Heuristics and Biases

0.0(0)
Studied by 0 people
call kaiCall Kai
learnLearn
examPractice Test
spaced repetitionSpaced Repetition
heart puzzleMatch
flashcardsFlashcards
GameKnowt Play
Card Sorting

1/36

flashcard set

Earn XP

Description and Tags

How we make decisions, utility theory, heuristics, heuristics gone awry, review and evaluation

37 Terms

1
New cards

What does the area of judgement and decision making study?

How people select between options (maximising good + minimising bad) → tied to problem of induction (having finite/incomplete data + make a conclusion based on these → no guarantee that it's right)

2
New cards

What example illustrates how not all inferences are equally likely (in the problem of induction)?

  • Have a 2D projection of the world on your retina but the world is 3D → lose a dimension in the process of encoding what's out there in the 3D world 

  • Making inferences about what you see from incomplete info → visual illusions 


3
New cards

What is the best/most ideal inductive inference?

Rational inductive inference → based on likelihood estimates from probability theory

  • Still mostly right (but not all the time)


4
New cards

Do we make rational inductive inferences? Why/why not?

No

  • Probability based on probability theory is hard to calculate → don't have time to work out the maths + no brain area/module to do this for us

    • Evolution → brittle bio organism that exists in space-time → no time to gather the requisite evidence + amount of data needed to make good inductions via probability theory (prioritise survival)

    • E.g. poisonous mushroom → would eat it again to collect more data → wouldn't survive


5
New cards

What did Newell and Simon (1972) argue about problem solving?

Is systematic + uses heuristics (general rules) 

6
New cards

What are heuristics?

  • Herbert Simon → satisficing (good enough approach to a solution)

  • Foundations of human decision making (Tversky + Kahneman) → solver assesses the difference between the current state + the goal state → applies an operator to reduce that difference (often by setting sub-goals)

    • Means-end analysis)

  • Give decent inductions most of the time without gathering lots of data (limited resources)

  • BUT not necessarily optimal or guaranteed to be right

    • Often general purpose/broadly applied → trial + error (works in different contexts)

    • Can be based on prior experience → educated guess


7
New cards

What did Gigerenzer and Goldstein (1996) argue about heuristics?

Are adaptive tools (not just errors) + we use them because they are useful (not because we're incapable of using anything else) 

  • E.g. if one recognizes one option but not the other, the recognized one is often the correct choice (e.g., larger city population) -> can be more effective than complex analysis in real-world environments 

  • Produce a satisfactory decision (not perfect) 


8
New cards

What does the Hobbits and Orcs problem illustrate?

Humans struggle with counter-intuitive steps that are logically necessary 

9
New cards

What are the properties of heuristics?

  • Work well (at least most of the time)

  • Almost always feel compelling → is a bias to us, feels naturally correct

    • BUT so compelling we don’t use standard probability theory even when we have the necessary info (e.g. T + K, 1973)

      • PPTs rated p(lawyer) at 0.5 even though they were told the base rate was 70% → base rate neglect


10
New cards

What is the difference between normative and descriptive theories of bias?

How decisions should be made VS how decisions are made

11
New cards

Give an example of a normative theory of decision making?

Utility theory (max expected utility)

  • Subjective Expected Utility (SEU) → allows for subjective probabilities and values 

  • BUT humans rarely follow these rational calculations in real life 


12
New cards

Give an example of a descriptive theory of decision making?


Prospect theory (Kahneman and Tversky, 1979)
 

  • People evaluate outcomes as gains or losses relative to a reference point 

  • Loss aversion = losses feel more painful than equivalent gains feel good 

    • Risk-aversion for gains + risk-seeking for losses 

  • Framing effects = decisions change based on presentation 

    • E.g. people are more willing to buy a replacement ticket if they lost cash (separate mental accounts) than if they lost the ticket itself (same mental account), even though the financial loss is identical ($20) 


13
New cards

How are decisions made?

With the goal of maximising utility → most likely to get positive utility (+ avoid negative utility)

14
New cards

What is utility?

Internal measure of goodness, function of value

  • 0 utility = neutral 

  • Positive utility (gains) = good 

    • Everyone’s utility is different → e.g. cookie = good BUT whether its 100 or 1000 utility depends on the person (but as long as it's good it'll be above 0 for everyone)

  • Negative utility (losses) = bad 


15
New cards

What is value?

How much you gain in some objective term (e.g. money) 

16
New cards

What is utility space?

How good or bad they feel to you (feelings of gain/loss)

17
New cards

Does utility scale linearly with value?

