Decision Making On the Reality of Cognitive Illusions

Decision Making On the Reality of Cognitive Illusions (Kahneman & Tversky, 1996) On Narrow Norms and Vague Heuristics: A Reply to Kahneman and Tversky (Gigerenzer, 1996) A reexamination of the evidence for the somatic marker hypothesis: What participants really know in the Iowa gambling task (Maia & Mc. Clelland, 2004) Expectations and outcomes: Decision-making in the primate brain (Mc. Coy & Platt, 2005) 1

Kahneman & Tversky (1996) VS. Gigerenzer (1996) 2

On the Reality of Cognitive Illusions (Kahneman & Tversky, 1996) • Judgmental heuristics – Useful BUT sometimes lead to errors or biases • Main goal of studying heuristics and biases : understand the cognitive processes that produce valid/invalid judgments. • Criticism : portrays human mind in an overly negative light 3

Base-Rate Neglect • Base rate that is known to subject, at least approximately, is ignored or significantly underweighted – Due to use of representativeness heuristics. – Representativeness: “Fit” or similiarity He must be a business man 4

Base-Rate Neglect • Diagnosing whether a patient has rare (25%) or common (75%) disease (Bluck & Bower, 1988) – Judgments of relative frequency of the two diseases were determined entirely by diagnosticity of symptom, with no regard for base-rate frequencies of the diseases – K & T : Our mind is not a frequency monitoring device! – G : Misinterpretation of the original study (“baserate info is not ignored, only underused”) 5

Base-Rate Neglect • Outcome Ranking paradigm • Give case data about person, ask subject to rank outcomes (e. g. , occupations) by different criteria (e. g. , representativeness, probability, base rate) – Rankings by representativeness and by probability were nearly identical – K & T : The role of representativeness in prediction is critical (underusing base-rate). – G : Use of random sampling makes this disappear. • → K & T: No! 6

Conjunction Errors • ∩ If A includes B then the probability of B cannot exceed the probability of A; A B implies P(A) ≥ P(B) – Conjunction Rule: P(A & B) ≤ P(A) – Because representativeness and availability are not constrained by this rule, violations are expected in situations where a conjunction is more representative or available than one of its components 7

Conjunction Errors • “Linda” Experiment • Imagine a young woman, named Linda, who resembles a feminist, but not a bank teller • Which is more probable? – (a) Linda is a bank teller – (b) Linda is a bank teller who is active in the feminist movement – K & T : Although (a) is more likely than (b), judgment based on representativeness results in conjunction error. – G : Content-blind! 8

Conjunction Errors • “Linda” Experiment – K & T : Although (a) is more likely than (b), judgment based on representativeness results in conjunction error. – G : Content-blind! – Sound reasoning starts w/ investigating content of problem • • What is 'probable' ? ? The meaning of AND in this context – According to K&T, content of problem is irrelevant and only terms probable and is important. 9

Conjunction Errors How detection of inclusion and appreciation of its significance affect conjunction errors • Estimating frequency of each category facilitates detection of inclusion – e. g. , estimate # of bank tellers & # of feminist bank tellers • → According to K & T : There are conditions under which the correct answer is made transparent, but the phenomenon is still very interesting! • → According to G : Why is this? ? What is the underlying process that makes the correct answer transparent? (critique about vague heuristics) 10

Overconfidence Mean confidence exceeds overall accuracy • Which city has the larger population? • • (a) Tokyo (b) New York How confident are you? (e. g. , 0% – 100%) Illusion of validity • Average confidence for single items VS. estimates of the percentage of items they answered correctly – Inside view (single-case approach) VS Outside view (frequentistic approach) • Does not contradict K&T's theoretical position: Diff perspectives yield diff estimates 11

Critiques of heuristics & biases Ex) Representativeness remain vague, undefined, and unspecified with respect to conditions that elicit them and to the cognitive processes that underlie them. <Narrow Norms> • Problems for “Linda” example: (1) prob. theory is imposed as a norm for single event (e. g. , whether Linda is a bank teller) (2) norm is applied in content-blind way <Vague Heuristics> (1) One-word-labels traded as explanation by redescription (2) 12

A reexamination of the evidence for the somatic marker hypothesis (Maia & Mc. Clelland, 2004) Somatic Marker Hypothesis (Bechara, Damasio, & Damasio, 2000) • Affective somatic states associated with prior decision outcomes are used to guide future decisions • Optimal decision making is not simply result of rational, cognitive calculation of gains & losses but based on good/bad emotional reactions to prior outcomes of choices (Hinson, Jameson, Whitney, 2002) 13

