depends heavily on circumstance, not fixed character.
The psychology of violent crime
Shaw’s discussion of serial offenders and murder is deliberately anti-sensational. She argues that loneliness is a recurring feature of serial killers’ profiles — not merely as motive, but because social isolation removes the correcting presence of other people who would otherwise perform ‘reality monitoring’: catching a person’s drift into a delusional or radicalised interpretation of events before it hardens. The same mechanism, she argues, underlies both psychotic symptoms and ideological radicalisation.
Most real murders, she stresses, are not premeditated; they are arguments and fights that escalate over trivial stakes, not the calculated work of a scheming mastermind. She attributes the public appetite for the opposite narrative to what she calls the victimization gap: the consequences for a victim (irreversible) and for a perpetrator (at most, imprisonment) are so mismatched that society compensates by preferring stories in which the perpetrator’s culpability is total and exotic. She cites a low recidivism rate for homicide (around 1–3%) against much higher rates for fraud and sexual violence, and argues that sentencing driven by a folk sense of just deserts is often poorly calibrated to what would actually make society safer.
Constructing false memories of committing a crime
The conversation’s scientific centre is Shaw’s PhD research, which combined two previously separate lines of work — Elizabeth Loftus’s research on false memories and Saul Kassin’s research on false confessions — into a single question: could a person be led not just to falsely confess to a crime, but to genuinely remember, with confidence and sensory detail, committing one that never happened?
Her protocol was procedural and carefully staged. Participants (recruited with their parents’ cooperation, to source real biographical detail and rule out any genuine history of the target events) first spent twenty minutes discussing a true childhood memory, to establish rapport and a template. The interviewer then introduced a second, fabricated memory — for example, an assault with a weapon at age fourteen that led to police contact — using the same structure. Techniques included the ‘illusion of transparency’ (acting as though the participant’s cooperation was already assumed) and guided imagination (‘close your eyes and picture what could have happened’), with every detail the participant supplied validated and encouraged. Because the interviewer genuinely knew nothing about the participant’s real life, any specific detail in the resulting account — who was hurt, where, how — had to come from the participant’s own imagination, woven from real people, places, and emotions into an event that never occurred. Roughly 70% of participants in this specific sample came to report detailed, confident false memories; the study was stopped early once its effect size became clear.
Shaw is careful, in the conversation, to caveat the finding: the 70% figure describes one sample and six specific fabricated events, not a general law of how easily any memory can be implanted, and she notes an ongoing scientific dispute over how to draw the line between a ‘false memory’ (something a person believes they experienced, with the subjective quality of a real recollection) and a ‘false belief’ (something a person merely thinks might have happened). She now uses the finding practically: training police interviewers, military and intelligence personnel, and investigators at the International Criminal Court on protocols that avoid leading and suggestive questioning, and on the importance of recording an account immediately and individually — because once a group compares notes, later recall reliably reflects the most recent shared version, not the original one, without anyone noticing the drift.
The everyday version of this same reconstructive process, she argues, is state-dependent memory: retrieval is easier for memories that match a person’s current emotional state, which produces both an ‘embarrassment spiral’ (dwelling on one humiliation surfaces others) and a general rosy reminiscence bias in how people look back on their lives. The corrective — cognitive restructuring — is a deliberate technique for changing a memory’s emotional interpretation (‘what did this teach me?’) without denying that the underlying event happened.
AI, memory, and the case for a scientifically grounded interview protocol
Shaw extends her research directly to generative AI, which she calls ‘the ultimate false memory machine’. A chat-based AI system is, structurally, a tailored conversational partner that tells a person what it predicts they want to hear, largely uncritically — the same social configuration she uses to implant false memories experimentally, but running continuously, at scale, without an ethics board. She argues the distortion runs in both directions: the AI can shape a person’s account of their own past, while a person’s leading or suggestive prompting can in turn shape what the AI itself represents as fact, with neither side reliably correcting the other.
Her proposed response is not to avoid the technology but to encode the cognitive interview — the decades-old, empirically validated, non-leading protocol used in UK police training — directly into AI systems, since a fixed script is exactly the kind of task a language model can execute faithfully. She co-founded a company, Spot, originally to build a machine-administered version of the cognitive interview for workplace incident reporting; she argues the same approach could extend to recording important personal memories more broadly, and expresses frustration that AI labs building conversational systems rarely consult the social scientists who understand how these interactions distort memory and belief.
A note on aphantasia
An unplanned aside surfaces a genuine individual-differences finding: Shaw recently discovered she has aphantasia — the inability to generate mental imagery. Asked to picture a red apple, she reports seeing only black. This explains, she suggests, her lifelong inability to use memory-palace-style mnemonic techniques, which rely on constructing vivid mental images, and she hypothesises (without strong evidence, by her own admission) that aphantasia may correlate with more analytical or concept-driven modes of thought.
See also
- Julia Shaw — speaker
- Lex Fridman — host
- Charan Ranganath on Human Memory, False Memories, and How the Brain Reconstructs the Past — the wiki’s companion memory-science episode; Ranganath supplies the neuroscience of reconstruction, Shaw the applied interviewing mechanism
- False Memory — concept page created from this source