Cognition Lecture 6

Cognition Lecture 6

1. Analogical Problem Solving

   • Previously experienced problem - base (LTM)
   • Current problem - target (WM)
   • Successful transfer -
     • Recognition
     • Abstraction
     • Mapping
   • Recognition and abstraction
     • Kintsch and Van Dijk (1978) - stored memory structures consist of "gist" - abstraction of essential content
       • Deletion process
       • Generalisation process
       • Construction process
   • Mapping
     • Solution of the base problem must be translated into the solution of the target
     • Gick and Holyoak - pp may generate and develop partial solutions to target problem
   • Extensions to classroom learning
     • Chen and Klahr (1999) - kids trained in designing and evaluating experiments did better on test
     • Chen (2002) - superficial, structural and procedural similarity between base and target are important
     • But
       • Similarities in everyday problems are less obvious
       • People choose structural rather than superficial similarities
       • Individual differences under-research, sleep?

2. Expertise

   • Novices and experts
     • Gradually accumulate experience
     • Previously acquired knowledge helps understand and solve problems
     • Differences between experts in different areas
   • Chess
     • Deep Blue - relies on speed of processing and online memory capacity - 90 billion moves per turn
     • Charness (1981), Saariluoma (1990) - human chess playing strategy mix of laziness and efficiency
       • 30 moves in total + ~4 first moves
     • Chase and Simon (1973) - masters recalled the most when board was organised to be mid game, but the least when organised randomly
     • Gobet and Clarkson (2004) - templates allow masters to recognise good moves rapidly - but info unclear
     • Holding and Reynolds (1982) - masters have an independent evaluative knowledge
     • Hatano and Inagaki (1986)
       • Routine expertise
       • Adaptive expertise
   • Physics
     • Experts build better representations of the problems than novices
       • Chi, Feltovich and Glasser (1981) - novices sorted problems according to surface similarities, experts sorted according to deep structure.
     • Larkin et al. (1980) - experts work forward through problem, novices work backwards.
   • Computer programming
     • Adelson (1981) - experts grouped together lines from same program; novices grouped together similar looking lines
     • Adelson (1984) - experts more likely to form representations based on higher-level principlas of program
     • McKeithen, Reitman, Reuter and Hirtle (1981) - recall increased when lines of code organised normally, but no difference when scrambled.
   • Medical expertise
     • Medical experts rely on implicit reasoning rather than explicit reasoning
     • Eye-tracking data - experts rapidly fixate on relevant information
     • Kundel et al. (2007) - less than 1 second by experts to fixate a cancer region on mammogram
     • Melo et al. (2012) - medical experts use a predominantly visual strategy to compare x-ray with stored patterns
     • Kulatunga-Moruzi et al. (2004) - expert dermatologists better at purely visual diagnosis
     • Mamede et al. (2010) - analytic thinking beneficial for experts with complex cases
     • Real world - both fast and analyticla processes
   • Brain plasticity
     • Development of expertise related to brain modifications?
     • Woolett and Maguire (2009) - experienced cabbies have greater volume of grey matter in posterior hippocampus
     • Zatorre (2013) - changes to volume and auditory cortex with musical training

3. Expert Systems

   • Many attempts to embody human expertise in computer programs
     • MACSYMA - maths
     • DENDRAL - structure of organic molecules
     • MYCIN - diagnoses and treatments of bacterial diseases
       • ~500 production rules
       • Outperformed many medical experts BUT ethical and legal issues
   • Problems
     • Can't learn for themselves
     • Difficult to get knowledge from human experts
     • Expensive to develop
     • Rely on domain-specific knowledge
     • Need to move away from rule-based approach
     • Incorporated into Bayesian Network models

4. Learning to be an Expert

   • Practice Makes Perfect
     • Hayes (1985) - experts have at least 10 years of practice
     • Ericsson et al. (1993) - best violinist students estimated ~7000 hours of lifetime practice
     • Distributed practice is best
       • Zachmeister and Shaughnessy (1980) - memory for words better if repeated, esp. when spaced out
       • Dempster (1987) - longer gap = better recall
     • 'Learning-curve' - follows a power-law function
       • Kolers (1979) - initial period of rapid learning very soon followed by diminishing returns with further practice
   • Chunking
     • Rosenbloom and Newell (1986) - experts use rule 1 (a to f) to decrease time taken to solve problem
   • Knowledge compilation
     • Fitts and Posner - 3 stages
       • Cognitive stage - declarative knowledge
       • Associative stage - recognition and elimination of errors, procedural knowledge
       • Autonomous stage - procedures automatic
     • Anderson (1982, 1983, 1987, 1993) - ACT*
       • Working memory - activated knowledge
       • Declarative memory - LTM of world
       • Procedural memory - LTM of production rules
       • Knowledge becomes procedural through proceduralisation