Learning & Memory

Learning & Memory

Part 1 of 4 — Classical Conditioning: Pavlovian Learning & Its Boundaries

The Core Vocabulary (get the labels mechanical)

Classical (Pavlovian, respondent) conditioning pairs a neutral stimulus with one that already triggers a reflexive response:

Labeling algorithm: find the stimulus that would work on day one with no training — that is the US, and its response is the UR. Whatever predicts the US and comes to evoke the response on its own is the CS. The CR and UR are usually similar, but the CR is often weaker.

Timing & Contingency

• Forward (delayed) conditioning — CS starts, US arrives while CS is still on: strongest learning
• Trace conditioning — CS ends, gap, then US: works, weaker
• Simultaneous and backward (US before CS): weak to negligible — the CS must predict the US
• Contingency beats contiguity (Rescorla): a CS paired with the US but no more informative than background context supports little conditioning. The animal learns predictive information, not mere co-occurrence.

The Lifecycle of a CR

• Acquisition: repeated CS–US pairings build the CR
• Extinction: CS repeatedly presented alone → CR fades. Extinction is new inhibitory learning, not erasure — the original CS–US association is suppressed, not deleted
• Spontaneous recovery: after a rest period, the extinguished CR partially returns, proving the original association survived
• Generalization: stimuli similar to the CS evoke the CR (Little Albert feared the rat, then rabbits and fur — Watson & Rayner's ethically infamous demonstration)
• Discrimination: with differential training (CS+ paired, CS− unpaired), responding narrows to the true predictor
• Higher-order conditioning: an established CS acts as a "US" to condition a second stimulus (light → bell → salivation); the second-order CR is weaker

Biological Boundaries (the MCAT's favorite exceptions)

• Taste aversion (Garcia effect): nausea conditions to a taste in one trial across a delay of hours — violating the usual need for repeated, close pairings — yet nausea will not readily condition to lights or sounds
• Preparedness: organisms are biologically primed to associate certain stimulus classes (taste→illness, in rats; color/visual cues→illness, in birds such as quail that pick food by sight) — learning is constrained by evolution, not infinitely flexible
• Clinically: chemotherapy patients develop anticipatory nausea to clinic smells and foods eaten before sessions — a direct Garcia-effect application

Worked Example — Labeling and Predicting in a Clinical Vignette

Vignette. A patient receiving chemotherapy eats a distinctive mint candy in the waiting room before each infusion. The drug reliably causes nausea an hour later. After four sessions, tasting the mint anywhere makes her queasy. Months after treatment ends, mints no longer bother her — until, a year later, one unexpected mint at a party produces a wave of mild nausea.

Step 1 — run the labeling algorithm. What causes nausea with no learning? The chemotherapy drug → US; drug-induced nausea → UR. The mint predicts the drug → CS; queasiness to mint alone → CR. The waiting room itself may become a weaker second CS (context conditioning).

Step 2 — explain the unusual efficiency. Conditioning succeeded in few trials across a one-hour delay, and attached to a taste rather than to the (equally present) waiting-room television. That is the Garcia pattern: prepared taste–illness association defeating the ordinary contiguity rules.

Step 3 — name the late events. Mints losing their effect after treatment = extinction (mint experienced repeatedly without the drug). The party episode = spontaneous recovery after a rest interval — evidence extinction suppressed, rather than deleted, the association.

Step 4 — anticipate the study question. A passage might ask how to prevent anticipatory nausea. The literature answer: give a novel, expendable "scapegoat" flavor before infusions so the aversion attaches to it instead of to the patient's regular diet — an intervention that only makes sense once you see conditioning as prediction-learning.

Classical Conditioning Foundations 🎯

Key Takeaways — Part 1

• Label mechanically: the stimulus that works untrained is the US; its reflex is the UR; the learned predictor is the CS; its acquired response is the CR
• Forward/delayed pairing conditions best; prediction (contingency) matters more than mere co-occurrence
• Extinction is new inhibitory learning — spontaneous recovery after rest proves the original association survives
• Generalization spreads the CR to similar stimuli; discrimination training (CS+ vs CS-) narrows it; higher-order conditioning chains a second, weaker CS
• Garcia taste aversion: one trial, hours-long delay, taste-specific — biological preparedness constrains what can be learned, and it explains anticipatory nausea in chemotherapy

Learning & Memory

Part 2 of 4 — Operant Conditioning: Consequences, Schedules & Learning Without Reinforcement

Classical vs. Operant — the First Fork in Every Item

• Classical: behavior is elicited (reflexive, involuntary); learning links two stimuli that occur before the response
• Operant (Thorndike's law of effect; Skinner): behavior is emitted (voluntary); learning links the behavior to its consequence

