Classical Conditioning: Pavlov, Contingency and Applications

Concept and theories of learning (Behaviourists models); The Processes of extinction, discrimination and generalization.

Official Syllabus: Paper I · Unit 6 Learning

“Learning is a relatively permanent change in behavioural potentiality that occurs as a result of reinforced practice.”

— Kimble (1961)

Classical conditioning is the simplest arrangement in which an organism learns that one event predicts another. It gave psychology its first precisely measurable unit of learning, and the century of work since has turned a salivating dog into an account of how fears are acquired, how preferences form, how drug tolerance builds and how anxiety is treated.

Concept of learning, and what it is not

• Learning is inferred from performance, never identified with it: a person may have learnt a route and not walk it. The definition’s “behavioural potentiality” carries that distinction.
• Maturation is excluded. A child walks when the nervous and muscular systems are ready, and no practice produces walking before that point.
• Training is narrower than learning, and the distinction is worth stating precisely: training is a deliberately arranged programme aimed at a specified competence, whereas learning is the underlying change that may occur with or without any such arrangement, intentionally or incidentally. Training is one route to learning; most learning is not training.
• Reflexive behaviour is the raw material. A reflex is an involuntary, unlearned stimulus-response connection — salivation to food, an eyeblink to a puff of air. Classical conditioning does not create reflexes; it transfers an existing reflex to a new signal, which is why only responses that already exist as reflexes can be classically conditioned.

Pavlov’s experiment and the four elements

• Ivan Pavlov (1849–1936), a Russian physiologist, won the Nobel Prize in 1904 for work on digestion. Conditioning began as an interference in that work: his dogs salivated at the sight of the dish and the attendant’s footsteps, which he first called “psychic secretions”.
• His decisive move was methodological — treating anticipatory salivation as a measurable reflex rather than a mental event, and recording drops of saliva against a controlled signal.
• Unconditioned stimulus (UCS) — elicits a response with no prior learning. Food in the mouth.
• Unconditioned response (UCR) — the reflex itself. Salivation to food.
• Conditioned stimulus (CS) — the formerly neutral stimulus that, after pairing, elicits the response. The bell.
• Conditioned response (CR) — the learned response to the CS. Typically weaker than the UCR, and sometimes compensatory, opposing rather than mimicking it.

Before:  [Food UCS] -> [Saliva UCR]
         [Bell  NS] -> (nothing)

During:  [Bell] + [Food] -> [Saliva]

After:   [Bell  CS] -> [Saliva CR]

The bell changes status, not the food. Food remains the UCS throughout.

• Conditions that make it work: the CS must precede the UCS; the interstimulus interval should be short, around half a second for many reflexes; the CS should be distinctive; and pairings should repeat, though one intense trial suffices for taste aversion.
• The arrangements, by timing: short-delay conditioning, where the CS begins just before the UCS and overlaps it, is most effective; trace conditioning, where the CS ends before the UCS starts, is harder; simultaneous is weak and backward conditioning is weaker still or inhibitory.
• Latent inhibition (Lubow and Moore, 1959) — a stimulus made over-familiar by unpaired presentations conditions poorly. Novelty is part of what makes a signal usable.

The processes of extinction, discrimination and generalization

• Acquisition — the stage in which pairing builds the CR. Gains are largest on early trials and flatten towards an asymptote.
• Extinction — presenting the CS without the UCS weakens the CR until it ceases. It is not erasure: what is learnt is a new inhibitory association competing with the original, and three findings show the original survives.
  • Spontaneous recovery — after a rest interval the CR returns, weakened, with no further pairings.
  • Renewal — an extinguished CR reappears when testing moves outside the extinction context (Bouton, 1993).
  • Rapid reacquisition — re-pairing restores the CR far faster than original learning took.
• Stimulus generalization — stimuli resembling the CS evoke the CR, more strongly the closer the resemblance, producing a generalization gradient. This is what makes conditioning useful outside the laboratory, where situations never repeat exactly.
• Stimulus discrimination — the complement. When one stimulus is reinforced and a similar one is not, responding becomes selective. Pavlov used discrimination training to measure animals’ sensory acuity.
• Higher-order conditioning — an established CS serves as a UCS for a new signal. Clapping paired with the bell comes to elicit salivation though it was never paired with food, which is how conditioning reaches words, money and brand names.
• Conditioned inhibition — a stimulus signalling the absence of the UCS actively suppresses the CR, showing that negative relations are learnt too.

