Hull-Spence, Rescorla-Wagner & Attention theories (L3,5&6)
Hull-Spence Model
this model depicts the idea that the US becomes less surprising on every trial (the participant is better at predicting it), Assumes that CSs are learned about independently, cannot predict conditioned inhibition and blocking effects - Strength of association (V) determines the strength of the conditioned response (CR) i.e. [ A = light -> US = food ] Strong connection = Strong response.
Selective Attention & Blocking
Unlike the Rescorla-Wagner explanation for blocking, which states that blocking occurs because of associative surprise, Mackintosh suggests that; - Attention to the blocking stimulus (A) increases during Phase 1, leaving little attention available for the blocked stimulus (B) - which is in line with the inverse law of selective attention. - They learn that A is important, therefore giving it a lot of attention. - So when B is presented, there is little attention towards it. However, in a condition where both are presented at the same time - A + B equally fight for the attention of the animal.
Rescorla-Wagner Model
a cognitive model of classical conditioning and an updated version of Hull-Spence; it states that the strength of the CS-US association is determined by the extent to which the unconditioned stimulus is unexpected. Can predict conditioned inhibition and blocking effects.
Blocking
(Kamin 1969) [ A -> US | A*B* -> US = | *B* (weak CR) ] - 'B' in the blocking condition develops a weak association with the 'US' because 'A' already signals the 'US' (learning about 'B' has been blocked by 'A') - Learning about 'B' is not independent of 'A'. *Hull-Spence & blocking* - Completely ignores the presence of 'A' - Predicts that 'B' will gain associative strength i.e. 'B' will not be blocked by 'A' (incorrect) - Can only focus on one stimuli at a time and does not allow them to interact with one another. *Rescorla-Wagner & blocking* - Acknowledges both 'A' & 'B'. - Predicts that 'B' will stay neutral i.e. 'B' will be blocked by 'A' (correct) - this model can predict blocking because *all* CS's contribute to the expectation of the US.
Selective Attention Theory
(Mackintosh 1975) Conditioning is not a one stage process as proposed by the Hull-Spence & Rescorla-Wagner where you just 'update associations' and automatically pay attention to *all* stimuli presented. This model acknowledges that some stimuli might be ignored. (Stage 1) You need to pay attention to some of the stimuli that are presented. (Stage 2) Then form associations between stimuli that are attended to. *Inverse law of selective attention* There is a limited attentional resource. If one stimulus receives a lot of attention, then all other stimuli that are also present will be attended to a lesser extent. *Attention allocation rules* If a stimulus is attended to and is reinforced (i.e., it is followed by an unconditioned stimulus), then it will be more likely to be attended again. If the stimulus is not reinforced, then attention will decline.
The Summation Test
(Testing Inhibition #1) A -> US | A*B* -> no US | = C*B* (weak CR) ..............................C -> US | = C (strong CR) - All animals get trained with A, AB, and C, and they are all tested on C and CB. - If B is inhibitory, then it should suppress the expectation that the US should follow. This inhibition should transfer to a novel compound: CB. - C by itself should generate a strong conditioned response, and B's inhibition should reduce it on a CB test.
The Retardation Test
(Testing inhibition #2) A -> US | A*B* -> no US | *B* -> US (slow learning) ..................................................D -> US (fast learning) - If a stimulus is a true conditioned inhibitor it must have the opposite motivational properties to those of an excitor so it should be relatively difficult to turn an inhibitor into and excitor. - Thus, in a retardation test for conditioned inhibition, the supposed inhibitor is paired with the US and the rate of conditioning is compared with that of a novel CS (D) - B should be 'retarded' (slower) in acquiring a conditioned response because it used to signal the absence of reinforcement. So it has to lose its inhibition before it becomes excitatory.
Conditioned Inhibition (Hull & Rescorla)
*Hull-Spence Model* - Predicts that B will stay neutral. - Assumes that CSs are learned about independently. - Cannot predict conditioned inhibition and blocking effects. *Rescorla-Wagner Model* - Predicts that B will lose associative strength and will become inhibitory (correct prediction). - Learning about one CS might influence learning about other CSs. - It predicts conditioned inhibition and blocking effects.
Mackintosh vs. Pearce-Hall
*Mackintosh* • Two stages: pay attention, then update associations. • Attention is proportional to associative strength (V). • Stimuli with a stronger associative strength receive more attention. *Pearce-Hall* • Two Stages: pay attention, then update associations. • Attention is proportional to surprise (l-V). • Stimuli that are followed by unpredictable outcomes receive more attention.
