Learning & Memory.

OpeningWhat memory is.

Memory is not a video recording. It is a reconstructive process that uses partial cues to assemble something that feels like a recording.

The 20th century overturned the common-sense view of memory as a storage device. The contemporary view, supported by a hundred and forty years of research from Ebbinghaus through Bartlett through Loftus through Schacter, is that memory is reconstructive, fallible, and continuously edited by retrieval.

This deck covers the history (Ebbinghaus to the multi-store model), the case studies that revealed memory's structure (H.M., K.C., the case of the missing forms), the major findings (forgetting curves, the testing effect, sleep consolidation), the seven sins (Schacter), and the contemporary picture.

Chapter IThe forgetting curve.

Hermann Ebbinghaus's 1885 monograph established that memory could be studied scientifically. He designed nonsense syllables (consonant-vowel-consonant trigrams without prior associations: ZUF, BAK, NIM) to control for prior learning, memorised lists, and tested himself at intervals.

The result: memory decay follows a predictable curve. Most forgetting happens in the first hour after learning; the rate of forgetting then slows. The curve is approximately exponential. After 24 hours, about 70% of newly learned material is gone if not rehearsed; after a week, around 80%. The retention curve flattens substantially after that — what remains tends to remain.

Ebbinghaus also discovered the spacing effect: distributed practice (study sessions spread out) produces better long-term retention than massed practice (cramming). This is the empirical foundation of all subsequent spaced-repetition systems.

Chapter IISensory, short-term, long-term.

Richard Atkinson and Richard Shiffrin's multi-store model proposed three memory stores:

Sensory memory. Very brief (about 250 ms for visual; up to 4 seconds for auditory). Holds a high-fidelity, modality-specific record of incoming perception. George Sperling's 1960 partial-report experiments demonstrated iconic (visual) memory; subsequent work demonstrated echoic (auditory) memory.

Short-term memory. Lasts ~15–30 seconds without rehearsal. Capacity famously limited — George Miller's 1956 paper "The Magical Number Seven, Plus or Minus Two" set the canonical figure, though later work suggests a smaller working capacity (3–4 items, in Cowan's revised estimate) when rehearsal is prevented. Chunking increases functional capacity: a chess master can remember a meaningful 25-piece board position because the pieces form a few high-level chunks rather than 25 individual items.

Long-term memory. Lasts indefinitely. No known capacity limit. Subdivides into multiple systems (next page).

Chapter IIIWorking memory.

Alan Baddeley and Graham Hitch (1974) argued that "short-term memory" was not a passive store but an active workspace with multiple specialised components. Their model:

Phonological loop. Holds verbal and acoustic information through subvocal rehearsal (the inner voice repeating a phone number). Capacity bounded by how much can be rehearsed in about 2 seconds. Hence longer words are harder to remember (the word-length effect), and similar-sounding words interfere with each other (the phonological-similarity effect).

Visuospatial sketchpad. Holds visual and spatial information. Independent of the phonological loop — you can do a verbal task and a spatial task simultaneously without much interference.

Central executive. Allocates attention; coordinates the slave systems; switches between tasks. Heavily implicated in dorsolateral prefrontal cortex.

Episodic buffer (added in 2000). Integrates information from the slave systems and from long-term memory into multimodal episodes.

Chapter IVH.M.

In 1953, surgeon William Beecher Scoville performed a bilateral medial temporal lobectomy on a 27-year-old man with intractable epilepsy. The seizures improved. Henry Molaison was also left with profound anterograde amnesia — he could not form new long-term explicit memories. His IQ remained normal. His personality was preserved. He could carry on a conversation but would not remember it minutes later.

Brenda Milner and Suzanne Corkin studied H.M. for 55 years (Milner from 1955; Corkin from 1962). The findings reorganised the field.

What H.M. revealed

1. Memory is dissociable from intelligence and personality. A person with amnesia is not generally impaired.

2. The medial temporal lobes are necessary for new explicit memory. H.M.'s remaining short-term memory and long-term memories from before surgery were intact; only new long-term explicit memory formation was impaired.

3. Procedural memory is dissociable from declarative memory. H.M. could learn new motor skills (mirror tracing) without remembering having practised them. The dissociation revealed multiple memory systems.

Chapter VMultiple memory systems.

The contemporary taxonomy of long-term memory:

Declarative (explicit) memory — consciously accessible:

  Episodic: specific events. "Where I was when I heard about the news on Tuesday."

  Semantic: general world knowledge. "Paris is the capital of France."