The Consequence 2×2 (define by effect, not by feel)

• "Positive/negative" mean add/remove, never good/bad; "reinforcement/punishment" are defined by whether the behavior actually increases or decreases
• Escape learning terminates an ongoing aversive stimulus; avoidance learning prevents it entirely (a warning cue signals the response window) — both are negative reinforcement
• Primary reinforcers satisfy biological needs (food, warmth); secondary (conditioned) reinforcers gain value by association (money, tokens, grades); token economies apply this clinically

Shaping and Extinction

• Shaping: reinforce successive approximations of a target behavior — how complex behaviors that would never occur spontaneously get built
• Operant extinction: consequence stops → behavior fades, often after a brief extinction burst (the vending machine gets hit harder before being abandoned)

Schedules of Reinforcement (know the response signatures)

• Continuous reinforcement → fastest acquisition; partial (intermittent) reinforcement → slowest extinction (the partial reinforcement extinction effect)

Learning Without Obvious Reinforcement

• Latent learning (Tolman): unrewarded rats still built cognitive maps of a maze — performance jumped as soon as reward appeared, proving learning had occurred silently. Learning ≠ performance.
• Observational learning (Bandura's Bobo doll): watching a model — attention → retention → reproduction → motivation; vicarious reinforcement/punishment modulates imitation; mirror neurons proposed as a substrate
• Instinctive drift (the Brelands): trained behavior degrades toward species-typical fixed patterns (raccoons "washing" tokens instead of depositing them) — biology constrains operant learning just as preparedness constrains classical

Worked Example — Reading an Operant Passage's Graph and Design

Passage. Researchers train rats to lever-press. Group A is reinforced for every press; Group B on a variable-ratio 10 schedule; Group C receives food at unpredictable times regardless of pressing. After training, all food delivery stops. A cumulative-response graph shows Group A responding fastest during early training but quitting quickly in extinction; Group B pressing at the highest sustained rate and persisting longest after food stops; Group C pressing at chance levels throughout.

Step 1 — map groups to concepts. A = continuous reinforcement (fast acquisition, fast extinction). B = VR (highest steady rate; the partial reinforcement extinction effect explains its persistence — the animal cannot easily detect that the contingency ended). C = noncontingent delivery: food is not a consequence of pressing, so no operant learning of pressing is expected (C is the control demonstrating that contingency, not mere food, drives the behavior).

Step 2 — anticipate the "superstition" distractor. Noncontingent reinforcement can produce idiosyncratic repeated behaviors (accidentally reinforced actions), but it cannot sustain the specific trained response at high rates — do not confuse the control group with a reinforcement schedule.

Step 3 — the transfer item. "A casino's slot machine and a factory's per-piece pay differ how?" Slot machine = VR (count unpredictable → steady rapid play, extinction-resistant); piecework = FR (predictable count → high rate with post-reinforcement pauses). Both are ratio schedules — the difference is variability, and variability is what buys persistence.

Step 4 — spot the classical/operant boundary. If the passage adds that rats began salivating when the food hopper clicked, that response is classical (click = CS predicting food), running in parallel with the operant lever-pressing. Real preparations almost always contain both — assign each response to its own paradigm.

Operant Conditioning & Schedules 🎯

Key Takeaways — Part 2

• First fork: elicited response linking stimuli = classical; emitted behavior shaped by consequences = operant
• The 2x2 is definitional: positive/negative = add/remove; reinforcement/punishment = behavior up/down; escape and avoidance are both negative reinforcement
• Shaping builds novel behavior via successive approximations; extinction often begins with a burst
• Schedule signatures: VR = highest rate, most extinction-resistant; FI = scallop; VI = slow and steady; continuous = fast acquisition but fast extinction (partial reinforcement extinction effect)
• Learning without direct reinforcement is real: latent learning (cognitive maps; learning vs performance), observational learning (Bandura; vicarious consequences), and instinctive drift shows biology constrains it all

Learning & Memory

Part 3 of 4 — Memory: Encoding, Storage & Retrieval

The Multi-Store Architecture (Atkinson–Shiffrin, upgraded)

Working memory (Baddeley) replaces the passive "short-term store": a central executive allocates attention across the phonological loop (inner speech), the visuospatial sketchpad, and the episodic buffer. Prediction it earns: two tasks using different subsystems (verbal + spatial) interfere far less than two tasks sharing one.