No → not the case that every unit of value produces an equal amount of utility (depends where you're starting from) 

  • Utility curve asymptotes on both sides → approaches the max + as value increases/approaches extreme values, the change in utility becomes smaller

    • Equal changes in value = ever smaller changes in utility


<p>No → not the ca<span style="background-color: inherit; line-height: 20.7px; color: windowtext;">se that every unit of value produces an equal amount of utility (depends where you're starting from)</span><span style="line-height: 20.7px; color: windowtext;">&nbsp;</span></p><ul><li><p class="Paragraph SCXO74043166 BCX0" style="text-align: left;"><span style="background-color: inherit; line-height: 20.7px; color: windowtext;">Utility curve asymptotes on both sides → approaches the max + as value increases/approaches extreme values, the change in utility becomes smaller</span></p><ul><li><p class="Paragraph SCXO74043166 BCX0" style="text-align: left;"><span style="line-height: 20.7px; color: windowtext;">Equal changes in value = ever smaller changes in utility</span></p></li></ul></li></ul><p></p>
18
New cards

Utility-Value Graph

  • Gains + losses compound at decreasing rates: 3>2>1

    • BUT (2-1) >( 3-2)

  • Losing is worse than winning is good → |-1| > 1

    • E.g. gain in 1000 was worth some utility, but losing 1000 was worth losing much more utility


<ul><li><p>Gains + losses compound at decreasing rates: 3&gt;2&gt;1</p><ul><li><p>BUT (2-1) &gt;( 3-2)</p></li></ul></li><li><p>Losing is worse than winning is good → |-1| &gt; 1</p><ul><li><p><span style="background-color: inherit;">E.g. gain in 1000 was worth some utility, but losing 1000 was worth losing much more utility</span></p></li></ul></li></ul><p></p>
19
New cards

How do we decide what is likely to get us positive utility?

Estimate probabilities → max expected utility

20
New cards

What is expected utility?

  • Amount of value we can expect from an event

    • How much over the long run we would expect to gain by making the same decision over and over again

  • Product of the utility of an outcome + probability of that outcome

  • Requires:

    • Value in utility

    • (Estimated) probability


<ul><li><p>Amount of value we can expect from an event</p><ul><li><p>How much over the long run we would expect to gain by making the same decision over and over again </p></li></ul></li><li><p>Product of the utility of an outcome + probability of that outcome</p></li><li><p>Requires:</p><ul><li><p>Value in utility</p></li><li><p>(Estimated) probability</p></li></ul></li></ul><p></p>
21
New cards

What is the formula to calculate the epected utility of a choice?

knowt flashcard image
22
New cards

Why do we make bad choices?

Bad at estimating probabilities → biased in JDM

23
New cards

What are our heuristics of judgement and decision making?

  • Representative → judge something to be likely to the extent it’s rep of things you’re familiar with

  • Availability → judge something to be likely to the extent that it’s easy to think of examples of that thing

    • Like rep but you have to think about specific examples 

  • Anchoring + adjustment → when estimating magnitudes, start with initial estimate + adjust it as new info arrives


Our estimates usually aren’t far off based on these techniques

24
New cards

According to Kahnemann and Tversky (1983), how can representativeness become biased?

Conjunction fallacy = judge the likelihood of joint events to be more likely than either one in isolation

<p>Conjunction fallacy = judge the likelihood of joint events to be more likely than either one in isolation </p>
25
New cards

According to Kahnemann and Tversky (1983), how can availability become biased?

  • Conjunction fallacy

  • Over/under-estimate based on news reports + what we’re aware of


<ul><li><p>Conjunction fallacy</p></li><li><p>Over/under-estimate based on news reports + what we’re aware of</p></li></ul><p></p>
26
New cards

According to Kahnemann and Tversky (1983), how can anchoring and adjustment become biased?

Biased by initial anchors, even if those anchors are generated by unreliable/biased sources (e.g. in biddings)

  • Low anchor → adjust lower

  • High anchor → adjust higher


<p>Biased by initial anchors, even if those anchors are generated by unreliable/biased sources (e.g. in biddings)</p><ul><li><p>Low anchor → adjust lower</p></li><li><p>High anchor → adjust higher</p></li></ul><p></p>
27
New cards

How did Tversky and Kahneman (1973) investigate base rate neglect (again)?