Non-declarative (implicit) memory — not consciously accessible but expressed in performance:

  Procedural: motor and cognitive skills. Riding a bicycle.

  Priming: prior exposure facilitates later processing of related stimuli.

  Classical conditioning: learned associations.

  Habituation and sensitization: simple non-associative learning.

The case of K.C. (Kent Cochrane), studied by Tulving, demonstrated the dissociability of episodic and semantic memory: K.C. retained semantic knowledge after a motorcycle accident but lost the ability to recall any specific personal events from his life. Other patients show the reverse dissociation (loss of semantic memory with preserved episodic).

Chapter VIEncoding depth.

Fergus Craik and Robert Lockhart (1972) argued the multi-store model had the wrong primary variable. What predicts long-term retention is not how long material spent in short-term memory but how deeply it was processed.

Shallow processing: focus on perceptual features (Is the word in capital letters?). Phonemic processing: focus on sound (Does the word rhyme with cat?). Semantic processing: focus on meaning (Is this word a type of animal? Could it apply to me?).

The classic Craik & Tulving (1975) finding: words processed semantically are remembered far better than words processed phonemically, which are remembered better than words processed perceptually — even when processing time is held constant.

The self-reference effect

Even deeper than semantic: information processed in relation to oneself ("Does this adjective describe me?") shows the strongest retention of any encoding strategy. Rogers, Kuiper, & Kirker (1977). The effect is a useful study tip: ask how new material relates to your own experience.

Chapter VIIContext matters.

The encoding specificity principle: what is retrieved depends on what was encoded, including the context. The most famous demonstration is Godden & Baddeley's 1975 scuba-divers study. Divers learned word lists either underwater or on a beach; later they were tested either underwater or on a beach. Recall was best when the encoding and retrieval contexts matched. Words learned underwater were recalled better underwater.

State-dependent learning

The same effect for internal states. Material learned while moderately intoxicated is better recalled while moderately intoxicated than while sober (the opposite is also true; the literature is robust). Mood-congruent recall: depressed individuals recall more depressing memories than positive ones; the bias compounds the disorder.

Practical use

Studying in conditions similar to those of the eventual test improves performance. The original experiments are decades-old; the principle has been replicated repeatedly.

Chapter VIIIMemory is reconstructive.

Frederic Bartlett's 1932 work demonstrated that memory is not a high-fidelity recording. Participants asked to retell a Native American folk tale (The War of the Ghosts) systematically distorted the story across successive retellings — adding rationalising detail, removing supernatural elements, smoothing the unfamiliar to fit Western narrative conventions. Bartlett called the underlying mental templates schemas.

The reconstructive view has been extensively confirmed. Memory uses general schemas (what restaurants are like, how arguments unfold, how meetings are structured) to fill in details that were never specifically encoded. This is highly efficient and often correct, but it produces predictable errors.

The misinformation effect

Elizabeth Loftus's (1974) classic demonstration: showing participants a video of a car accident, then asking different questions about it. Asked "How fast were the cars going when they smashed into each other?" they reported higher speeds than asked the same question with hit; a week later they reported seeing broken glass that wasn't in the video. Memory had been edited by the questioning.

Chapter IXLoftus.

Elizabeth Loftus is the central figure in the contemporary scientific understanding of false memory. Her programme of research demonstrated that memories can be:

1. Distorted by post-event information (the misinformation effect).

2. Implanted wholesale through suggestion ("Lost in the Mall," 1995). Subsequent variations have implanted memories of meeting Bugs Bunny at Disneyland (impossible — Bugs is Warner Bros.), of being attacked by an animal, of nearly drowning.

3. Held with high confidence even when objectively false. Confidence is not a reliable marker of accuracy.

Implications

The legal-system implication is enormous. Eyewitness testimony — long considered the most reliable form of evidence — is in fact one of the least reliable. The Innocence Project's exoneration database (DNA evidence overturning convictions) shows that 70% of overturned cases involved mistaken eyewitness identification.

Loftus's role in the 1990s "memory wars" — particularly her testimony against the recovered-memory movement around alleged childhood sexual abuse — was scientifically vindicated and personally costly.

Chapter XThe seven sins.

Daniel Schacter's taxonomy of memory's predictable failures. Schacter's argument is that the failures are not bugs but features — side effects of an otherwise efficient adaptive system.

1. Transience. Memory fades over time. (Ebbinghaus's curve.)

2. Absent-mindedness. Failures of attention. The keys are forgotten because attention was elsewhere when they were put down.

