Ranganath, Charan

memory researcher, UC Davis; author of Why We Remember

Charan Ranganath: Human Memory, Imagination, Deja Vu, and False Memories

Lex Fridman Podcast · May 25, 2024

• Charan Ranganath
• human memory
• episodic vs semantic memory
• false memories
• déjà vu
• forgetting
• imagination and memory
• ADHD and attention
• fMRI methodology
• memory and AI

Core ideas

Ranganath opens by reframing memory around Daniel Kahneman's distinction between the "experiencing self" (who lives moment to moment) and the "remembering self" (who curates a narrative), arguing that since the remembering self is what actually guides future decisions and sense of identity, it's rational to deliberately seek out memorable — even momentarily unpleasant — experiences, because their long-term "story value" outweighs the discomfort of living them. This sets up his broader thesis: memory isn't a videotape of the past but a predictive machine, constantly building and updating an internal model of the world so it can forecast what happens next, with prediction errors (moments that violate expectations) driving learning and marking what gets encoded. He explains childhood/infantile amnesia through this lens — the hippocampus is still developing, cortical plasticity is running at a massive rate, and the sense of self needed to organize autobiographical memory is immature — and extends this into a "stages of life" theory: childhood is an exploratory, no-fixed-goal mode suited to an immature prefrontal cortex; young and middle adulthood shift into goal-directed, prefrontal-dominant behavior; and old age pivots toward transmitting accumulated semantic knowledge to younger generations, illustrated by orca pods led by post-reproductive grandmothers who carry the group's ecological knowledge.

On forgetting, he rejects the idea of memory as simple decay, describing it instead as competition among distributed neural memory pools, missing or weak retrieval cues, and insufficient neuromodulator-driven consolidation — and situates episodic memory within Alan Baddeley's working-memory architecture and semantic memory, using event boundaries (moments like a birthday party's transitions) as points where hippocampal connectivity bursts and marks optimal encoding opportunities. He's skeptical of most popular memory hacks — Ebbinghaus's roughly 60%-loss-in-a-day forgetting curve remains essentially unbeaten, and he's dubious of general "brain training" and attention-training claims (citing both Amishi Jha's mindfulness research and Adam Gazzaley's video-game-based training as promising but limited) — while endorsing techniques with real mechanistic backing: the Memory Palace/method of loci (which works via distinctiveness, like a memorable Post-it note among blank ones), using stories and songs as mnemonic scaffolds, spaced repetition (tied to a computational model of context-dependent memory decontextualization), and the "testing effect" (retrieval practice), which he likens to backpropagation — repeatedly generating and correcting output strengthens the underlying representation more than passive re-exposure does.

A recurring theme is the deep entanglement of memory and imagination. Drawing on Marcia Johnson's reality-monitoring research and Frederic Bartlett's framing of remembering as "imaginative construction," he explains that remembering and imagining the future share overlapping default-mode-network machinery, and that amnesia patients characteristically struggle to imagine future scenarios, not just recall past ones. He discusses Solomon Shereshevsky (via Alexander Luria) — a mnemonist whose synesthesia-driven imagery was so vivid it eventually blurred the line between imagination and reality, which Ranganath speculatively links toward psychotic-like experience — and reports that memory-competition athletes (per Joshua Foer's Moonwalking with Einstein) are consistently made through training rather than born with innate gift. On the neuroscience side, he walks through fMRI methodology in real depth: the BOLD signal's reliance on the differential magnetic properties of oxygenated versus deoxygenated hemoglobin, the roughly six-second hemodynamic lag, an emerging theory that astrocyte glutamate-clearing metabolism (not neuronal spiking directly) may underlie the signal, and the technique's practical limitations. He describes his lab's multivariate "pattern" analyses — decoding neural signatures that distinguish memory for a person's identity from memory for a location — and a newer cross-species, cross-method research program (via lab member Tyler Brown) unifying rodent electrophysiology, human fMRI/intracranial recordings, and computational vision/language models through shared "state space" analysis.

On déjà vu, he traces the phenomenon from Wilder Penfield and Hughlings Jackson's classic epilepsy-stimulation case studies to Anne Cleary's modern work using matched-but-differently-themed VR environments that reliably trigger déjà vu without any conscious recognition of the underlying structural match — supporting a "fluency"/partial-match theory of the sensation, illustrated by the way rare words (like "rutabaga") versus common ones (like "apple") produce different feelings of familiarity. On false memories, he insists on abandoning any true/false binary in favor of viewing memories as blends of fact and inference, discussing Elizabeth Loftus's misinformation-effect studies, the social "contagion" of shared memories, the 2000 McCain/Bush South Carolina primary push-poll as a case study in manufactured false belief, and — pivoting to collective/cultural memory — Viet Thanh Nguyen's account of the asymmetry between the Vietnamese "American War" and the American "Vietnam War" as competing national narratives, alongside research showing that diverse, non-dominant-narrator group discussion produces more accurate collective memory than homogeneous groups anchored to a single loud voice. A long, dark digression covers false confessions and coercive interrogation: he lays out a "recipe" — feed someone the desired information, physically and psychologically stress them, apply an authority figure's pressure or offer a face-saving narrative — that reliably produces false memories of committing crimes, as documented in CIA enhanced-interrogation research and wrongful-conviction cases.

The conversation closes on a wide-ranging set of adjacent topics. On technology and AI, he argues that building genuine episodic-plus-semantic memory into AI is "superficially not that hard, but at a deeper level very, very hard," invoking Steve Grossberg's Stability-Plasticity Dilemma (when should a system overwrite old knowledge versus preserve it while learning new exceptions) and stressing that human memory's power comes from flexibly combining episodic memory (specific past events) with semantic memory (generalized knowledge) to handle novel situations without erasing prior learning — a capacity current AI systems lack, along with the survival-driven motivational and emotional machinery (fight-or-flight, reward) that shapes what humans prioritize remembering in the first place. He connects this to ADHD, discussing his own probable undiagnosed case from childhood and describing the condition as fundamentally an attentional flexibility/stability imbalance (too-easy distraction paired with difficulty disengaging from absorbing tasks) rather than a core memory deficit — people with ADHD remember what interests them vividly and forget what doesn't — and shares personal strategies (timeboxing email vs. writing vs. "goofing off" time) for managing it, echoing Cal Newport's warnings about task-switching costs. He also discusses his life as a musician (trumpet, then a late shift to punk/post-punk guitar after seeing Sonic Youth break musical "rules," playing "math rock" and unconventional time signatures in his band Pavlov's Dogs, whose members are mostly memory researchers), reflects on nostalgia's double edge (a coined-as-"disease" term that can bring genuine happiness or curdle into toxic idealization of "the good old days"), and on how monotonous, undifferentiated stretches of time (like pandemic lockdown) compress subjectively because too few event boundaries are created to anchor memory. He ends on mortality, his own post-50 awareness of time's asymmetry, comparisons of Buddhist/Stoic detachment against the Bhagavad Gita's alternative of acting fully while renouncing attachment to outcomes, and a closing meditation on the beauty of the brain's "internal model" — the invisible, unifying structure that turns disconnected sensory fragments into a coherent, navigable sense of self and world.

Excerpts