PowerPointPrsentation

1
12 pt Is it possible to read this? A lot of text
can be put into such a small font.
14 pt Hopefully a little better
Calibration slide
28 pt The smallest font used
44 pt Titles
2
?
Cognitive Psychology Spring 2005 -Discussion
Section-
3
Full plate - again

• Decision making
• Reasoning
• Problem solving
• Exams
• Evaluations

4
Cognitive functions

• Perception

• Attention

Emotion Motivation Action

• Memory

• Imagery

• Language

• Reasoning, problem-solving

problem-solving
Reasoning,

• Decision-making

• Decision-making

5
Decision making

• Old economical theories
• New psychological theories
• ? Bounded rationality

6
Decision making

• Economy Expected utility theory
• Normative theory Yields optimal payoff
• Based on expected value
• EV ? ( pi x vi ) Sum of prob times value
• Applicable to many situations (Renting an
  apartment, chosing a major, deciding where to go
  for holidays, etc.)

7
Example Expected utility theory
8
Problems

• This obviously works only for small, well defined
  problems.
• Most real life problems have many options
• Are hard to quantify both in value and
  probability
• Need quick decisions.
• Even if people were able to collect all that
  information, they were probably unable to combine
  it appropriately to act optimally.

9
Conclusions

• People do generally not use expected utility
  theory to make their decisions.
• The model helps economists to make predictions.
• The predictions are usually not very accurate
• The model has to be enriched with psychological
  mechanisms.

10
Alternative view

• People use heuristics in decision making (not
  algorithms!)
• They suffer from psychological biases

11
Algorithms and Heuristics

• Algorithms yield optimal solutions (if they
  exist). They provide certainty. But are often
  practically impossible to apply. Example Solving
  chess.
• Heuristics are rules of thumb. They yield a
  solution that is good enough. But no certainty
  that it will work. Quick and dirty. Example
  Doing what everyone else does.

12
Example Constrast of Algorithms and Heuristics

• Outfielder in Baseball. Objective Catch ball.
• Can either estimate ball speed, distance, wind
  speed, throwing angle, own speed, ball
  aerodynamics, etc. (in a short time, then start
  running).
• Or One can keep the angle of the ball in the
  viewing field constant by running. ? Will catch
  ball AND will already be running.

13
Availability heuristic

• People are not good at estimating probabilities
• They use shortcuts
• Basically Instances that are overrepresented in
  the mental representation and hence easier to
  call to mind are deemed to be more probable.
• This may reveal something about the structure of
  our mental representations!
• A priori, it is unclear why we should have a
  harder time thinking of words ending in p than
  words starting with p. But people think the
  former case is much more frequent than the
  latter.
• This heuristic can be both very efficient or
  misleading. Depending on the correlation with the
  criterion.

14
Representativeness heuristic

• If something is easier to represent in a certain
  way, people will think that it is more likely
  that the things are physically like the mental
  representation.
• Example Repetition avoidance in generating
  random numbers.
• Example Illusory effectiveness of rain dances,
  many medications and therapies.

15
Knowledge heuristic (availability)

• The less is more effect
• More citizens in Baltimore or Seattle?
• More citizens in Minneapolis or Sacramento?
• More citizens in Berlin or Leipzig?
• More citizens in Duisburg or Munich?
• If the probability of you knowing about it is
  correlated with the criterion asked, you will be
  better off, if you just chose the one you know.

16
Biases

• Cognitive illusions
• Like perceptual illusions, it is hoped that we
  can understand decision making by investigating
  its illusions. Like in Perception.
• Unlike Heuristics, they generally diminish the
  quality of the decision.

17
Framing bias

• The way a certain problem is phrased influences
  peoples behavior in response to it.
• Examples 90 Chance of survival vs. 10 Chance
  of Death influences surgeons decision to operate.
• Example Losing vs. Gaining not the same.

18
Anchoring bias

• The initial starting point of our considerations
  will have a big effect on the final
  estimate/decision.
• Example 1x2x3x4x5x6x7x8x9
• vs 9x8x7x6x5x4x3x2x1

19
Sunk Cost Bias

• Once an investment has been made, the probability
  to continue the undertaking is higher,
  proportional to the magnitude of the investment.
• Rationally, it should NOT influence our decision
  to continue.
• In the real world, investment often increases
  probability of success (its an estimate of that
  probability).
• Also, this is structurally similar to cognitive
  dissonance.
• Example Continue to fight wars that one cant
  win.

20
Illusory correlation bias

• People see structure wherever possible.
• Even in random noise (some people see faces)
• Probably an offshot of the representativeness
  heuristic What fits the mental model is seem to
  be more likely. These features then get more
  attention and are overepresented in the mental
  representation.

21
Hindsight bias

• Hindsight is always 20/20
• Prime example 9/11. Wasnt it obvious that
  something like that would happen soon, given that
  they already tried? Why did two presidents do
  nothing about it?
• People think their original expectations of what
  will happen were closer to what actually
  happened.
• Evidence for constructiveness of mental rep.