The partial return after rest is the evidence that extinction suppresses an association rather than deleting it.

Contingency and the cognitive approach

The classical view held that contiguity — pairing in time — builds an association. From the late 1960s that was dismantled, and what replaced it treats conditioning as the learning of predictive information.

• Rescorla (1968) varied the probability of the UCS in the CS’s presence and in its absence, independently. Conditioning tracked the difference between those probabilities, not the number of pairings.
  • At zero contingency — the UCS equally likely with or without the CS — animals receive many pairings and learn nothing. Contiguity is not sufficient.
  • At negative contingency the CS becomes a conditioned inhibitor despite never being paired with the UCS. Contiguity is not necessary either.
• Kamin (1969) showed the same within one preparation. Train a light to signal shock, then present light-plus-tone with shock: little is learnt about the tone, because the light already predicts the shock and the tone is redundant. Organisms learn from what is surprising.
  • Overshadowing, described by Pavlov and investigated by Kamin, is the related case where the more salient element of a compound is learnt at the weaker one’s expense.
  • Wagner, Logan, Haberlandt and Price (1968) made the point as relative validity: when cues compete, the better predictor wins.
• Rescorla and Wagner (1972) formalised it. Associative strength changes in proportion to prediction error — the gap between the UCS that occurred and the UCS already predicted by the stimuli present.
  • A fully predicted outcome gives zero error and no learning, however many pairings follow. Blocking falls straight out.
  • All stimuli present, the background context included, compete for a limited pool of associative strength.
  • Schultz (1997) found midbrain dopamine neurons firing to unexpected rewards and falling silent to predicted ones — a physiological implementation of exactly this quantity, and the bridge to machine reinforcement learning.
• Rescorla (1988) put the correction bluntly: conditioning is not what most psychologists think it is. The organism extracts the predictive structure of its environment rather than accumulating pairings.

S-R or S-S: what is associated

• The S-R position (Hull, Guthrie) holds that the CS connects directly to the response.
• The S-S position (Tolman) holds that the CS connects to a representation of the UCS, and the response follows from what the UCS now means.
• UCS devaluation decides it. Condition a CS signalling food, then make that food aversive by pairing it with illness in the CS’s absence. The CR to the CS then falls — though nothing touched the CS-response link.
• On an S-R account the response should persist. Its decline shows the CS was pointing at the food itself, so Pavlovian conditioning is better read as S-S learning — knowledge, not habit. This fits the contingency findings directly.
• Sensory preconditioning supplies the second line of evidence. Pair two neutral stimuli with each other — a tone and a light, with no UCS present — then condition only the light to food. The tone now elicits the CR, though it was never paired with food and no response was ever reinforced in its presence.
  • On an S-R account nothing could have been learnt in the first phase, because no response and no reinforcer were involved. The animal evidently learnt a stimulus-stimulus relation and then chained it.
• Higher-order conditioning works the same way and is the mechanism by which conditioning reaches stimuli with no biological significance at all: a second-order CS acquires its power from a first-order CS rather than from any UCS, which is how words, logos and money come to carry conditioned affect.
• The form of the CR is further evidence against a simple reflex-transfer account. Where the UCR is a physiological disturbance, the CR is often compensatory — opposing the UCR rather than copying it — which makes sense only if the CS signals the UCS’s imminent arrival and the body prepares for it. This is the mechanism behind conditioned drug tolerance, treated under Applications.

