Primary Readings
Gluck, Mercado & Myers Chapter 9; Ott & Nieder (2019) — dopamine and cognitive control. The Goldberg zebra finch dopamine study was lecture-emphasized and is likely exam-relevant.
Section 1
Course Arc: Where Module 11 Fits
Working memory (WM) is the "operating system" that all reinforcement learning algorithms run on. Without WM, the agent cannot hold the current goal in mind long enough to execute goal-directed behavior.
- PFC uses dopamine to gate what we hold in mind — the D1/D2 balance determines what enters WM, how stably it is maintained, and when it influences action.
- Disruption of PFC dopamine signaling causes collapse from goal-directed (model-based) to habitual (model-free) responding.
- Connect backward: actor-critic, basal ganglia as the actor, WM as the gating system that tells the actor which problem it is currently solving.
- Connect forward (Module 12): the same PFC gating logic regulates emotional expression via the amygdala.
The Unifying Metaphor (Memento)
Leonard Shelby cannot update his WM with new information — he is forever locked in a single frozen representation ("find John G."). This is an extreme caricature of D1 over-stabilization: the goal state is maintained so rigidly it cannot be updated by new evidence.
Section 2
Memory Systems — Behavioral Framework
| System | Duration | Capacity | Key Feature |
|---|---|---|---|
| Sensory (Iconic / Echoic) | <1 second | Very large — holds ALL visual items briefly | Sperling partial-report: attention selects from it; capacity far exceeds STM |
| Short-Term Memory (STM) | ~15–30 s without rehearsal | 7 ± 2 items (Miller) | Atkinson-Shiffrin: passive temporary storage; gateway to LTM |
| Working Memory (WM) | Variable — maintained actively | ~4 chunks (modern estimate) | Baddeley: active, goal-directed maintenance and manipulation under ongoing cognitive demands |
| Long-Term Memory (LTM) | Potentially permanent | Effectively unlimited | Declarative (hippocampus) + Non-declarative (BG, cerebellum) |
WM ≠ STM. STM is passive temporary storage. WM is active, goal-directed — it involves ongoing manipulation of information while simultaneously doing something else. The phonological loop rehearsing a phone number while you walk to write it down is WM, not STM.
Baddeley's Working Memory Components
| Component | Function |
|---|---|
| Phonological Loop | Verbal/auditory info; phonological store (~1-2 s) + subvocal articulatory rehearsal |
| Visuospatial Sketchpad | Spatial and visual information; mental rotation, spatial navigation |
| Central Executive | Attention-controlling component; coordinates subsystems, focuses attention, switches tasks |
| Episodic Buffer | Integrates information across subsystems and with LTM; binds multi-modal representations |
Sperling Partial-Report — Know This
Sperling flashed a 3×4 letter grid for 50 ms. Whole-report: subjects recalled only ~4 of 12. Partial-report (cue which row after offset): subjects could recall ANY cued row — proving that ~12 items were briefly available in iconic memory. Attention selects from a large sensory buffer; STM capacity is the bottleneck, not iconic memory.
Section 3
The Prefrontal Cortex
| Region | Primary Role | Key Evidence |
|---|---|---|
| DLPFC (Dorsolateral PFC) | WM maintenance (delay-period activity) + manipulation; top-down cognitive control | Goldman-Rakic monkey recordings; WCST failures after lesion → perseveration |
| VLPFC (Ventrolateral PFC) | Retrieval from LTM; response inhibition | Verbal WM tasks; go/no-go inhibition |
| mPFC (Medial PFC) | Emotional regulation; social cognition; self-referential processing | Default mode network; connects to amygdala |
Goldman-Rakic's Foundational Finding
In monkey recordings during delayed-response tasks, DLPFC neurons maintained elevated firing THROUGHOUT the delay period — even seconds after the cue was gone. This "delay-period activity" is the direct neural signature of WM: the circuit sustains a representation without external input, via recurrent excitation forming attractor states.
WCST (Wisconsin Card Sorting Task): Cards can be sorted by color, shape, or number. The rule changes periodically. DLPFC-damaged patients continue sorting by the old rule even after being told it changed — they perseverate. They cannot update WM and switch behavioral rules.