  • 114 uni students given personality sketch + rate whether Tom was likely to be in CS or humanities

  • Overwhelmingly chose CS even though:

    • Judged humanities grad students to be 3x more than CS grad students

    • Same students said they thought projective tests were unreliable

    • People change over time (may have been into computer when he was younger but not anymore)

  • Ignored prior odds even though they believed the info given was invalid


<ul><li><p>114 uni students given personality sketch + rate whether Tom was likely to be in CS or humanities</p></li><li><p>Overwhelmingly chose CS even though:</p><ul><li><p>Judged humanities grad students to be 3x more than CS grad students</p></li><li><p>Same students said they thought projective tests were unreliable</p></li><li><p>People change over time (may have been into computer when he was younger but not anymore)</p></li></ul></li><li><p>Ignored prior odds even though they believed the info given was invalid</p></li></ul><p></p>
28
New cards

How can problem solving be improved?

  • Know your stuff → domain knowledge helps, though too much can cause rigidity

  • Follow a plan → set + revise sub-goals (Means-End Analysis)

  • Look for links → use analogies based on procedural, structural, or superficial similarities 

  • Change your POV → alter the mental rep of the problem (e.g., from numbers to images) 


29
New cards

What are concepts?

Mental reps of classes of objects

30
New cards

What are categories?

The actual class of objects in the real world

31
New cards

What is the classical approach to categorisation? What are the weaknesses of this approach and the consequence of these?

  • Based on defining attributes → every concept has a list of defining features + to be a member of a category, an object must possess all these features (SO all members of a category are considered equal in status) 

  • Originates with Aristotle 

  • BUT McCloskey + Glucksberg (1978) → humans are inconsistent in applying these rules (e.g., disagreeing on whether a penguin is a bird or if advertising is a science) 

  • Precise + interpretable (people understand what it says and can grasp the details) + testable (it is worded in such a way that it is possible to find evidence against it) BUT fails to account for evidence regarding typicality + inconsistent categorisation 

  • BUT struggles with abstract or "fuzzy" concepts (like "having the X Factor") where clear defining features are impossible to list + members aren't equally representative → typicality 


32
New cards

What is typicality?

Some members of a category are more typical than others 

  • Influences how we think about things (in terms of speed of learning + decision-making) 

    • Rips, Shoben and Smith (1973) →  we are able to decide whether something is typical of a category more quickly than whether it is atypical (e.g. robin vs penguin as birds)  

    • Mervis and Pani (1980) → we are able to learn new concepts quicker if given examples of typical members rather than atypical members 


33
New cards

What is the prototype approach to categorisation (Rosch and Mervis, 1975)? What are the weaknesses of this approach?

  • Based on typicality 

  • Categories are represented by a prototype = a central description or a weighted list of characteristic features (known as family resemblances) 

    • NOT strict defining attribute 

  • Items don't need to match every feature to belong → just need to be sufficiently similar to the prototype (e.g. a penguin is a bird BUT not a typical bird) 

  • BUT category membership can change based on context (e.g., viewing a penguin at a zoo vs. a bird sanctuary) → not explained by prototype model 

  • BUT difficult to define prototypes for abstract concepts (e.g., "belief" or "instinct") 


34
New cards

What is the exemplar approach to categorisation? What are the weaknesses of this approach?

  • Based on specific past instances → concepts are formed from specific instances stored in memory from past experiences 

    • When judging if something belongs to a category, humans compare it to remembered examples (exemplars) rather than an abstract rule 

  • Challenges idea of abstract summaries (prototypes) 

  • Explains why people are quicker to classify a robin as a bird than an animal (it resembles more stored bird exemplars) 

  • Good for complex concepts BUT may be less efficient for simple concepts (Smith + Minda, 2000) → likely use a compromise between prototype + exemplar approaches depending on the situation (Van Paemel + Storms, 2008) 


35
New cards

How are we not good problem solvers?

  • Negative set = tend to use one type of solution + can't see around it

  • Functional fixedness = unable to think about novel ways of using things


36
New cards

What was Gestalt psychology’s early approach to problem solving?


(Wolfgang) Köhler's Apes → experiments with an ape named Sultan
 

  • Demonstrated insight learning → sudden "ah-ha" moment where the solution appears instantly, rather than through gradual trial and error 

  • E.g. Sultan joined two short sticks to reach bananas outside of his cage (higher level problem-solving) (previously given one long stick + used this → demonstrated basic-level problem solving) 


37
New cards

How did early approaches to problem solving involve insight? What is a weakness and consequence of this?

A moment in which a novel approach to a problem, or a new way of looking upon things, is suddenly realised 

  • Sometimes good problem solving comes from an insight that uses an analogy → recognising that a current problem is similar to a past one (may then have similar solutions) 

    • E.g. Archimedes' Eureka → solving a problem (measuring the crown's volume) by linking it to a known phenomenon (water displacement) 

    • BUT Gick + Holyoak (1980) → people often miss analogies unless surface similarities are obvious or hints are given 

  • BUT good problem solving is much more systematic than this → heuristics