Learning I: The Fundamental Theories of Thorndike, Guthrie and Hull, and Classical Conditioning — Procedure, Phenomena and Related Issues
2. Guthrie: contiguity and one-trial learning
Edwin R. Guthrie reduced learning to one principle: "A combination of stimuli which has accompanied a movement will on its recurrence tend to be followed by that movement." Contiguity alone is sufficient — no reward is needed — and an association reaches full strength on a single pairing (one-trial learning). What is learned is movements, not acts; skills improve with practice only because an act consists of many movements, each needing its own association in many stimulus situations. By the recency principle, whatever was done last in a situation is what will be done next time.
Reward works, Guthrie argued, only because it changes the stimulus situation and so protects the last response from being unlearned. Guthrie and Horton (1946) photographed cats escaping from a puzzle box and found highly stereotyped escape movements repeated trial after trial. Forgetting is caused by new learning (interference), not by decay. Guthrie described three ways to break a habit: the threshold method (introduce the stimulus so weakly that the response does not occur, then increase it), the fatigue method (evoke the response until it stops) and the incompatible-response method (present the cue with something that forces an incompatible response). W. K. Estes later formalised Guthrie's ideas as stimulus-sampling theory.
3. Hull: drive reduction and the hypothetico-deductive system
Clark L. Hull (Principles of Behavior, 1943) built the most formal learning theory: a hypothetico-deductive system of postulates and theorems with precisely defined intervening variables linking stimulus inputs to response outputs. Reinforcement is drive reduction: a response that reduces a need-based drive is strengthened, increasing habit strength (sHr) as a function of the number of reinforced trials. Habit alone does not produce behaviour; it must be energised by drive (D), and the two combine multiplicatively.
| Term | Meaning |
|---|---|
| sEr = sHr × D (1943) | Reaction (excitatory) potential is habit strength multiplied by drive; if either is zero, no response occurs. |
| sEr = V × D × K × sHr (1952) | The later version adds V, stimulus-intensity dynamism, and K, incentive motivation — added to explain the Crespi effect, in which rats shifted to a larger reward ran faster at once, too fast for habit to have changed. |
| Inhibition: Ir and sIr | Reactive inhibition (a fatigue-like tendency not to repeat a response, which dissipates with rest — explaining spontaneous recovery and the advantage of spaced practice) and conditioned inhibition (learned not-responding). Effective reaction potential is sEr minus inhibition, subject to oscillation (sOr) and a threshold. |
| rg–sg and the goal gradient | The fractional anticipatory goal response — a small part of the goal response occurring earlier in the chain — produces stimuli (sg) that guide behaviour toward the goal; responses nearer the goal are learned more strongly (goal gradient). |
4. Classical conditioning: procedure and phenomena
In Pavlov's procedure an unconditioned stimulus (UCS, food) reliably elicits an unconditioned response (UCR, salivation). A neutral stimulus (a tone) is paired with the UCS until, alone, it elicits a conditioned response (CR); the tone is now a conditioned stimulus (CS). The timing of CS and UCS matters greatly.
| Procedure | Timing | Effectiveness |
|---|---|---|
| Delay (forward) | CS begins before the UCS and overlaps it | Most effective; the optimal interval for many responses is about half a second |
| Trace | CS ends before the UCS begins | Effective with short gaps; weaker as the gap grows |
| Simultaneous | CS and UCS begin together | Weak |
| Backward | UCS precedes the CS | Little or no excitatory conditioning; the CS may become inhibitory |
- Acquisition, extinction and spontaneous recovery: presenting the CS without the UCS extinguishes the CR, but after a rest it reappears — extinction is new inhibitory learning, not erasure. Disinhibition: a novel stimulus during extinction briefly restores the CR.
- Generalisation and discrimination: stimuli similar to the CS elicit the CR along a gradient; reinforcing one stimulus but not another produces discrimination. Pavlov's "experimental neurosis" followed an impossible circle–ellipse discrimination.
- Higher-order conditioning (a CS1 serves as the UCS for a new CS2) and sensory preconditioning (two neutral stimuli are paired first, then one is conditioned, and the other also elicits the CR).
- Latent inhibition (pre-exposure to the CS alone retards later conditioning), overshadowing (the more salient of two CSs gains more strength) and blocking (Kamin, 1969: a CS already predicting the UCS prevents a new CS added to it from being conditioned).
- Conditioned emotional responses: Watson and Rayner (1920) conditioned fear of a white rat in "Little Albert" by pairing it with a loud noise, and the fear generalised to other furry objects.
5. Related issues: contingency, preparedness and the Rescorla–Wagner model
Three findings overturned the view that conditioning is automatic stamping-in by contiguity. Rescorla (1968) showed that what matters is contingency: a tone paired with shock produced no fear if shocks were just as likely without the tone, because then the tone predicted nothing. Garcia and Koelling (1966) showed that rats readily associated taste with illness and light-and-sound with shock, but not the reverse, and taste aversion forms in a single trial with delays of hours — evidence for biological preparedness (Seligman, 1970) and "belongingness". Kamin's blocking showed that a CS is learned only if the UCS is surprising.