3. Blocking. Tip-of-the-tongue states; temporary inability to retrieve a known fact.

4. Misattribution. Recalling a fact correctly but misattributing its source. (Source-monitoring failures.)

5. Suggestibility. Vulnerability to misleading post-event information. (Loftus.)

6. Bias. Current beliefs and feelings reshape memory of past events. People remember opinions they held in the past as more similar to their current ones than they actually were.

7. Persistence. Unwanted memories that won't go away. The traumatic flashback; the embarrassing intrusive memory.

Chapter XIRetrieval practice.

The most important applied finding in memory research. Henry Roediger and Jeffrey Karpicke's 2006 paper compared three study conditions for college students reading prose passages:

(a) Read the passage four times.

(b) Read the passage three times, then write down everything they could recall.

(c) Read the passage once, then write down everything they could recall, three times.

On a test five minutes later, condition (a) performed best. The students preferred condition (a). On a test one week later, condition (c) performed best by a substantial margin. The students who had spent the least time studying and the most time testing themselves remembered the most.

The effect is robust, large, and counterintuitive. Active retrieval — even from a source you have not yet "fully learned" — strengthens memory more than additional encoding. The implication for studying: stop re-reading; start testing yourself.

Why students don't do this

Re-reading feels productive (the material gets easier). Self-testing feels unproductive (it surfaces what you don't know). The metacognitive judgment is wrong; the actual learning gain is in the discomfort.

Chapter XIISpacing.

The spacing effect: distributed practice produces more durable learning than massed practice. The effect is one of the most robust and replicated findings in memory research; it has been shown across age groups, content domains, and time-scales from minutes to years.

Optimal spacing

Cepeda et al. (2008) — the meta-analysis. The optimal gap between practice sessions depends on the desired retention interval. As a rough rule: the optimal gap is about 10–20% of the retention interval. To remember something for a year, study it again about every 30–60 days.

Implementation: spaced-repetition systems

Software like Anki and SuperMemo automate the scheduling. The user creates flashcards; the algorithm schedules each card for review at increasing intervals based on past success. Successful retrievals push the next review further out; failures bring it closer. The algorithms (SM-2 in Anki, more sophisticated FSRS variants now) approximate optimal spacing for each individual card.

Medical students, language learners, and Wikipedia editors have made Anki near-standard. The empirical case for spaced retrieval as a study tool is essentially settled.

Chapter XIIIInterleaved practice.

When practising multiple types of problem (e.g., volume calculations for cylinders, cones, spheres), interleaved practice — mixing the types within each session — produces better retention than blocked practice — doing all the cylinders, then all the cones.

Rohrer & Taylor (2007); Kornell & Bjork (2008) provide the empirical foundation. The effect transfers across many domains: math, motor skills, art-style identification.

Why it works: interleaving forces the learner to retrieve and select the appropriate procedure for each problem, which improves discrimination between procedures. Blocked practice lets the learner apply the same procedure repeatedly, which feels efficient but produces shallower learning.

The metacognitive trap

Like the testing effect, interleaving feels worse during practice (fewer correct answers, more confusion) and produces better outcomes. Students reliably prefer blocked practice and judge themselves as having learned more from it. They are wrong, by the test that matters.

Chapter XIVSleep consolidates memory.

Memory consolidation — the process by which new memories are stabilised and integrated with existing knowledge — depends substantially on sleep. The mechanism is now relatively well-understood at the level of neurophysiology.

During slow-wave sleep, hippocampal neuron firing patterns from waking hours are replayed at faster speed, in a coordinated dialogue with the neocortex. Bruce McNaughton, Matt Wilson, and others demonstrated this in rats in the 1990s; the technique has been extended to humans.

Behaviourally: Robert Stickgold and Matthew Walker's programmes have repeatedly shown that sleep after learning improves retention more than equivalent waking periods. A nap can produce the consolidation benefit. Sleep deprivation impairs both the encoding of new memories the next day and the consolidation of memories formed before the sleep loss.

Targeted memory reactivation

An emerging technique: presenting a sound or smell during learning, then re-presenting it during slow-wave sleep, selectively strengthens the associated memories. The effect is small but real and has been replicated across multiple labs.

Chapter XVReconsolidation.

The traditional view: a memory, once consolidated, is stable. Karim Nader and colleagues' 2000 paper in Nature overturned this. They showed that retrieving a fear memory in rats made it temporarily susceptible to disruption — protein-synthesis inhibitors administered immediately after retrieval prevented the memory from being re-stored.