Section 4
Dopamine and the PFC: D1/D2 Balance
| Receptor | Primary Effect on WM | Mechanism | Clinical Extreme |
|---|---|---|---|
| D1 (Gs-coupled) | Stabilizes WM representations — deepens attractor basins; increases resistance to distractors | cAMP → PKA → closes HCN channels, potentiates NMDA → deeper attractor | Too much D1 → rigidity, locked representations (schizophrenia) |
| D2 (Gi-coupled) | Facilitates updating and flexibility — flattens attractor basins; allows new info to enter | Inhibits cAMP cascade → reduces attractor depth → easier to update | Too much D2 tone / too little D1 → distractibility (ADHD) |
Inverted-U Relationship
Both too little AND too much dopamine impair WM performance. Optimal dopamine (moderate level) → optimal WM. This inverted-U applies across PFC-dependent tasks: WM capacity, cognitive flexibility, distractor resistance. Moving off the peak in either direction impairs performance.
Earl Miller's DLPFC distractor experiments: While monkeys maintained a visual cue in WM through a delay period, irrelevant distractor stimuli were flashed. DLPFC neurons maintained the cue representation through the distractor; D1 stimulation increased this resistance. D1-mediated attractor deepening keeps WM contents stable against competitive inputs.
D1 Stabilization Pathway
DA release → D1 receptor → Gs protein → adenylyl cyclase ↑ → cAMP ↑ → PKA activation→ HCN channels close (↑ membrane resistance) + NMDA potentiation
→ Deeper attractor basin → WM representation MORE stable
Section 5
Stability-Flexibility Tradeoff
Three-Gate Model of Working Memory
INPUT GATE (D2-mediated) → [WM Maintenance — D1-mediated stability] → OUTPUT GATEWhat enters WM What stays in WM What influences action
| Gate | Dopamine Receptor | Function |
|---|---|---|
| Input gate | D2 — facilitates updating | Controls what NEW information enters WM; too little D2 activity → WM too sticky; too much → promiscuous updating |
| Maintenance | D1 — stabilizes | Keeps current WM contents active against interference; too little → distraction; too much → cannot update (perseveration) |
| Output gate | Both | Controls when WM contents influence downstream action selection; disruption here dissociates knowing-from-doing |
| Condition | D1/D2 Imbalance | Behavioral Consequence |
|---|---|---|
| Schizophrenia | Excess D1 (or hyperactive D1 circuits) → too much stability | Representations locked — cannot update WM with new disconfirming information → hallucinations and delusions persist |
| ADHD | Insufficient D1 tone → too much flexibility (or too little stability) | WM contents replaced too easily by distractors; difficulty maintaining goal across time |
Section 6
Neural Mechanisms: Persistent Activity and Attractors
Delay-period firing: PFC neurons maintain elevated firing rates throughout a delay between cue offset and required response — even in the absence of any external stimulus. This is the neural signature of active WM maintenance.
Attractor states: recurrent excitatory connections between PFC neurons create self-sustaining activity patterns. Once a representation is activated, the recurrent circuit keeps neurons firing each other — no external input required. These stable patterns are the neural implementation of WM.
| Parameter | High D1 (Deep Basin) | Low D1 / High D2 (Shallow Basin) |
|---|---|---|
| Stability | High — representation resists noise and distractors | Low — representation easily disrupted or replaced |
| Updatability | Low — hard to shift to a new representation | High — easy to update with new inputs |
| Clinical analog | Schizophrenic locked thoughts | ADHD distraction |
Why Attractor Models Explain WM Better Than Buffer Models
A simple buffer needs constant external refreshing. Attractor models generate persistent activity spontaneously through recurrent excitation — the network is self-sustaining. D1 makes the attractor basin deeper (representation harder to knock out); D2 flattens the basin (easier to update). Noise in the dopamine system can cause a representation to escape the attractor — this is how WM "forgets."
Section 7
The Goldberg Zebra Finch Studies (Lecture-Emphasized)
Exam-Relevant Lecture Content
The Goldberg finch dopamine study was emphasized in lecture. Know: (1) what the performance prediction error signal encodes, (2) what happens in the presence of a female, and (3) the analogy to actor-critic and WM gating.
