Contingency And Contingency Theory Of Conditioning

69

70

71

Rescorla – Wagner Model 1.

There is a maximum associate strength that can develop between a CS & a US - limit determined by US

2.

Associative strength increases with each training trial – the amount of increase on each trial is dependent on level of prior training

3.

Rate of conditioning varies depending on CS & US used.

4.

The level of conditioning on a particular trial is influenced not only by the amount of conditioning to that stimulus, but also by the level of conditioning to other stimuli associated the US (i.e. blocking)

- same l (asymptote) - a & b are different (rates of change)

72

- different l (asymptote) - a & B same (rates of change)

direction of learning is determined by the relationship between V and V max. 1. 2.

When V is less than V max then associate strength increases when V is greater than V max then associative strength decrease

73

3.

When V = V max then associative strength does not change.

-

eg –rat receives 1OV shock – if not expecting shock then ↑ in fear – if it expected 2OV shock then Ø in fear – if expected OV shock no change in fear level

Rescola/Wagner Explaining the old: Equation can predict the shape of the acquisition curve. Equation can explain extinction, the decrease in associative strength with presentations of the US without the CS. Explains Blocked (as shown above) Model can also explain the effects of contingency. Why does presenting the US without a CS interfere with conditioning?

74

-

-

-

The US is not presented in a Vacuum! – there are many stimuli from the environment that can be associated with the US. – these environmental cues are also present when CS predicts US so...CSe Æ US, e Æ US, CSe Æ US, e Æ US, initially conditioning occurs to both CS & e, eventually although VCS and Ve are each less than Vmax, together they exceed Vmax so the US will actually Ø the fear associated with CS & e, and when e Æ US is next presented it is below V max so will ↑ associative strength – so background will be stronger association than CS! seen in ↑ fear of apparatus by animal no unsignalled US’s not so much fear shown,

1.

the strength of conditioning depends not just on the characteristics of the US, but also on whether it was expected.

2.

Conditioning occurs not just to the nominal CS but also to background cues that may be present. (Context conditioning) -

Predicting the New Theory also predicts that under some circumstances CS’s will lose associative strength despite being paired with US! - keep in mind model views learning as an adjustment of expectancies to outcomes.

75

Phase 1 A Æ US

Phase 2 [A+B] Æ US

Test A ?

B Æ US -

B

common sense says no change in A or B, but R/W model predicts that if an animal receives less of an US than expected – then a decrease in US expectation or Associative Strength is predicted! END OF Phase 1 VA -= l, VB = l Start Phase 2 VAB = VA + VB = 2l End of Phase 2 VAB = l; VA = _ l, VB = _ l

Model Limitations - some predictions incorrect Latent Inhibition Lubow & Moore (1959) Gp 1 Gp 2 1) Flashing light – 2) light Æ shock light Æ shock What effect will pre-exposure to light have? R/W model says when light is presented alone no conditioning will occur because no US is presented

76

-

V(max) = 0 when no US presented DV = ab (Vmax – V) = ab (0 - 0) = 0!

So exposure to light should have no effect – but Lubow & Moore found conditioning significantly slower in the group pre-exposed to the light - they called this Latent Inhibition because they believed that the CS becomes inhibitory - however no evidence for inhibition – just difficult to condition - perhaps habituation decreases selective attention – maybe attention should be build into a model! Configural Learning - model says – associative strength of a compound stimulus is equal to the sum of its parts A US = .5 Vab = Va + Vb = .5 + .5 = 1.0 B US = .5 - so predicts when A & B presented together should get greater responding. - Bellingham, Gillete – Bellingham & Kehoe (1985) tone Æ food light Æ food tone/light Æ - rats responded to each alone on 90% of trials - when presented together responding dropped to 30% - rats behaved as if compound was a unique, novel stimulus – now termed a “configuration”

77

Occasion Setting Rescorla 1985 trained pigeons Noise Æ light Æ food Light Æ ______ -noise “set the occasion” for when the light would be followed by food. - Rescorla then presented light 144 without foodunder normal CC this would lead to extinction. In this case when noise and light were presented together the birds responded three times more often then if light alone was presented. - So presenting light alone had no effect on its role as occasion setter. - Holland hypothesized that occasion setters are not directly associated with the US, but serve to modulate or facilitate the association between the CS and the US. So the Wagner/Rescorla model accounts for many aspects of learning, but not all. There are several alternate theories: MacIntosh’s Theory of Attention! Masses of info coming in all the time - can’t attend to it all – must be selective - cocktail party phenomenon – can attend to 1 conversation despite large amt of competing input

78

-

theories of attention propose that on a conditioning trial when several stimuli are presented together the subject will attend to only 1 (or only a few) of those stimuli, and learning will occur only for those stimuli attended to.

Mackintosh ’75 proposed a theory of attention and classical conditioning that is the major competitor of R/W model. -

-

animals seek out info that predicts the occurrence of biologically significant events (US’s) once it has found a stim that reliability predicts an event – it ignores all other stimuli thus the animal plays an active role in the conditioning process. Animal must recognize the correlation (or lack of correlation) between events (CS + US) CS Preexposure – animal learns CS is irrelevant – will stop attending to it + will have difficulty learning that CS is correlated with US. “learned irrelevance” evidence for this – Baker & MacKintosh (1977) - significantly greater interference is produced by uncorrelated presentations between CS & US prior to conditioning than to either CS or US presentations alone.