Each retrieval of a memory, in the contemporary view, opens a window during which the memory can be modified, weakened, or strengthened. The process is called reconsolidation. It explains how memory can be reconstructive (each retrieval slightly changes the memory) and provides a possible therapeutic target.

Therapeutic implications

If retrieval makes a fear memory temporarily labile, then administering a treatment during the reconsolidation window (within roughly 6 hours) could weaken the memory. Two clinical applications have been investigated: propranolol (beta-blocker) administered after PTSD memory retrieval; extinction learning during the reconsolidation window. Results are mixed; the basic phenomenon is robust, the clinical translation has been harder than initially hoped.

Chapter XVIThe repressed-memory controversy.

In the late 1980s and early 1990s, a number of adults, mostly women in therapy, came to believe they had been sexually abused as children — memories that had supposedly been repressed for decades and "recovered" through hypnosis, guided imagery, and suggestive questioning. Many entered lawsuits and criminal accusations against family members.

Experimental memory research — Loftus's work, Ofshe and Watters's Making Monsters (1994), Pendergrast's Victims of Memory — established that the therapeutic techniques used could induce false memories indistinguishable from real ones. The American Psychological Association's working group (1996) acknowledged that recovered memories could not be reliably distinguished from confabulation without external corroboration.

The episode left lasting damage to families and clinical practice. The current consensus: real childhood trauma is rarely fully forgotten and later "recovered" — the clinical pattern is usually one of partial memory of well-remembered abuse. The strong-form repressed-memory model is not supported.

Trauma can certainly affect memory in many ways (van der Kolk's work, the Clinical deck). But the specific 1990s recovered-memory framework, as practised, has not held up.

Chapter XVIIMemory techniques.

Methods that exploit the brain's stronger memory for spatial and visual material:

Method of Loci (Memory Palace). Imagine a familiar location (your childhood home, a route through your city). Mentally place the items you want to remember at specific locations. To recall, mentally walk the route. Used by Roman orators to memorise speeches; used by every modern memory champion.

Major System / Phonetic Number System. Convert digits into consonants by a fixed mapping; add vowels to make memorable words. Used to memorise long numbers (digits of pi, phone numbers).

Peg System. Pre-memorise a list of pegs (one-bun, two-shoe, ...) then attach new items to the pegs through vivid imagery.

Chunking. Group disconnected items into meaningful units. A phone number is easier as 555-867-5309 than as 5558675309.

Joshua Foer

Joshua Foer's Moonwalking with Einstein (2011) is the readable trade book on memory-championship technique. Foer trained for one year and won the 2006 USA Memory Championship. The techniques work; they require sustained practice; they do not generalise to other types of memory.

Chapter XVIIILearning styles do not work.

The widespread educational belief that students have characteristic "learning styles" (visual, auditory, kinesthetic, etc.) and that teaching matched to those styles improves learning is not supported by evidence.

Harold Pashler, Mark McDaniel, Doug Rohrer, and Robert Bjork's 2008 review in Psychological Science in the Public Interest examined the empirical literature and concluded: there is no rigorous evidence that the learning-styles approach improves learning outcomes; the studies that purport to show it use weak designs; and the educational resources spent on learning-styles-based curriculum redesign are likely wasted.

What does work: matching teaching to content rather than to learner. Visual material is best taught visually; verbal material is best taught verbally. The optimal modality is content-driven, not student-driven. Most learners benefit from multimodal presentation, not single-modality matched to a self-reported preference.

The myth persists. About 90% of teachers across multiple international surveys still report believing in learning styles. The gap between research and practice in this domain is one of the largest in education.

Chapter XIXWhat changes, what doesn't.

Age-related memory change is selective rather than uniform. Episodic memory (specific events) shows the steepest decline starting in the 60s; working memory and processing speed decline gradually from the 30s onward. Source memory (where you learned something) declines faster than item memory (the fact itself).

What stays robust: semantic memory, vocabulary, procedural memory, autobiographical memories from early adulthood (the "reminiscence bump" — disproportionate vivid memory for events from age 10–30, robust across cultures).

Mild cognitive impairment vs dementia

About 15–20% of adults over 65 meet criteria for Mild Cognitive Impairment (MCI) — measurable cognitive decline that does not yet significantly impair daily functioning. About 10% of MCI cases progress to dementia per year, though many remain stable or revert.

Alzheimer's disease — the most common dementia — typically presents first with progressive episodic-memory loss, eventually progressing to global cognitive decline. The 2023–24 disease-modifying drugs (lecanemab, donanemab) are the first treatments to slow progression, though modestly.