The Rescorla–Wagner model (1972) formalises this: on each trial the change in associative strength is ΔV = αβ(λ − ΣV), where λ is the maximum strength the UCS can support, ΣV is the total strength of all CSs present, and α and β are learning-rate parameters for the CS and UCS. Learning is driven by the prediction error (λ − ΣV). Worked: with αβ = 0.5 and λ = 1, on trial 1 V rises from 0 by 0.5 × (1 − 0) = 0.5; on trial 2 it rises by 0.5 × (1 − 0.5) = 0.25, to 0.75; on trial 3 by 0.125, to 0.875 — a negatively accelerated acquisition curve. In blocking, the old CS already makes ΣV ≈ λ, so the error and hence learning about the new CS are near zero. Pavlov's own stimulus-substitution view (the CS comes to act like the UCS) is contradicted by drug-tolerance findings: Siegel showed that cues for a drug elicit a compensatory response opposite to the drug's effect.
Key takeaways
- Thorndike: trial-and-error learning in puzzle boxes, connectionism, and the laws of effect (later truncated), exercise (abandoned) and readiness.
- Guthrie: contiguity is sufficient, learning is complete in one trial, movements are learned, reward merely preserves the last response, and habits are broken by the threshold, fatigue and incompatible-response methods.
- Hull: drive reduction strengthens habit, sEr = sHr × D (later V × D × K × sHr), reactive and conditioned inhibition, and the fractional anticipatory goal response.
- Delay conditioning is most effective and backward least; extinction, spontaneous recovery, generalisation, discrimination, higher-order conditioning, latent inhibition, overshadowing and blocking are the core phenomena.
- Contingency (Rescorla), preparedness (Garcia; Seligman) and blocking (Kamin) show that conditioning is learning to predict; Rescorla–Wagner's ΔV = αβ(λ − ΣV) makes prediction error the engine of learning.
Practice questions (10)
Attempt each one before opening the answer. Every explanation names the tempting wrong option as well as the right one, because that is where marks are lost.
Which theorist held that an association reaches full strength on a single pairing of stimulus and movement, and that reward works only by changing the stimulus situation?
Show answer
Answer: D — Edwin Guthrie
Guthrie's contiguity theory proposes one-trial learning and treats reward as protecting the last response from being replaced. Hull made drive reduction the basis of reinforcement, Thorndike the satisfying state of affairs.In Hull's 1943 system, what happens to reaction potential when a well-trained animal is completely satiated (D = 0)?
Show answer
Answer: C — It becomes zero, because habit and drive multiply
sEr = sHr × D, so with D = 0 the product is zero whatever the habit strength: learning is present but not performed. This multiplicative rule is the core of Hull's performance equation.Which of the following are among the methods Guthrie proposed for breaking a habit? Select all that apply.
Show answer
Answer: A — Threshold method; C — Incompatible-response method; D — Fatigue method
Guthrie described the threshold, fatigue and incompatible-response methods, each of which makes the unwanted response fail to occur in the presence of its cue. The token economy is an operant technique from later behaviour modification.Rats first learn that a tone predicts shock. A light is then added so that tone and light together precede the shock. Later the light alone produces little fear. This is:
Show answer
Answer: D — Blocking
Kamin's blocking: because the tone already predicts the shock, the shock is not surprising and the added light acquires little strength. Overshadowing involves two new stimuli of unequal salience conditioned together from the start.In the Rescorla–Wagner model, αβ = 0.5 and λ = 1, and a new CS starts with V = 0. What is its associative strength after two conditioning trials? Type a number.
Numerical answer — type the value.
Show answer
Answer: 0.75
Trial 1: ΔV = 0.5 × (1 − 0) = 0.5, so V = 0.5. Trial 2: ΔV = 0.5 × (1 − 0.5) = 0.25, so V = 0.75. Each increment is smaller because the prediction error shrinks.The classical-conditioning arrangement in which the CS begins before the UCS and continues until the UCS appears, generally the most effective, is:
Show answer
Answer: A — Delay conditioning
In delay (forward) conditioning the CS precedes and overlaps the UCS and so predicts it best. In trace conditioning a gap separates them, and simultaneous and backward pairings produce weak or no excitatory conditioning.Garcia and Koelling (1966) found that rats readily associated taste with illness but not with shock. This finding is most closely associated with the concept of:
Show answer
Answer: B — Biological preparedness
Some CS–UCS combinations are learned far more readily than others because evolution has prepared animals to link tastes with illness; Seligman called this preparedness. The finding also challenged the idea that any CS can be linked equally to any UCS.Assertion (A): A CS that is paired with a UCS on every trial may still produce no conditioning. Reason (R): According to Rescorla (1968), conditioning depends on the contingency between CS and UCS — whether the CS changes the probability of the UCS — not on the number of pairings. Choose the correct option.
Show answer
Answer: A — Both A and R are true, and R is the correct explanation of A
Both are true and R explains A: if the UCS is equally likely without the CS, the CS carries no information, so even consistent pairings produce no conditioned response.Thorndike's revision of the law of effect after 1930 held that:
Show answer
Answer: C — Reward strengthens connections, but punishment does not weaken them to the same degree
The "truncated" law of effect kept the strengthening effect of satisfiers but dropped the symmetrical weakening effect of annoyers. The law of exercise, the idea that practice alone strengthens, was abandoned at the same time.Which findings show that extinction does NOT erase the original conditioned association? Select all that apply.
Show answer
Answer: B — Disinhibition by a novel stimulus; C — Spontaneous recovery after a rest; D — Rapid reacquisition when pairings resume
If extinction erased learning, the CR could not return after rest, reappear with a novel stimulus, or be relearned faster than originally. Latent inhibition concerns pre-exposure before conditioning, not extinction afterwards.