Performance prediction error: dopaminergic neurons in the finch's basal ganglia fire when the bird's song output matches its internal template (song is not distorted → "I sang well") and pause when the output is distorted ("I sang poorly"). This encodes a self-generated performance quality signal — analogous to TD prediction error but applied to self-monitoring.
The female effect: when a female is present ("performance mode"), this internal dopamine signal disappears. The bird switches from internal learning feedback to external social feedback (the female's behavioral responses).
Two Modes — Context Gates the Learning Signal
REHEARSAL MODE: Internal error signal active → dopamine encodes self-assessed song quality → drives vocal learningPERFORMANCE MODE (female present): Internal error signal gated OFF → social feedback from female replaces it → no self-revision
Actor-Critic Analogy
The critic adjusts the actor's policy (vocal learning) during rehearsal. During performance, the critic is silent — the actor executes without self-correction. This is identical to the output gate logic: context gates which feedback signal operates. The WM gating framework predicts exactly when internal vs. external feedback should dominate.
Section 8
Clinical Perspectives
| Condition | Dopamine Mechanism | WM Effect | Treatment Logic |
|---|---|---|---|
| Schizophrenia | Excess dopamine in D1/subcortical circuits → PFC representations locked; subcortical D2 hyperactivity generates hallucinatory "noise" | Representations too stable — cannot update with new evidence; delusions and hallucinations persist | D2 antagonists (antipsychotics) reduce subcortical noise → allow PFC to update more normally |
| ADHD | Insufficient PFC dopamine → D1 hypo-activation → WM attractor basins too shallow | WM contents easily replaced by distractors; cannot sustain goal across time; impulsivity | Methylphenidate (Ritalin) boosts synaptic dopamine → more D1 activation → deeper attractor → WM stabilizes |
| Aging | Gradual loss of PFC dopamine with age (dopaminergic neurons decline) | Reduced WM capacity; more perseveration; harder to update rules | No validated pharmacological reversal; cognitive training; lifestyle factors |
Why D2 Antagonists Help Schizophrenia Despite a D1 Problem
The prevailing view: excess dopamine in subcortical D2 circuits generates the hallucinatory "noise" signal that floods PFC. Blocking D2 receptors reduces this noise → PFC dopamine levels can operate in a more normal range → WM representations can be updated. The PFC D1 instability is a separate route to the same phenomenology.
Section 9
Neurotransmitter Basics (Lecture-Only Section)
Neurotransmitter Release Sequence
Calcium-Mediated Vesicle Fusion
Action potential arrives at presynaptic terminal→ Membrane depolarization → voltage-gated Ca²⁺ channels OPEN
→ Ca²⁺ influx (high outside, low inside cytoplasm)
→ SNARE complex (synaptobrevin, syntaxin, SNAP-25) mediates vesicle docking and fusion
→ Exocytosis: transmitter released into synaptic cleft
→ Binds postsynaptic receptors → effect
| Receptor Type | Mechanism | Speed | Examples |
|---|---|---|---|
| Ionotropic | Ligand-gated ion channel — transmitter directly opens channel | Fast (milliseconds) | AMPA (Na⁺/K⁺), NMDA (Na⁺/Ca²⁺), GABA-A (Cl⁻) |
| Metabotropic (GPCR) | G-protein coupled → second messenger cascade → indirect channel modulation | Slow (seconds to minutes) | D1 (Gs → cAMP ↑), D2 (Gi → cAMP ↓), mGluR, GABA-B |
| Transmitter | Main Action | Key Receptor | Ion / Effect |
|---|---|---|---|
| GABA | Main inhibitory neurotransmitter | GABA-A (ionotropic) | Cl⁻ influx → hyperpolarization → reduced firing probability |
| Glutamate | Main excitatory neurotransmitter | AMPA + NMDA (ionotropic) | Na⁺ (AMPA) + Ca²⁺ (NMDA) influx → depolarization → LTP |
| Dopamine | Neuromodulator — slow, modulatory | D1, D2 (metabotropic GPCRs) | Via cAMP-PKA cascade → modulates channel states & synaptic strength |
D1 and D2 Are Metabotropic — Practical Consequence
Because D1/D2 signal via G-proteins and second messengers (not direct ion channel opening), their effects are slow (seconds to minutes) and modulatory — they change the excitability and synaptic strength of neurons over longer timescales. This makes dopamine a neuromodulator, not a classical fast transmitter. It tunes the gain of PFC circuits, not individual action potentials.