79

Pearce & Hall (1980) Proposed a theory of CS effectiveness -

CS’s become ineffective whenever the US is already well predicted if the situation is changed so that the US is again surprising (i.e. by making the US more intense) the CS’s will quickly regain effectiveness

Comparitor theories -

other theories we’ve discussed have 2 things in common 1) predictions are based on trial by trial calculations and 2) assume the presence of one CS can interfere with a subjects learning about other CS’s (blocking)

-

Comparator theories (Gibbon & Balsam, 1981; Miller & Schachtman 1985) – assume that the animal compares the likelihood that the US will occur in the presence of the CS with the

80

-

likelihood that the US will occur in the absence of the CS. What’s important is the overall long-term correlation between CS & US comparison of CS and context does not affect the learning of the CR, but rather its performance. e.g. p (US/CS = .5 p (US/noCS) = .5

comparator theory says that this CS will not elicit a CR – NOT because CS has no excitatory strength – but because both the CS and the contextual stimuli have acquired equal strengths because both paired with US 50% of the time. -

theory assumes CS will not elicit CR unless it has greater excitatory strength than the contextual stimuli so assume that animal has learned something about CS – that the US sometimes occurs in its presence – but will not respond to CS unless it’s a better predictor of US than context.

Test – after condit keep animal in context w/o US does this extinguish context component and make CS better predictor? Yes – extinction of context does increase responding to the CS.

81

What is learned in C.C.? -

when a tone is paired with food what is it that is associated? Pavlov – simultaneous activation of tone & food centers results in the formation of a new pathway between them. so when tone presented the excitation it produces will be transmitted to the food center Hull (1943) – when tone is followed by salivation an association forms between tone & salivary centers – so tone will directly elicit salivation

Pavlov’s approach called S-S Theory – because it assumes association is between two stimuli – the CS & the US Hull’s approach is the S-R theory because it assumes a direct link between the CS and the response. To test – what if – after conditioning we make the food aversive? Pavlov - TONE Æ FOOD Æ SALIVATION - if FOOD ≠ Saliv then tone won’t either Hull

- TONE Æ SALIVATION - If food ≠ saliv – no change!

Holland & Straub 1979 – 1 Osec Noise Æ FOOD pellets

82

-

conditioning measured by activity level after condit activity to tone doubled after condit made food unattractive by pairing with rotation at high speed later present tone- what will happen with activity? - significant decrease in activity to tone so supports S-S theory -However – another test is 2nd order conditioning CS1 Æ US

CS2 Æ CS

SS Learning

SR Learning

CS1 ↑ CS2

CS2 Æ CR Ø CR

Test by extinguishing CS1 – what happens to CS2? - Results are mixed – in some experiments 2nd order condit. involves S-R learning – in others S-S learning - so research is focused on which conditions support which forms of learning?! Is the CS a signal or a substitute for the US? Tolman (1932) believed that the CS signaled food or set up the expectation of food “Oh boy, I’m going to get food?”

83

There is evidence to support the idea that the CS acts as a signal that the US is coming. Colwill and Motzkin (1994) CS1 Æ sucrose CS2 Æ food pellets When either CS presented rats approach the food hoper The researchers then “devalued” one of the US’s – this means they paired on the CS’s with lithium (illness= taste aversion). If the rats know which CS predicted which US then since CS1 predicted sucrose they should now avoid it, but still respond on CS2. This is what happened- providing evidence that the CS led to an expectation of a specific reward. Pavlov – CS-US centers linked, so excitation in CS produced excitation in US – so stimulus substitution “Oh boy what great food?” - Pavlov found that after pairing light with food when dog released from harness it tried to lick bulb Auto shaping (sign tracking) – pigeons responses to food and water are different - food – peck mouth open & eyes closed - water – peck beak closed, tongue down, eyes open - Jenkins & Moore (’73) – autoshaped to food or water – what form does peck to the key-light take. - support for stimulus substitution – pigeons pecked key as if it was food or H20!

84

However if stimulus substitution only way, then CR should always be identical to UR but this is not the case

85

Fear condit UR to shock – jump in air CR to shock – freeze UR to shock - ↑ in respiration CR to shock - Ø in respiration In salivation – chemical composition of saliva to US differs to saliva to CS -

Modified Stimulus – substitution model CS elicits innately programmed responses that prepare the organism for the forthcoming US animals have UR’s to important events, also have preparatory responses to prepare for events in some cases preparatory response = UR, in others it differs..

The Two-System Hypothesis -evidence for both stimulus substitution and expectation -hypothesis is that there are two systems of learning- a primitive stimulus substitution system and a more sophisticated cognitive system based on anticipation/expectation. There has been a gradual shift in thinking about CC- from a simple contiguity association idea to a more complex system of determining the most effective predictors of US’s. This means it is a sophisticated system of detecting

86

relationships so that we can learn to anticipate important events and be prepared for them. `Medical Diagnosis Gluck and Bower 1988- gave university students 250 sets of medical records (fictitious) -patients had some combination of: bloody nose, stomach cramps, puffy eyes and discolored gums. - students asked to decide which of two diseases the patient had. After each diagnosis they were told whether they were right or wrong. - after reading all 250 records students were shown each symptom and asked to estimate the probability whether the patient had disease 1 or disease 2. (the symptoms were designed so that they had predictive values ranging from 01. -this can be seen as CC with each symptom acting as a CS and the disease as a US. -if this is true then the Rescorla/Wagner model could be used to predict subjects probability estimates. -model’s predictions were more accurate than more complex cognitive information processing models. -