Lecture 5 Flashcards
1
Q
What is David Hume’s Empiricism?
A
- Causal relations cannot be perceived directly
- Sensory experience is the source of all our knowledge e.g touch does not reflect causality and there is no causality measurement e.g thermometer
- Causal knowledge is constructed using non-causal input
- Relevant input is the presence and absence of events
2
Q
What are the three factors?
A
- Temporal and spatial contiguity between cause and effect: mix of space and time
- Temporal priority of cause before effect
- Constant conjunction of cause and effect
3
Q
What are the Caveats of Empiricism?
A
- Correlation: just because things occur together does not mean they are causation e.g violence on TV and violent behaviour
- Probability of the event occurring with the cause subtracted by the absence of the cause is a good indicator of causal strength = ceiling and floor effects
- e.g when this equation equals 0, you would not conclude that there is no relation, instead experiment is flawed
4
Q
What is tracking the invariants of nature?
A
- Distal stimulus is the thing we want to look at in the real world, proximal stimulus is the 2d image in the brain
- Cannot be done directly, ends up being a combination and reconstruction of something similar to the distal stimulus
5
Q
What is the significance of Funes the Memorious?
A
- Funes had an accident and has infallible perception and memory - but Is also an affliction
- Struggles to track invariance as everything changes all the time - cannot comprehend general things
6
Q
What is the summary of causal judgement and learning?
A
- Causal Powers are distal stimuli
- Observable contingencies e.g stats patterns are proximal stimuli
- We build up representation of the distal stimuli via proximal
- Goal is to track distal (invariant) rather than the proximal stimulus
- Proximal stimulus changes according to circumstance e.g base rate and context
- Causal learning goes beyond associative learning: associations cannot capture causal asymmetries
7
Q
What is the difference between strength and structure of correlation?
A
- Structure: Are variables related?
- Given there is a relationship, how strong is it?
- Different causal structures can produce the same patterns of correlation
8
Q
What are Causal Bayesian Networks?
A
- Use statistical evidence to construct causal models of the world
- Causal dependencies imply various patterns of conditional dependencies
- Intervening on a system to change one variable renders it independent from the other factors = basis of systematic experimentation and hypothesis testing
9
Q
How do heuristics affect bayesian networks?
A
- Probability questions are substituted with something easier
- Which statement makes more causal sense as information flows from cause to effect
10
Q
What are inappropriate Application of Causal Models?
A
- Stereotyping/Prejudice
- Belief preservation
- Confirmation Bias
- Misperception of Randomness
- Not-hand Fallacy: in sport and gambling: if player scores three times, people believe he is on a streak and so others pass to him
11
Q
When does causality interact with time?
A
- Asked people questions between different historical events and asked how long time differed between them
- There was some correlation between some times
- Asked people how strong they thought there was a causal relationship between the two events: the higher the time, the less the causal strength rating but the lower the time estimate, the causal strength is high
- BECAUSE: attribute sub: temporal question - attribute is substituted by causal info
12
Q
What are perceptual distortions?
A
- Causality warps subjective time and space
- Time perception task: people perceived two lights flash at short time intervals
- Had to reproduce temporal interval by holding down a key = very close to accurate
- Other condition is the ppt presses a key and light shows up: when they reproduce = underestimate a lot
- Reproduced with primary school children and found children anticipate causal outcome earlier
13
Q
What is Newcomb’s Paradox?
A
- Demon can give you 1 mill in one box and 1000 guaranteed in another box, 1 mill disappears if you take both boxes, demon predicts and then you make ur decision
- Expected Utility analysis makes sense: probabilities to get 1 mill
- BUT causal analysis makes sense as shows independence and guaranteed 1000
- Contradicting analyses
- Can sever links to dictate independence so you can change mind after demon makes prediction
14
Q
Why does causal structure matter?
A
- Choices are interventions
- Probabilities derived from observations are often not useful
- Interventional probabilities are useful
- e.g out of 100 men who do chores, 82 are in good health whereas 32 of 100 men who do not help with chores are = doing chores is additional exercise vs men concerned about equality are also concerned about health and help with chores and eat healthier food = should your friend do chores?
- Observed prob/correlations are identical, same evidential relations, different causal = people choice clearly reflect awareness of causal consequences
15
Q
What was an example of why causal structure matters?
A
- Two types of car and both have marketing campaign but can only launch one
- Main competitor selling same cars, and promotes each car 95% of the time the same time as you: he has decided and cannot change his decision
- Regardless of what he does, his decision cannot be changed, so must choose an outcome to maximise YOUR sales
- If he hasn’t decided, you should go off comparisons because there is a direct link between campaigns - base rates come into play