Encoding — Getting Information In

• Depth of processing (Craik & Lockhart): semantic encoding (meaning) beats acoustic (sound), which beats visual (appearance) — judged by later recall, even when study time is equal
• Self-reference effect: relating material to yourself deepens encoding further
• Elaborative rehearsal (linking to existing knowledge) builds durable memory; maintenance rehearsal (rote repetition) mostly refreshes short-term memory
• Spacing effect: distributed practice beats massed practice (cramming); testing effect: retrieval practice beats rereading
• Mnemonics (method of loci, pegwords, acronyms) work by adding organization and imagery — i.e., forced elaboration
• Dual coding: items encoded both verbally and visually get two retrieval routes

Long-Term Memory Systems (the dissociation map)

• Explicit (declarative) — conscious recollection; hippocampus-dependent
  • Episodic: events with time-and-place tags (your last birthday)
  • Semantic: facts without source tags (Paris is in France)
• Implicit (nondeclarative) — expressed through performance, not recollection
  • Procedural skills (typing, mirror tracing): basal ganglia, cerebellum
  • Priming: prior exposure speeds processing without awareness
  • Conditioned associations (Parts 1-2) are also implicit
• Patient H.M. (bilateral medial temporal/hippocampal resection): dense anterograde amnesia for new explicit memories, yet improved daily at mirror tracing while denying ever having done it — the dissociation cornerstone. The hippocampus consolidates new explicit memories into cortex; long-term potentiation (LTP) — persistent strengthening of synapses after high-frequency stimulation — is the leading cellular mechanism.

Retrieval — Getting Information Out

• Recall (generate from scratch) is harder than recognition (pick from options); relearning savings is the most sensitive index that a trace remains
• Retrieval cues: context-dependent memory (physical setting match — divers recall word lists best in the environment where they learned them), state-dependent memory (internal physiological state match), and mood-congruent retrieval (current mood favors matching memories)
• Serial position effect: primacy (early items rehearsed into LTM) + recency (last items still in working memory). A filled delay before recall abolishes recency but spares primacy — the classic evidence for two stores.

Worked Example — A Levels-of-Processing Experiment, Fully Read

Passage. Participants see 60 words. For 20, they judge whether the word is in capital letters (visual); for 20, whether it rhymes with a target (acoustic); for 20, whether it fits a sentence (semantic). Judgment time is equated. A surprise recognition test follows. Results: semantic 78% correct, acoustic 55%, visual 33%. A second experiment repeats the design with a filled 10-minute delay before an oral free-recall test and finds recall of the final list positions drops to the level of middle positions, while early positions are unaffected.

Step 1 — identify the design's point. Study time is controlled, so differences cannot reflect exposure duration — only the DEPTH of processing each orienting task forces. The semantic > acoustic > visual gradient is the levels-of-processing signature; note participants never intended to memorize (incidental encoding), which strengthens the claim that depth, not intention, drives encoding quality.

Step 2 — read the serial-position manipulation. The filled delay eliminates the recency advantage (working memory has been flushed by the distractor task) but leaves primacy intact (early words got extra rehearsal into long-term memory). One manipulation moving one component while sparing the other is a dissociation — the evidence pattern the MCAT most often asks you to interpret.

Step 3 — pre-empt the trap answers. "Semantic words were seen longer" — excluded by design. "The delay caused general forgetting" — no, it selectively removed recency. "Recognition and recall are interchangeable" — they are not; recognition provides the item as its own cue, which is why experiment 2 needed free recall to expose the serial-position curve.

Step 4 — the application item. A student who rereads notes (maintenance rehearsal, shallow) versus one who writes practice questions linking concepts to cases (semantic elaboration + testing effect): the passage's data predict the second student's advantage, and naming BOTH mechanisms is what separates the best answer from the near-miss.

Encoding, Storage & Retrieval 🎯

Key Takeaways — Part 3

• Stores: sensory (iconic under 1 s, echoic a few seconds) → working memory (about 7 items, extended by chunking; loop + sketchpad + episodic buffer under a central executive) → LTM
• Encoding gradient: semantic > acoustic > visual; elaborative beats maintenance rehearsal; spacing and testing effects are the two highest-yield study principles
• LTM map: explicit (episodic + semantic, hippocampus-consolidated, LTP as mechanism) vs implicit (procedural, priming, conditioning) — H.M. is the dissociation proof
• Retrieval: recall < recognition < relearning in sensitivity; context-, state-, and mood-matching all serve as cues
• Serial position: primacy = LTM, recency = working memory; a filled delay kills recency only — know that dissociation cold

Learning & Memory — Deep Dive

Part 4 of 4 — MCAT Integration: Forgetting, Memory Errors & Reading Learning-Memory Studies

Why We Forget — Locate the Failure Stage

The cueing test is the great divider: if recognition or cued recall recovers the item, the memory was stored and the failure was retrieval. MCAT passages engineer exactly this contrast.

