21 - Attention with learning

Question 1

Rescorla-Wagner recap

Answer 1

• Salience of CS, US & prediction error (actual vs expected)
  a) Established connection between CSa and US
  b) No learning with new CSb (when it’s paired with CSa and US)
• Learning helps us predict the future, to anticipate upcoming events
• This functional perspective suggests that the importance of [a CS], and hence the degree of processing it receives, should be related to its power as a predictor of [the US]

Question 2

Why this is incomplete? Latent Inhibition

Answer 2

• Latent Inhibition: Learning proceeds more slowly to a stimulus that has been pre-exposed compared to a novel stimulus
• This is not predicted by the Rescorla-Wagner model AT ALL
• The only thing that differs is the exposure to the CS

Question 3

Mackintosh’s attentional theory

Answer 3

• Background - Dickinson (1980): “The core of this idea can be illustrated by considering the latent inhibition phenomenon”
• “A theory which maintains that the processing of [the CS] depends upon its predictive power would argue that during the pre-exposure phase the animal learns that [the CS] predicts nothing of significance.”
• “As a result, its ability to command processing capacity in the learning mechanism is severely reduced, thus retarding any subsequent associative learning involving [the CS]”
• Mackintosh: If a cue is the best predictor of an outcome … then pay it more attention
• If a cue isn’t the best predictor of an outcome… then ignore it
• So, attention isn’t fixed – it adapts based on experience
• Blocking occurs because subjects learn to ignore the added element if it’s a redundant predictor of a reinforcement

Question 4

Changes in attention during blocking

Answer 4

• Blocking occurs because subjects learn to ignore the added elements if it’s a redundant predictor of reinforcement
• If this were true, interference with conditioning to the added element could only develop as subjects learned it was redundant
• Since this would require at least one trail, we should expect to find no evidence of blocking on this first trial
• Stage 1: light = shock
• Stage 2: light + tone = shock
• Test: tone?
• The DV in this exp is conditioned suppression (how much do scary stimuli supress responding)
• With this DV smaller numbers = more conditioning
• Results: higher suppression to tone as there was little conditioning
• Conclusion: tone was ignored during learning because it was redundant. No strong association between tone and chock was formed (blocking)

Question 5

Changes in the processing of cues: learned irrelevance

Answer 5

• Group 1: stage 1 – water, stage 2 – tone = water
• Group 2: stage 1 – tone, stage 2 – tone = water
• Group 3: stage 1 – tone & water randomly presented, stage 2 – tone = water
• Results: group 3 indicated poor learning
• But maybe the rats in the Tone/Water group 3 just learned in Stage 1 that the tone predicted No Water. Wouldn’t need to appeal to changes in attention to explain the results? WRONG
• New stage 2: light = water, tone + light = no water
• Results: rats were not good at learning the tone means no water, meaning this cannot be learnt
• Learned Irrelevance training slows excitatory and inhibitory learning
• Consistent with Mackintosh’s (1975) theory
• Attention declines to the tone in group 3 Tone/Water during stage 1

Question 6

Changes in the processing of cues: learned predictiveness in humans

Answer 6

• We learn to pay more attention to cues that have been useful (predictive)
• Learning phase: cues (e.g. risotto) are predictive of an outcome, others (e.g. brownie) are irrelevant
• Predictive cues = high attention, irrelevant cues = low attention
• Attention phase: triangle appears on the screen and pp’s have to respond quickly to its location
• Congruent = triangle appeared in the position recently occupied by the predictive cue (e.g. Risotto)
• Incongruent = triangle appeared in the position just occupied by the irrelevant cue (e.g. Brownie)
• Results: faster reaction times to the congruent condition since attention is already focused there
• Conclusion: we attend more to cues that have been predictive in the past

