Associative Learning

Associative learning involves forming connections between events so that one predicts the other. Classical (Pavlovian) conditioning is a major form, in which a neutral stimulus becomes linked with a meaningful stimulus and begins to elicit a learned response.

Foundations of Associationism

Contiguity: Temporal and Spatial Closeness

Contiguity refers to how closely in time or space two events occur.
Earlier theories assumed that the closer the CS (conditioned stimulus) and US (unconditioned stimulus) occur together, the stronger the association formed.

Historical Roots

• Aristotle: Proposed associationism — learning occurs through contiguity, similarity, and contrast. Knowledge arises from experience (empiricism).
• Descartes: Introduced dualism and the idea of reflexes as automatic S → R responses.
• John Locke: Viewed the mind as a tabula rasa (blank slate). Complex ideas arise from combinations of simple sensory experiences.

These ideas shaped behaviourist assumptions centuries later.

Classical Conditioning

Core Components of Classical Conditioning

Component	Description
US (Unconditioned Stimulus)	Naturally elicits a response.
UR (Unconditioned Response)	Automatic, innate reaction to the US.
CS (Conditioned Stimulus)	Initially neutral stimulus that gains meaning through association.
CR (Conditioned Response)	Learned response to the CS.

Second-Order Conditioning

A neutral stimulus becomes a CS by pairing with an already established CS (not directly with a US). Demonstrates layered learning and abstraction.

---

“Wrinkles” in Classical Conditioning

Classical conditioning is not purely mechanical. Several phenomena show limits to simple S–R pairing.

Overshadowing

When two cues predict a US simultaneously, the more salient cue acquires stronger associative strength, reducing learning about the weaker cue.

Blocking

If a CS already predicts the US, adding a second CS does not lead to learning about it. This occurs because the US is not surprising, so no new associative strength is needed.

Latent Inhibition

Prior non-reinforced exposure to a neutral stimulus slows later conditioning. The organism has learned the stimulus predicts “nothing,” reducing attention to it.

Conditioned Inhibition

A stimulus becomes a signal for the absence of the US. The CR elicited is inhibitory rather than excitatory.

Contiguity vs Contingency

Why Contiguity Alone Is Not Enough

Although CS and US must occur close together, contiguity is not sufficient for learning.

Contingency: The Predictive Relationship

Contingency refers to the degree to which one event predicts another:

• If the CS reliably predicts the US → learning occurs.
• If the US occurs with or without the CS → little or no learning.

Expectancy and Prediction Error

Conditioning involves forming expectations about the US based on the CS.

Role of Surprise

Learning occurs when there is a prediction error — a mismatch between what is expected and what actually occurs.

• US > expected → excitatory learning increases.
• US = expected → no learning.
• US < expected → inhibitory learning.

Determinants of Contingency

• Reliability: How often US follows the CS.
• Uniqueness: How often US occurs without the CS (background probability).

Quantifying contingency

Mathematical Condition for Learning

Learning occurs when:

p(US | CS) > p(US | no CS)

Three major contingency types:

Contingency Type	Meaning	Effect
Positive	CS predicts presence of US	Excitatory conditioning
Zero	CS does not change probability of US	No conditioning
Negative	CS predicts absence of US	Inhibitory conditioning

Rescorla (1968): The contingency experiment

Experiment Design

• Rats trained to lever-press for food (baseline behaviour).
• Tone (CS) paired with shock (US).
• All groups had equal CS–US contiguity, but differed in:
  • Probability of shock during the tone.
  • Probability of shock without the tone.

Findings

• Strongest conditioning occurred when shocks only occurred during the tone.
• Weaker or no learning occurred when shocks happened unpredictably during no-tone intervals.

Conclusion

Contiguity alone is insufficient — predictability drives conditioning.

The Rescorla–Wagner model

Core Assumptions

Learning is driven by prediction error — the discrepancy between expected and actual outcomes.

Formula

ΔV = α (λ − V)
Where:

• ΔV = change in associative strength
• α = salience/learning rate of CS
• λ = maximum associative value of the US
• V = current expectation (total associative strength)

Learning ends when λ = V (no prediction error).

Six Rules of the Rescorla-Wagner Model

1. Stronger-than-expected US → excitatory learning.
2. Weaker-than-expected US → inhibitory learning.
3. Expected US = actual US → no learning.
4. Learning magnitude depends on prediction error.
5. More salient CSs learn faster (higher α).
6. When multiple CSs occur together, their combined strength determines expectation.

Model Applications

Acquisition

Rapid early learning, slow later learning as expectations approach asymptote.

Blocking

If CS1 fully predicts the US, adding CS2 results in no learning — US is not surprising.

Extinction

CS without US → expected US > actual → negative prediction error → decrease in V.

Conditioned Inhibition

CS predicts reduced US → negative associative strength.

Overexpectation

Two strong CSs paired together lead to a predicted US greater than the actual → both lose associative strength.

Alternative models of conditioning

Theory	Core Idea	Mechanism	Key Findings / Explanations
Attentional Theories	Learning depends on attention to the CS, not just US surprise.	Attention (α) is dynamic and changes trial-by-trial depending on cue informativeness.	– Mackintosh (1975): Attention increases to the best predictors. – Pearce & Hall (1980): Attention increases when the US is unpredictable. – Dynamic α: Unlike RW’s fixed α, attentional models update α continuously. – Explains Latent Inhibition: Pre-exposure lowers attention to CS → slower later learning. RW cannot explain this because it predicts no learning during pre-exposure.
Comparator Theories	Learning may occur without being expressed; performance depends on how well the CS predicts the US relative to context or other cues.	The animal compares the strength of CS–US association with context–US association to determine CR.	– Learning ≠ performance; associations may be stored even if not expressed. – Explains Blocking: CS2 may have learned, but CS1 outcompetes it. Extinguishing CS1 reveals CS2’s learning. – Role of Context: Context forms its own associations; CS elicits CR only if better predictor than context. – Emphasises comparison at test rather than trial-by-trial updates.

Common themes across models

• Learning depends on informativeness, not just pairing.
• Predictiveness is judged relative to competing cues and context.
• Expectation, attention, and comparison all contribute to conditioning.