Section 10
Key Terms Checklist
- Working memory — active maintenance and manipulation; operating system for cognition
- Phonological loop — verbal/auditory WM subsystem; phonological store + subvocal rehearsal
- Visuospatial sketchpad — spatial/visual WM subsystem
- Central executive — attention-controlling WM component; coordinates subsystems
- Delay-period activity — persistent elevated PFC firing throughout delay; neural WM signature
- Attractor state — stable self-sustaining neural pattern via recurrent excitation
- DLPFC — WM maintenance + top-down control; Goldman-Rakic's key site
- VLPFC — retrieval and inhibition
- mPFC — emotional regulation; social cognition; connects to amygdala
- D1/D2 receptors — metabotropic GPCRs; D1 stabilizes, D2 updates
- Inverted-U — optimal WM at moderate dopamine; too little or too much impairs
- Stability-flexibility tradeoff — D1 maintains, D2 updates; core tension of WM
- Perseveration — failure to switch rules; DLPFC damage hallmark
- Wisconsin Card Sorting Task (WCST) — standard test of cognitive flexibility / rule switching
- Input/output gate — D2-mediated entry into WM; D1-mediated stable maintenance
- Performance prediction error — Goldberg finch: dopamine encodes song quality; silent during performance
- Schizophrenia — excess D1 stability → locked representations; treated with D2 antagonists
- ADHD — insufficient WM stability; treated with methylphenidate
- Ionotropic vs. metabotropic — fast/direct vs. slow/GPCR-mediated
- SNARE complex — synaptobrevin, syntaxin, SNAP-25; mediates vesicle fusion and exocytosis
- GABA — main inhibitory transmitter; Cl⁻ influx → hyperpolarization
- Glutamate — main excitatory transmitter; AMPA/NMDA → depolarization and LTP
- Goldman-Rakic — identified delay-period activity in DLPFC as neural signature of WM
- Ott & Nieder (2019) — review: D1/D2 balance, inverted-U, stability-flexibility, clinical implications
Key Terms — 25 Flashcards
Click any card to reveal its definition. Use the filters to focus on a category.
Behavioral
Working Memory
Click to reveal →
Behavioral
Active maintenance and manipulation of information; the "operating system" for cognition. Distinct from STM: WM is goal-directed and involves manipulation under ongoing cognitive demands, not merely passive storage.
Behavioral
Phonological Loop
Click to reveal →
Behavioral
WM subsystem for verbal/auditory information. Two parts: (1) phonological store holds sound-based info ~1-2 s; (2) articulatory rehearsal (subvocal repetition) refreshes the store. Explains why verbal rehearsal helps you remember a phone number.
Behavioral
Visuospatial Sketchpad
Click to reveal →
Behavioral
WM subsystem for spatial and visual information. Supports mental rotation, spatial navigation, and visual imagery. Operates in parallel with the phonological loop under central executive coordination.
Behavioral
Central Executive
Click to reveal →
Behavioral
Attention-controlling component of Baddeley's WM model. Coordinates the phonological loop, visuospatial sketchpad, and episodic buffer. Focuses and switches attention; analogous to DLPFC function in top-down control.
Behavioral
Perseveration
Click to reveal →
Behavioral
Failure to switch responses or rules after conditions change. Hallmark of DLPFC damage. Demonstrated on WCST: patients continue sorting by color after the rule changes to shape. Reflects inability to update WM and shift behavioral strategy.
Behavioral
Wisconsin Card Sorting Task (WCST)
Click to reveal →
Behavioral
Neuropsychological test of cognitive flexibility. Cards sorted by color, shape, or number; rule changes periodically. Frontal lobe / DLPFC damage → failure to shift sorting rules → perseveration. Standard clinical test of PFC-dependent rule-switching.
Neural
Delay-Period Activity
Click to reveal →
Neural
Persistent elevated firing of DLPFC neurons throughout a delay period after cue offset, even without external stimulus present. Goldman-Rakic's foundational finding in monkeys. The direct neural signature of active WM maintenance; sustained by recurrent excitation forming attractor states.
