outline.txt
4/7/2006
Outline for rational analysis as optimal data selection talk

I. Background
  A. Selection task
  B. Falsification account
  C. Probabilistic account

II. Bayesian approach
  A. Conditional entropy and information gain
  B. Assumptions of the abstract selection task
    1. p(p | M_d) = p(p | M_i) = a
    2. p(q | not(p),M_d) = p(q | not(p),M_i) = b
    3. M_i => p(p,q | M_i) = p(p | M_i)*p(q | M_i)
    4. M_d => p(p,not(q) | M_d) = 0
  C. Model behavior  for various parameter settings
    1. p card: informative insofar as p(q) is low; it is largely
      independent of p(p)
    2. q card: informative insofar as p(p) and p(q) are both
      small
    3. not-q card: informative to the extent that p(p) is large; 
      it is independent of p(q)
    4. not-p card: not informative
    5. Rarity assumption: humans believe p(p),p(q) are low

III. Application of Bayesian model to standard abstract results
  A. Table 2 results
  B. Expectation of Eig(x) where x is card type over region R

IV. Some other abstract experimental results

V. Thematic selection task
  A. Examples of selection task
  B. Reaction to Comides et al.
  C. Two different rule types (enforcer vs. actor)

VI. Utility-based model
  A. Rule testing
  B. Formulating utilities
  C. Computing expectations over utilities for a given rule type

VII. Application of utility model to thematic selection task
  A. Comparison of expect information gain to observed results

VIII. Discussion of probabilistic account of rationality
  A. Relation to rational analysis
  B. Compare with falsification account
  C. Compare with accounts of biases and content effects 
    (evolutionary account)

IX. Discussion questions
  A. Validity of rarity assumption?
  B. Why use a probabilistic,information-theoretic framework?
  C. Is the selection task a representative problem of accounting for
  rational thought?
  D. Is the selection task exhaustive?
  E.