Integrating Reasoning and Skill Flashcards
Can reasoning be learnt?
Can’t decrease reasoning errors with training, due to generalisability problem
Thorndike (1923): Little success in transfer across dissimilar domains
Teaching thinking skills in schools - mixed success
But if reasoning involves heuristics, then there’s some reason to think we can learn to use good heuristics.
What does Nisbett, Fong, Lehman and Cheng’s (1987) two-part study show about training reasoning?
- Gave people explanation of law of large numbers (as sample gets larger, less likely to be deviant from norm in audition) and training in several domains
- Read about director of ballet company who chose some ballet dancers, but they turned out to not be much better than others. Asked to reflect on how law of large numbers might explain this
FOUND: Training improved correct statistical reasoning from 42%-56% of problems
Replicated experiment, and tested immediately after training and two weeks later
FOUND: Some forgetting, but still do better than control
Teaching people general reasoning principles seems to have an impact.
Nisbett et al: Does teaching statistics affect inferential skills?
Does learning stats affect logic?
Postgrad vs 1st/3rd year students
Tested different kinds of inferential skills
FOUND: All students were similar in 1st year, but not 3rd year.
Psych did best (lots of stats), Chem didn’t do too well.
There are gains on logical reasoning, without direct training.
What does Evans’ (2003) Dual process Theory suggest about reasoning?
Reasoning uses two processes:
- Ancient system: relies on associative learning (implicit) e.g. intuition
- Biased by prior beliefs and knowledge
- Heuristic –> cog illusions
- Built by experience - System uses WM resources: Slow, sequential
- Evolutionarily recent
- Related to IQ
- Requires effortful intervention (instructions important)
- Built by training
System 1 is default, but reasoning tasks require intervention of 2 to consistently succeed.
How does Dual process approach compare to other reasoning approaches (Mental Models, Mental Logics, Bayesian)?
Evans (2003) in sum:
System 1 - Autonomous subsystems using domain specific knowledge
System 2 - Permits abstract reasoning, but constrained by WM capacity
Other approaches
- Mental models and mental logic: Emphasise working things through - More System 2
- Errors due to too little System 2, not too much
- Initial models or application of mental logic rules described as effortless - imply System 1
Bayesian approach: Emphasise System 1
How does the Dual process theory account for believability effects?
System 1: Believability appears to have an effect, because experience shapes reasoning
What evidence is there from believability bias to support the Dual process theory?
- Believability affected by instructions to focus on deduction (System 2) –> less effect of believability
- People shift from logic to belief-based reasoning when under time pressure (System 2)
- Logic performance declines with age
(System 1 faster –> Use belief-based reasoning, while System 2 uses cog resources) - Mental Models: Initial model produced is shaped by bias/knowledge
1st system produced by System 1, further models created by System 2 to reach correct conclusion
Klauer (2000): People produce just one model, and then apply other processes
What is other evidence that support Dual process approach?
Goel and Dolan (2003): Correct belief-biased responses associated with activation in right inferior pFC, incorrect with vmpFC. Suggests that there’s differential neural activation with different belief mindsets, and that belief biases uses different parts of brain
Context effects in Wason: experience as evidence for System 1
Are Skill and Reasoning two separate systems?
Many systems in cognition appear to use two systems
Stanovich (1999): 12 sets of researchers see reasoning as dual-system
System 1): Associative, holistic, automatic
System 2): Rule based, analytic, controlled
System 1 keeps providing answers that System 2 may accept/reject - but usually cooperate (not compete)
- Patterns of error depend on the system used
Many theories of thinking involve multi-system idea
e.g. Decision making (heuristics vs calculation)
Problem solving (hyp testing vs pushing towards goal)
Skill acquisition: Development of System 2, then 1 (1 hard to change and improve from)
Reasoning: Heuristics vs calculation
What is the relationship between IQ and reasoning? Refer to Stanovich and West’s (1998) study
Extent of IQ relation to reasoning depends on which system is engaged to answer questions
- System 1 = No correlation (as no learning occurs?)
Used SAT score as indicator of intelligence
Concreteness of reasoning syllogism corr with IQ:
Abstract (Sys 2) - Associated with SAT scores
Concrete (Sys 1) - no corr
The more you expect a problem to involve System 2, the larger the effect of SAT
How does Stanovich and West (2008)
Examined studies where reasoning tasks do/don’t corr with IQ
FOUND:
Application of heuristics - no corr with IQ
Logical reasoning does corr with IQ
System 2 needs both capacity and metacog to override System 1.
Usefulness of brief info
System 1 > System 2 when using brief info
Better reasoning/thinking in the first few seconds (e.g. chess) due to practice
We make assessments of people based on little info
- Probabilistic account: We are sensitive to distributions of info (unaware)
Make quick analysis
What does the study by Gottman and Levinson (2000) about couples and divorce demonstrate about our analysis of the environment?
Videotaped 85 couples doing 3 15min conversations about events of day, conflict, agreed pleasant topic.
Could predict:
- Divorce within 7 years: negative affect (defensiveness, contempt) but not positive affect
- Divorce at 14 years predicted by positive affect
Measured marital satisfaction: Divorce predicted with 93% accuracy
There is small amounts of information in the environment from which we can make automatic big conclusions.
How does Kahneman explain reasoning?
Mental life as interaction of two agents System 1 (fast, intuitive): Operates auto and quickly with little-no effort and no sense of voluntary control (he came from automaticity/attn background) - E.g stereotypes System 2 (slow, deliberative): Allocates attn to effortful mental activities that demand it. Operates with subjective experience of agency, choice, and concentration
Interaction:
System 1 always gives answers, even when undesirable
- Heuristics/biases part of System 1, but not necessarily
System 2 can always intercept these answers, if it recognises them as wrong
- Doesn’t imply System 2 is always right
What are problems with Kahneman’s approach to reasoning?
Have we solved anything by saying there are 2 systems? (More df = easier to fit data)
Can’t make precise predictions: not sure when which system will be engaged (Speed? Load?)
How exactly do the systems interact?