22
Confirmation bias

• We encountered this in the reasoning chapter.
• Most people look for information that confirms
  original expectations instead of information that
  is more likely to falsify it.

23
Overconfidence bias

• People generally have higher confidence in their
  judgement than is warranted by the data or by
  expectations.
• The problem with this bias is that it immunizes
  people to seek a fix for their other biases. They
  dont see the need.

24
A bad picture

• People suffer from all kinds of biases that
  prevent them from reaching the optimal decision.
  
• The sunk cost effect keeps them to stick with
  their decision, the overconfidence effect keeps
  them from realizing that they are in need of a
  better solution.
• Hindsight bias keeps them from learning in the
  long run. After all, our decisions werent so
  bad, right?
• Are they just bad and cant do any better?

25
Why?

• Action control
• Heuristics and biases enable us to reach a
  decision quickly, act on it and confidently STICK
  with it.
• In the long run, this might be more rational than
  wavering and behaving erratically the first time
  new information comes around.
• People need to act on information. These
  shortcuts often enough help them, particularly if
  they have to act on them (vs. just pondering
  them).

26
Example Mate choice

• Assumptions

• Can have everyone one wants

• Only one shot. Decisions are final

• Views candidates sequentially

• Is able to assign a value to everyone

• Doesnt compromise standards.

• From random distribution

27
Normally distributed pool
28
Equally distributed pool
29
Lessons Specific

• Make the cutoff at 20 of your expected pool

• Actual size and distribution of pool doesnt
  matter much.

• Subjective expected pool-size does!

• Breakup and divorce as means of optimization of
  utility in case of premature selection

• Could explain increased divorce rate by that
  (more people around, longer integration time)

30
Lessons General

• Decision making and modeling of decisions do have
  very important real-life implications and
  applications!

31
Problem Solving
32
Problem solving

• The three door problem

• Behind one of the doors is 1 Mio , nothing is
  behind the other 2.

• You choose a door.

• I open another door and show you that there is
  nothing

• Now you have the option to change. Do you?

33
Problems

• Well defined problems Clear start state, clear
  goal state, clear operations.
• Example Chess, Tower of Hanoi, Games in general.
• Ill-defined, complex problems Unclear start
  state, unclear goal state, unclear operations.
• Example Managing a company, waging a war,
  leading a good life.

34
Problem space

• Every well defined problem can be represented by
  a problem space.
• The space contains states and relations (steps)
  between them.
• Even well-defined spaces get huge very quickly
  (e.g. chess).
• Computers (AI) solve problems by exploring the
  problem space.

35
Strategies of problem solving

• Generate and Test Come up with a lot of
  solutions and then test them sequentially until
  success.
• Means-End Analysis Comparing the starting state
  with the goal state and generating intermediate
  steps that reduce the discrepancy.
• Working backward Determine the last step before
  the goal step and then generate steps from there
  on until one reaches the start position. Useful
  if there is only possible solution.
• Backtracking Making assumptions, deliberating
  as-if and undoing them if it turns out that it is
  a dead-end.
• Using analogies Realize the same underlying
  structure between a familiar situation and a
  problem situation and act as if the familiar
  situation would apply. Famous example Tumor
  problem, the gamma knife.

36
Detriments to good solutions

• Mental set
• Constrained problem space

37
Detriments to good solutions

• Functional fixedness
• Inadequate mental representations

Candle Box of nails How to tack the candle to the
wall?
38
Detriments to good solutions

• Lack of knowledge and expertise

• Experts pay attention to relevant information
• They have deep (vs. shallow) categories
• They can bring an enormous memory to bear,
  basically pattern matching (e.g. in chess).
• Experts are more likely to check for errors.

39
Creativity

• Particularly useful to solve ill-defined
  problems.
• Practice (10 year rule)
• Productivity (the more, the merrier)
• High IQ
• Good mental representations
• Willingness to take risks
• Personality factors
• Nothing mystical Artificial creativity! Aaron.

http//www.kurzweilcyberart.com/KCATaaron/STAFsamp
le
40
Reasoning
41
Hypothesis testing
2 4 6...
42
Wason
If vowel on side one, even number on other
p p q q
...
43
Base rate neglect
Prevalence of breast cancer in population is 1.
Mammograph detects present breast cancer 99 of
the time. If no breast cancer is present, the
false-positive rate is 10.
After a test, the mammograph of a woman is
positive. What is the probability that she has
breast cancer?
44
Base rate neglect
Consider 1000 people.
10 of them do have breast cancer.
All 10 are detected.
990 dont have breast cancer.
10 of them are detected as false-positives 99.
10 out of 10 99 109 roughly 9 !
Test parameters great, but baserate low!
45
The II. MIDTERM
46
Sections
47
Gains
48
Individual Changes
49
Evaluations
50
Conundrum
May 27th, 2005
8