Kaye & Pearce
- Measured attention by assessing Rat's orienting responses to a light (CS). They often stand on their hind paws and sniff the light. - However, these orienting responses gradually diminish if the light is never followed by something important (i.e. Habituation) *Continuous Reinforcement Group:*- Light always followed by US (100%). *Partial Reinforcement Group:*- Light followed by US on half of the trials on a random schedule (50%). *No Reinforcement Group:* -Light never followed by US (0%). *Mackintosh* - Group Continuous should exhibit most orienting responses because the Light in this group is the strongest predictor of the US, followed by Group Partial. - Attention should be proportional to associative strength: Light (100%) > Light (50%) > Light (0%). *Pearce-Hall* - Group Partial (50%) should exhibit most orienting responses because the US is unpredictable in this group. - Attention should be proportional to surprise (or unpredictability of the US): Light (50%) > Light (100%) = Light (0%). (correct)
Partial Reinforcement Extinction Effect
Continuous Reinforcement Group: Light always followed by US (100%). Partial Reinforcement Group: Light followed by US on half of the trials on a random schedule (50%). - Instead of measuring attention, it's measuring extinction. How quickly it loses it's conditioned response to the light (CS). Both groups undergo Extinction following the Acquisition phase: Light no longer followed by the US. Continuous | Light -> US (100%) | Light -> no US (fast extinction) Partial | Light -> US (50%) | Light -> no US (slow extinction) -The conditioned response to the Light persists in extinction for a longer period of time in the Partial Group. - This is inconsistent with the Rescorla-Wagner model: The Light should be weaker in the Partial Group, hence, during the Extinction Phase it should reach zero associative strength faster. - Also inconsistent with the Pearce-Hall model: The Light in Group Partial should get most attention, hence it should extinguish faster. -The effect is consistent with Mackintosh's theory: Group Continuous: The continuously reinforced Light should receive more attention, hence learning about it should be faster. This is true of BOTH acquisition and extinction: In extinction, the Light should lose its strength faster.
Experiment summary
Latent Inhibition: Mackintosh & Pearce-Hall. Pearce-Hall (little to large): Only Pearce-Hall Kaye & Pearce: Only Pearce-Hall. Partial Reinforcement: Only Mackintosh. - Rescorla-Wagner cannot predict any of these effects.
Conditioned Inhibition
Learning that a 'CS' signals the absence of a 'US' [ A -> US | A*B* -> no US | = B ] - 'B' is a conditioned inhibitor: It cancels the US that normally follows 'A'. - If 'B' is presented by itself, the animal will not expect the 'US' - therefore meaning that 'B' signals the absence of reinforcement. Example: [Jane = fun | Jane + Lisa = No fun = | Lisa] - Lisa is inhibitor and signals that there's no fun. (Jane = excitatory, Lisa = Inhibitor for fun) - Jane + Lisa cancel each other out in the Rescorla-Wagner model. (neutral) *Summation test*: Pair a different US with Lisa (B) to see if she's really an inhibitor. *Retardation test*: Slow learning that Lisa will be fun because of your previous learning (turning an inhibitor into an excitor should take longer than turning a neutral cue into an excitor).
Selective attention & Latent inhibition
Mackintosh's model explains latent inhibition because it states that attention to the light diminishes during exposure because it is not reinforced - therefore, explaining the slower conditioning.
Latent Inhibition
Pre-exposed group -> Light -> Light -> food (slow conditioning) Control group ->..............................Tone -> food (fast conditioning) - (Habituation) If a CS (light) is repeatedly presented without consequence during the exposure phase, then animals subsequently learn to associate it with a US more slowly (compared to a novel CS, the tone) - The *Rescorla-Wagner* and *Hull-Spence* models cannot explain this effect: According to these models, nothing happens to the Light during the Exposure phase (its associative strength stays zero), so it should condition as fast as the Tone (which also starts with an associative strength of zero). They cannot predict habituation and how if effects conditioning. - The animals habituate to the Light during the Exposure phase, causing them to pay less attention to the Light than to the Tone during training, explaining the slower conditioning to the Light.
Pearce-Hall Attentional Model
Pretrain | A -> little US | A -> large US (slow learn) No Pretrain |.....................| A - > large US (fast learn) - Both the Rescorla-Wagner model and Mackintosh's attention model incorrectly predict faster learning in Group Pretrain: *Rescorla-Wagner:* Group Pretrain should have a head start at the beginning of Phase 2 since A already acquired some strength in Phase 1. *Mackintosh:* Group Pretrain should already pay attention to A at the beginning of Phase 2 since A was a good predictor of the US in Phase 1. - However, Group Pretrain learns more slowly in Phase 2, perhaps because the animals in Group Pretrain stopped paying attention to A in Phase 1 once they knew what A predicted. *Assumptions of Pearce-Hall* (1) Learning is proportional to attention (like Mackintosh). (2) Attention is proportional to the amount of surprise. (3) Unpredictable events enhance attention. (4) Once a US is fully predicted by a CS, attention to the CS diminishes.