Neural
Attractor State
Click to reveal →
Neural
Stable neural pattern maintained by recurrent excitation among PFC neurons; the neural implementation of WM. D1 deepens attractor basins (harder to disrupt); D2 flattens them (easier to update). Noise can knock the representation out of the attractor — equivalent to "forgetting."
Neural
DLPFC
Click to reveal →
Neural
Dorsolateral prefrontal cortex. Site of WM maintenance and top-down cognitive control. Neurons show delay-period activity. Damage → perseveration on WCST. Goldman-Rakic's foundational recording site. High D1 receptor density — critical for stability-flexibility balance.
Neural
D1 Receptor
Click to reveal →
Neural
Metabotropic dopamine receptor (Gs-coupled). Stabilizes WM representations by deepening attractor basins via cAMP-PKA pathway: closes HCN channels, potentiates NMDA receptors. Increases resistance to distractors. Too much D1 → rigidity (schizophrenia); too little → distractibility (ADHD).
Neural
D2 Receptor
Click to reveal →
Neural
Metabotropic dopamine receptor (Gi-coupled). Facilitates WM updating and flexibility by flattening attractor basins (inhibits cAMP cascade). Controls the input gate — allows new information into WM. Too much D2 / too little D1 → ADHD-like distractibility; D2 antagonism → reduces subcortical noise in schizophrenia.
Neural
Input Gate
Click to reveal →
Neural
D2-mediated process that controls what information enters WM. When D2 activity facilitates updating, the input gate is open and new representations can displace old ones. When D2 is low (relative to D1), the gate is closed and existing WM contents are maintained against competing inputs.
Neural
Output Gate
Click to reveal →
Neural
Process that controls when WM contents influence behavior/action. Dissociates knowing from doing — a representation can be held in WM without necessarily driving current behavior. Disruption here can cause WM contents to inappropriately gate actions, or fail to do so when needed.
Neural
Performance Prediction Error
Click to reveal →
Neural
In Goldberg's zebra finch studies: dopamine signal encoding the match between song output and internal template. Fires when bird sings undistorted (good match), pauses when distorted (mismatch). Critically, this signal disappears in performance mode (female present) — replaced by external social feedback.
Neural
Ionotropic Receptor
Click to reveal →
Neural
Ligand-gated ion channel; direct, fast (milliseconds). Transmitter binds → channel opens → ions flow. Examples: AMPA (Na⁺/K⁺ → depolarization), NMDA (Na⁺/Ca²⁺ → depolarization + LTP), GABA-A (Cl⁻ → hyperpolarization).
Neural
Metabotropic Receptor
Click to reveal →
Neural
G-protein coupled receptor (GPCR); indirect, slow (seconds to minutes). Transmitter binds → G-protein activation → second messenger cascade (e.g., cAMP-PKA) → channel modulation or gene expression changes. Examples: D1, D2, mGluR, GABA-B. Dopamine is a neuromodulator — tunes gain, not individual action potentials.
Neural
SNARE Complex
Click to reveal →
Neural
Protein machinery mediating vesicle fusion and neurotransmitter exocytosis. Key proteins: synaptobrevin (vesicle), syntaxin (presynaptic membrane), SNAP-25 (presynaptic membrane). Ca²⁺ influx via voltage-gated channels triggers SNARE-mediated docking and membrane fusion → transmitter released into cleft.
Neural
GABA
Click to reveal →
Neural
Main inhibitory neurotransmitter. GABA-A is ionotropic: GABA binds → Cl⁻ channel opens → Cl⁻ influx → membrane hyperpolarization → reduced firing probability. GABA-B is metabotropic (slow, via K⁺ channels). Critical for inhibitory control and balancing excitation in cortical circuits.
Neural
Glutamate
Click to reveal →
Neural
Main excitatory neurotransmitter. AMPA receptors: Na⁺ influx → fast depolarization. NMDA receptors: Na⁺ + Ca²⁺ influx (requires depolarization to relieve Mg²⁺ block) → depolarization + LTP induction via Ca²⁺ signaling. Glutamate's recurrent excitation underlies attractor states in PFC.