Interference — Forgetting Caused by Other Learning

• Proactive interference: OLD learning disrupts NEW (last year's locker code intrudes on this year's)
• Retroactive interference: NEW learning disrupts OLD (this year's code erases access to last year's)
• Direction mnemonic: the prefix names which memory does the damage relative to the target — proactive = prior material attacks forward; retroactive = recent material attacks backward
• Sleep after learning reduces retroactive interference and supports consolidation

Memory Is Reconstructive (the error catalog)

• Misinformation effect (Loftus): post-event information rewrites reports — "smashed" vs. "hit" changes speed estimates and later false memory of broken glass
• Source monitoring errors: content remembered, origin misattributed (was it seen, imagined, or told?)

• Cryptomnesia = unconscious plagiarism: a source-monitoring error in which a remembered idea is experienced as one's own new idea

• Flashbulb memories: vivid and confidently held memories of emotional events — confidence stays high while accuracy decays like ordinary memory (the confidence-accuracy dissociation)
• False/recovered memory debate: suggestive techniques can implant detailed, sincerely believed events — central to eyewitness-testimony reform
• Clinical anchors: Alzheimer's disease (early hippocampal/acetylcholine involvement → anterograde-first explicit loss); Korsakoff syndrome (thiamine deficiency, often with confabulation — fluent, unintentional fabrication); retrograde amnesia (loss of old memories, typically temporally graded) vs. anterograde (cannot form new ones)
• Normal aging: recall declines more than recognition; procedural and semantic knowledge remain robust

The Paradigm-Sorting Flowchart (for any learning passage)

1. Is the response reflexive and triggered by a predictive stimulus? → classical conditioning (label US/UR/CS/CR)
2. Is a voluntary behavior changing with its consequences? → operant (identify add/remove × increase/decrease, then the schedule)
3. Did behavior change after mere observation of a model? → social/observational learning
4. Is performance improving or failing on a memory test? → name the stage (encoding/storage/retrieval), the store, and the interference direction Most passages mix two of these; assign each dependent measure separately.

Worked Example — An Eyewitness-Memory Passage, Fully Read

Passage. Participants watch a staged theft video. Group 1 then reads an inaccurate summary mentioning a "toolbox" (no toolbox appeared); Group 2 reads an accurate summary; Group 3 reads nothing. A week later, all take (a) a free-recall test and (b) a recognition test including the toolbox. Group 1 "recognizes" the toolbox 44% of the time versus 9% and 8% in Groups 2 and 3, and reports high confidence in those recognitions. A follow-up finds the effect shrinks when participants are warned about the summary's unreliability BEFORE reading it, but not when warned after a week.

Step 1 — name the effect and the control logic. Elevated false recognition only in the misled group is the misinformation effect. Group 2 controls for reading any summary; Group 3 for mere delay — so the false memory traces specifically to the misleading content.

Step 2 — interpret the confidence data. High confidence accompanying false recognitions reproduces the confidence-accuracy dissociation: jurors weight confidence heavily, but the data show it tracks the FEELING of memory, not its source. Expect an answer choice equating confidence with accuracy — it is always wrong.

Step 3 — use the warning asymmetry. A pre-exposure warning lets participants tag the summary as a separate, suspect source at ENCODING; a warning after a week fails because the event and post-event sources have already blended — a source-monitoring account. This asymmetry is the passage's strongest clue that the mechanism is misattribution, not simple overwriting or social compliance.

Step 4 — the application item. "Which police practice does this research most directly support?" Blind lineup administration and neutral, non-leading question wording — interventions that keep post-event information from entering the witness's memory in the first place, since the follow-up shows late corrections barely work.

Forgetting, Memory Errors & Paradigm Sorting 🎯

Key Takeaways — Part 4

• Diagnose forgetting by stage: cues/recognition restore it = retrieval failure; nothing restores it = encoding or storage; Ebbinghaus curve is steep early
• Interference directions: proactive = old attacks new; retroactive = new attacks old — identify which memory is the TARGET first
• Memory is reconstructive: misinformation effect, source-monitoring errors, and confident-but-drifting flashbulb memories; confidence never certifies accuracy
• Clinical anchors: Alzheimer's = anterograde-first explicit loss (hippocampus, acetylcholine); Korsakoff = thiamine deficiency with confabulation; aging spares recognition more than recall
• Passage flowchart: elicited-by-predictor = classical; consequence-driven = operant; model-driven = observational; test-performance = name stage, store, and interference direction

• Low-yield extras: cryptomnesia (unconscious plagiarism) is a source-monitoring error in which a remembered idea feels new and self-generated