Question 7

Why all this is useful

Answer 7

1. Exploring regularities in the world
   - E.g. cross the road safety
   - Predictive cues: green man on traffic light, cars
   - Are attention is captured to a reliable predictor, not just the most salient (e.g. big famous buildings)
2. schizophrenia
   - Symptoms: Difficulty concentrating, disorganised thinking = disrupts that tells them to pay attention to regularities in the world

Question 8

Why Mackintosh’s theory is implausible

Answer 8

• Core idea: we increase attention to cues that are goof predictors
• It could be argued that there is something intuitively implausible about the central idea of Mackintosh’s theory
• The idea that animals learn about an event to the extent that it has been a reliable and good predictor in the past
• Rather, one might expect the animal to devote most of its processing capacity to analysing events whose predictive significance is uncertain in an attempt to discover relationships involving these events
• Animals should focus on uncertain or ambiguous cues, not predictable ones

Question 9

The Pearce and Hall theory

Answer 9

• We pay more attention to cues when their consequences are uncertain or surprising
• We pay less attention to cues when outcomes are predictable
• Attention is driven by how with your expectation are (i.e. big surprise = more attention)
• Mackintosh: attend to good predictors
• Pearce-Hall: attend to uncertain predictors

Question 10

Evidence for Pearce-Hall theory: negative transfer in rats

Answer 10

• Group 1: tone = weak shock then tone = strong shock
• Group 2: light = weak shock then tone = strong shock
• Again, the DV in this exp is conditioned suppression
• Smaller numbers = more fear = more conditioning
• Results: Learning the Tone → Strong Shock association was slower in group 1 and faster in group 2 (negative transfer)
• Totally inconsistent with Mackintosh (1975) as a previously predictive cues didn’t enhance learning
• Entirely consistent with Pearce Hall (1980) as a previously predictive cue gets less attention making new learning slower

Question 11

Evidence for Pearce-Hall theory: uncertainty induced attention in humans

Answer 11

• Pp were sat in front of an eye tracker, pretending to be scientists
• Had to pretend to add molecules together and guess which ‘monster’ they would create
• A is a perfect predictor of outcome 1, B is a perfect predictor of outcome B
• Results: Overt attention/ more time on the predictive cues (a la Mackintosh, 1975) – goof predictors attract more attention
• Also, uncertainty (a la Pearce & Hall, 1980) – when outcomes were uncertain or changing, attention increased
• Conclusion: attention is not controlled by just one mechanism (predictiveness AND uncertainty)

Question 12

Two kinds of attention?

Answer 12

a) Exploitative attention: focuses on known, reliable predictors (Mackintosh)
b) Exploratory attention: focuses on uncertain or surprising cues (Pearce-Hall)
- In two experiments that used the same stimuli (auditory), rats underwent appetitive conditioning (learning about rewards), showed that learning was influenced by both exploitative and exploratory attention
- This is a ratio of responding to the CS+ vs the CS-
- 0.5 = no diff in responding btw CS+ and CS-
- > 0.5. means responding to CS+ > CS-
- Results: behaviour showed evidence of both types of attention at once

Question 13

Hybrid model: a unified model

Answer 13

• Attention is dynamic and driven by two complementary forces
  a) Exploitative attention: focus on cues that have been good predictors in the past (Mackintosh)
  b) Exploratory attention: focus on cues whose outcomes are uncertain or surprising (Pearce & Hall)
• If a previous reliable cue becomes unreliable = exploratory attention kicks in
• Allows flexible updating of associations
• Outcome processing (Rescorla & Wagner)

Question 14

Summary of attention with learning

Answer 14

• Studies of learning in human and non-human animals suggest two types of attentional learning:
  a) Exploitative attention - Allows organisms to take advantage of predictivity in the environment when it is available
  b) Exploratory attention – Deployed when there is prediction error in order to try to make sense of uncertainty in the world
• Ongoing research is looking at…
• Where, and how, in the brain these systems are, and how they interact.
• Their role in mental health issues (e.g. Uncertainty and anxiety; Predictivity and addiction)
• Different types of uncertainty – Expected and unexpected uncertainty