Theory
Inverted-U Relationship
Click to reveal →
Theory
Relationship between dopamine level and WM/PFC performance. Too little dopamine → impaired WM; too much → also impaired; optimal moderate level → peak WM performance. The U-shape applies across WM capacity, cognitive flexibility, distractor resistance, and most PFC-dependent tasks.
Theory
Stability-Flexibility Tradeoff
Click to reveal →
Theory
Core tension in WM: maintaining current contents (D1-mediated stability) vs. updating them with new information (D2-mediated flexibility). The system must balance these opposing pressures. This is the central synthesis of Ott & Nieder (2019): opposite perturbations of the same balance produce ADHD vs. schizophrenia.
Theory
Goldman-Rakic
Click to reveal →
Theory
Neuroscientist who identified delay-period activity in DLPFC as the neural signature of WM through foundational monkey single-unit recording studies. Established that PFC neurons maintain representations through recurrent excitation without ongoing external input. Foundational for the attractor state model of WM.
Theory
Ott & Nieder (2019)
Click to reveal →
Theory
Review paper on dopamine and cognitive control in PFC. Central synthesis: D1/D2 balance regulates the stability-flexibility tradeoff that is fundamental to adaptive goal-directed behavior. Perturbations in either direction produce distinct clinical phenotypes: excess D1 stability → schizophrenia; insufficient D1 stability → ADHD.
Clinical
Schizophrenia
Click to reveal →
Clinical
Characterized by excess D1-mediated stability → PFC representations locked in, cannot be updated by disconfirming evidence → hallucinations and delusions persist. Also: subcortical D2 hyperactivity generates noise. Treatment: D2 antagonists reduce subcortical noise → allow PFC to update more normally.
Clinical
ADHD
Click to reveal →
Clinical
Characterized by insufficient WM stability (insufficient D1 tone → shallow attractor basins). WM contents easily replaced by distractors; difficulty maintaining goals across time; impulsivity. Treatment: methylphenidate (Ritalin) boosts synaptic dopamine → more D1 activation → deeper attractors → WM contents stabilize.
Practice Multiple Choice — 20 Questions
Click an option to check your answer. Explanations appear automatically.
30
Questions
0
Correct
0
Incorrect
—
Score
Big Picture Synthesis
How the week's concepts connect across levels of analysis and to the course arc.
The Unifying Principle
"The same dopamine signal that teaches the actor-critic system which actions to reinforce also maintains the working memory context that tells the system WHICH problem it is currently solving." D1/D2 balance is not just a detail of receptor pharmacology — it is the mechanism by which the brain decides whether to hold on to the current goal or update it.
Levels of Analysis
Behavioral
WM capacity: ~4 chunks
Cognitive flexibility vs. perseveration
WCST: rule-switching failures after DLPFC damage
Neural
DLPFC delay-period activity
D1/D2 balance; attractor states
Goldberg dopamine signal; two modes
Molecular
cAMP-PKA cascade (D1)
SNARE-mediated exocytosis
GABA/Glutamate; ionotropic vs. metabotropic
Clinical
Schizophrenia: too stable
ADHD: too flexible
Aging: gradual D1 loss
Likely Exam Themes
- D1/D2 balance and the stability-flexibility tradeoff — core Ott & Nieder synthesis
- Inverted-U: optimal WM at moderate dopamine
- Delay-period activity as neural signature of WM (Goldman-Rakic)
- WCST perseveration = DLPFC damage signature
- Goldberg finch: performance prediction error disappears with female present
- Methylphenidate mechanism: boosts DA → D1 → WM stability
- Antipsychotics: D2 antagonism reduces subcortical noise
- SNARE complex and Ca²⁺ mechanism of transmitter release
- Ionotropic vs. metabotropic speed difference
Cross-Course Connections
- WM = model-based substrate; BG = model-free (Weeks 8-9)
- D1 stability = maintaining current goal = model-based control
- D2 flexibility = goal updating = model-free transition trigger
- Dopamine prediction error (Week 8) same DA system that modulates WM
- Actor-critic: PFC = meta-controller; Goldberg critic silent during performance
- Module 10: WM is the gating system the basal ganglia actor operates within
- Module 12 preview: PFC gates emotional expression via amygdala — same D1/D2 logic