Biological constraints

The behaviourists assumed equipotentiality, that any stimulus could associate with any other at roughly equal rates. It is false.

• Garcia and Koelling (1966) gave rats water that was simultaneously saccharin-flavoured and accompanied by light and noise, so both cues arrived with the same lick. Groups then received either illness (X-irradiation or lithium chloride) or footshock.
  • Rats made ill avoided the taste only. Rats shocked avoided the audiovisual cue only.
  • The double dissociation rules out one cue being merely more noticeable: each was learnt about, but only with the matching consequence. Which cue associates with which outcome depends on their biological relationship.
  • Limitation: the original lacked some controls, since supplied by later work.
• Garcia, Ervin and Koelling (1966) showed taste aversions forming across hours between tasting and illness — impossible on a contiguity account calibrated in seconds, and sensible for an organism whose poisons act slowly.
• Conditioned taste aversion therefore has distinctive properties: one-trial acquisition, long tolerable delays, cue-to-consequence specificity, and resistance to extinction.
• Seligman (1970, 1971) generalised this as preparedness, a continuum along which organisms are prepared, unprepared or contraprepared to form particular associations by evolutionary history.
  • Applied to phobias it explains their non-random distribution: snakes, spiders, heights and blood are feared easily, while electrical sockets and traffic, objectively more dangerous now, are not. The prepared fears match ancestral threats.

Conditioned emotional responses

• Watson and Rayner (1920) conditioned an eleven-month-old infant, Albert B., to fear a white rat by striking a steel bar behind him as he reached for it. Reported generalization extended to a rabbit, a dog, cotton wool and a fur coat.
• John B. Watson, founder of behaviourism, took this to show that emotional dispositions are built by conditioning rather than inherited.
• The study will not bear the weight placed on it, and the criticisms are the examinable part: one subject, no control, outcome measures that were observers’ impressions, inconsistent avoidance across sessions, a confounding of classical with instrumental conditioning, and no deconditioning — Watson left before any was attempted.
• Harris (1979) traced how the study has been distorted in retelling, including by Watson himself, and judged the data uninterpretable.
• The underlying claim survived on better evidence. LeDoux (1996) mapped the amygdala circuitry of conditioned fear, making it one of the best-specified phenomena in the subject — with the controls Watson and Rayner lacked.
• Rachman (1977) showed conditioning is only one of three pathways to fear, alongside vicarious acquisition by observing another’s reaction and verbal transmission of threat information. Many phobic patients report no conditioning episode at all.

Applications

• Jones (1924), with Watson as consultant, treated a boy called Peter who feared a rabbit by bringing it progressively nearer while he ate a liked food — counterconditioning, and the first demonstration of behaviour therapy.
• Wolpe (1958) developed systematic desensitisation from that logic, on the principle of reciprocal inhibition: relaxation and anxiety cannot occupy the organism at once. Three components — relaxation training, an anxiety hierarchy, and graded pairing of each step with relaxation. Jones’s work is the precursor; the standardised therapy is Wolpe’s.
• Modern exposure therapy descends from both, and its rationale has shifted: Craske (2014) reframed extinction as inhibitory learning, so treatment now varies contexts and violates expectancies to make safety learning retrievable, rather than aiming to reduce anxiety within a session.
• Aversion therapy reverses the arrangement, pairing an unwanted attraction with an unpleasant UCS — disulfiram in alcohol dependence works this way. Effects are often context-bound, and the ethics are treated below.
• Likes and preferences. Staats and Staats (1958) paired nonsense syllables and national names with evaluatively loaded words and shifted participants’ ratings — the founding demonstration of evaluative conditioning, by which a neutral object paired with something liked or disliked acquires that valence.
  • It operates with little awareness of the pairing, which is why incidental and subliminal exposures move evaluations.
  • Advertising is its systematic use: a product paired with music, humour or admired figures becomes the CS for the affect those already command, and celebrity endorsement is higher-order conditioning in commercial form.
  • The same mechanism builds prejudice, as repeated pairing of a group with negative portrayal conditions evaluations the holder may not endorse on reflection.
• Conditioned drug tolerance. Siegel (1976) showed that cues preceding a drug elicit compensatory responses opposing it, so the same dose has less effect in a familiar setting.
  • The corollary is well attested: a long-tolerated dose can prove fatal in an unfamiliar environment, where the protective conditioned response is absent.
  • The same mechanism drives cue-triggered craving and relapse on return to old surroundings.
• Ader and Cohen (1975) paired saccharin with an immunosuppressive drug; saccharin alone later suppressed immune response — conditioning reaching a system assumed autonomous, and the founding result of psychoneuroimmunology.
• Anticipatory nausea in chemotherapy is conditioning in a clinical setting: the cytotoxic drug is the UCS, the ward’s sights and smells become CSs, and nausea occurs before infusion. A substantial minority of patients report it, rising with successive cycles, and it persists despite antiemetics because it is learnt rather than pharmacological. Overshadowing interventions, giving a novel strong-tasting drink at infusion so the salient new cue absorbs the association, follow directly from the theory.