A reexamination of the evidence for the somatic marker hypothesis (Maia & Mc. Clelland, 2004) Iowa Gambling Task (IGT) • Task: select a card from 4 decks on each trial with a goal to have the best possible net outcome • Two $100 decks and two $50 decks • – $100 decks: longterm net loss (avg of $25/trial) – $50 decks: longterm net gain (avg of $25/trial) Provide evidence for somatic marker hypothesis – e. g. , skin-conductance response 14

A reexamination of the evidence for the somatic marker hypothesis (Maia & Mc. Clelland, 2004) Problems w/ previous studies using IGT • Prior studies (Bechera et al, 1997) asked subjects about explicit knowledge w/ questions such as “tell me all you know about what is going on in this game” – • Open-ended questions often fail to identify all of conscious knoweldge. (MAYBE THEY KNOW BUT THEIR KNOWLEDGE WASN'T ASSESSED PROPERLY) “Good Deck” VS “Bad Deck” – Good and bad decks should be determined on each trial based on mean net outcomes a subject has had w/ each deck up until that trial 15

A reexamination of the evidence for the somatic marker hypothesis (Maia & Mc. Clelland, 2004) • • Defining levels of conscious knowledge – Level 0. No conscious knowledge for preference – Level 1. Conscious knowledge specifying pref. for one of two best decks but no conscious knowledge about outcomes of decks – Level 2. Conscious knowledge specifying a pref. & about outcomes Question: Do individuals behave advantageously even when they are at Level 0 or 1? 16

17

A reexamination of the evidence for the somatic marker hypothesis (Maia & Mc. Clelland, 2004) • All of the verbal report measures demonstrate knowledge of the advantageous strategy for the majority of participants 18

A reexamination of the evidence for the somatic marker hypothesis (Maia & Mc. Clelland, 2004) • Participants were at least at Level 1 of knowledge when they behaved advantageously • Vast majority of cases, participants have both level 1 and level 2 knowledge. 19

A reexamination of the evidence for the somatic marker hypothesis (Maia & Mc. Clelland, 2004) • Seems that people know what they're doing in IGT. – Knowledge may be sufficient to explain behavior in IGT • Possibility that SCRs reflect emotional responses elicited by knowledge that is consciously accessible 20

Expectations and outcomes: decision-making in the primate brain (Mc. Coy & Plan, 2005) Expected Value Theory → choose option w/ highest expected value (probability x expected reward) • Humans/animals are sensitive to expected value of available options – → suggests that nervous systems somehow represent info about estimated costs and benefits of potential behav. to link sensation to action • Stages of Oculomotor decision-making Sensation → reward expectation → action → outcome evaluation 21

Expectations and outcomes: decision-making in the primate brain (Mc. Coy & Plan, 2005) Expected value and primate parietal cortex • The role of Lateral Intraparietal area (LIT) – When monkeys were permitted to freely choose between two eye movements (looking up VS down) • modulated by both the size of reward and the probability of reinforcement. Sensation → reward expectation → action → outcome evaluation 22

High Value Trials Low Value Trials 23

Expectations and outcomes: decision-making in the primate brain (Mc. Coy & Plan, 2005) Expected value and primate cortex • Dorsolateral prefrontal cortex, supplementary eye fields, substantia nigra pars reticulata, superior colluculus – Expected value systematically biases neuronal activity throughout the cortical and subcortical oculomotor afferents to the superior colliculus 24

Expectations and outcomes: decision-making in the primate brain (Mc. Coy & Plan, 2005) Learning from mistakes: dopamine and prediction error • Expected value of eye movements is rapidly updated in primate brain when reward contingencies change • Reward prediction error – – comparison of expected and actual reward dopamine neurons in substantia nigra & ventral tegmental area: elevated by delivery of unpredicted rewards, unchanged following predicted reward, depressed when expected rewards are withheld. → these neurons encode reward prediction error, thus determining the direction and rate of learning Sensation → reward expectation → action → outcome evaluation 25

Monkeys rapidly learn to choose the high value target 26

Expectations and outcomes: decision-making in the primate brain (Mc. Coy & Plan, 2005) Updating expectations: posterior cingulate cortex • • Neurons in posterior cingulate cortex (CGp) – carries info about the predicted and experienced reward value of eye movement – reward modulation of neuronal activity in CGp: When reward differs from expectation, learning occurs (useful for instructing “when” & “how rapidly” learning should occur) Role of cingulate cortex in linking motivational outcomes to action Sensation → reward expectation → action → outcome evaluation 27

unrewarded trials (when reward was expected) Monkeys change the strategy of their behavioral response in order to maximize their receipt of reward - Cingulate cortex and supplementary eye field plays an important role. 28