Classical and operant conditioning compared

• The distinction is real but not absolute, and that is the better answer to whether they differ basically.
  • Both show acquisition, extinction, spontaneous recovery, generalization and discrimination — the same functional laws, documented by Pavlov (1927) and Skinner (1938) alike.
  • Both are now analysed in terms of prediction and contingency rather than mechanical strengthening.
  • Brown and Jenkins (1968) demonstrated autoshaping: pigeons peck a key that merely predicts food, with no requirement to peck, so a classically conditioned signal generates an apparently instrumental act.
  • Mowrer (1947) two-factor theory needs both — classical conditioning establishes the fear, operant conditioning maintains the avoidance that reduces it. Persistent phobic avoidance requires the pair.

Criticisms and limitations

• Equipotentiality is false, so the general laws of learning the behaviourists sought are constrained by species-specific predispositions.
• Awareness. In humans, conditioning is often accompanied by and sometimes depends on explicit knowledge of the CS-UCS relation, and can be abolished by instruction. A purely non-cognitive account is untenable.
• Not all fear is conditioned — Rachman’s (1977) three pathways make the conditioning account one route among several.
• Reductionism. Explaining human behaviour by reflex association omits language, meaning and intention. It explains a conditioned nausea well and a career decision not at all.
• Preparation specificity. Findings from salivation, eyeblink, fear and taste aversion differ systematically, so “classical conditioning” names a family of processes rather than one.
• Ethics. The Watson and Rayner (1920) study would not be approved today: an infant was made fearful and never deconditioned.
• Cultural narrowness. Parameters were established almost entirely on Western samples and a few laboratory species, while the meaning of a CS — a food, an animal, a gesture — is culturally constituted in ways the paradigm does not model.

Indian context

• De-addiction practice. Aversion-based approaches including disulfiram are used in Indian de-addiction settings, and Siegel’s (1976) context-bound tolerance predicts their weakness: gains made in a treatment centre are vulnerable on return to the environment where use occurred.
  • This is the psychological case for the community-level follow-up the National Action Plan for Drug Demand Reduction is built around, and against treating detoxification as completion.
• Health communication. The pictorial warnings mandated on tobacco packaging aim to condition an aversive response to a product cue, and their design — graphic, salient, rotated periodically — tracks CS salience and latent inhibition closely.
• Ethical limits of aversive conditioning are a live Indian legal question. The Madras High Court in S. Sushma v. Commissioner of Police (2021) directed action against conversion “therapy”, and the National Medical Commission moved to prohibit it as professional misconduct; the Mental Healthcare Act, 2017 (Ministry of Health and Family Welfare) sets the consent framework within which any aversive procedure must now operate.