Lecture

1
Lecture 3 Human Cognition(Preece 3 Norman
3-5)
2
What Did You Learn Last Week?

• To impress your friends, suppose that you decide
  to sprinkle the following terms into your
  conversations
• "Conceptual model"
• "Gulf of evaluation"
• "Gulf of execution"
• "Direct manipulation"
• What are some example sentences that properly use
  these terms?

3
Why Do We Need to Know About Human Cognition?

• Interacting with technology involves cognitive
  processes
• Perceiving
• Remembering
• Learning
• Acting
• We need to understand the limits of those
  cognitive processes
• We need to identify and explain the nature and
  causes of problems users encounter
• We need theories, tools, and methods that help us
  do better design

4
Lecture Overview

• Part I Overview of Cognition
• Part II Models of Cognition

5
Part I Overview of CognitionAttention
PerceptionMemoryLearningMaking errors
6
Applying Cognitive Psychology to Human-Computer
Interaction
7
Core Cognitive Processes

• Attention
• Perception
• Memory
• Learning
• Motor behavior
• Reading, speaking and listening
• see Preece book
• Problem-solving
• see Preece book

8
Attention Make Salient Information Stand Out

• Everyone knows what attention is It is taking
  posession by the mind in clear and vivid form, of
  one out of what seem several simultaneously
  possible objects or trains of thoughtIt implies
  withdrawal from some things in order to deal
  effectively with others William James,
  ca. 1890
• Humans are limited with respect to what they can
  attend to at a given time attend to at a given
  time
• Design Implication Make salient information
  stand out using, e.g.,
• perceptual boundaries (windows)
• color
• reverse video
• Sound

9
How Can we Foster Accurate Perceptual Judgments?

• Remember Stage 5 of Norman model is Perceive
  state.
• What are some ways we can encode information
  (e.g., feedback) on the screen?
• Suppose we want to express quantitative
  relationships among objects of differing
  magnitudes.
• Which ways do you think will lead to accurate
  perceptual judgments?

10
Graphical Encodings

• Encoding 1 Angles

11
Graphical Encodings (cont.)

• Encoding 2 Areas

12
Graphical Encodings (cont.)

• Encoding 3 Lengths

13
Graphical Encodings (cont.)

• Encoding 4 Position on a common scale

14
Graphical Encodings (cont.)

• Encoding 5 Position on identical but unaligned
  scales

15
Graphical Encodings (cont.)

• Encoding 6 Angles with respect to horizontal

16
Graphical Encodings (cont.)

• Which type of encoding do you feel will yield the
  most accurate human judgments of differences?
• Why?

17
An Empirical Study of Graphical Encodings

• Cleveland McGill (1986) aimed to answer this
  question empirically
• Participants in the study were asked to make
  perceptual judgments using several of the
  encodings just presented
• angle, area, color hue, length, color brightness,
  position (on common scale), position (on
  identical but unaligned scales), color purity,
  slope, volume

18
An Empirical Study of Graphical Encodings (cont.)

• Sample Questions

19
An Empirical Study of Graphical Encodings (cont.)

• Results From best to worst, the accuracy of the
  encodings is as follows
• Position on a common scale
• Position along identical but unaligned scales
• Length
• Angle/Slope
• Area
• Volume
• Color properties

20
Color and Text Perception Limits

• Color Perception There are limits w.r.t.
• number of colors we can distinguish (7)
• the range of colors we judge to be a certain
  color (e.g., red)
• Text Perception There are limits w.r.t.
• the size of the font we can read
• The combinations of foreground/background colors
  that are legible

What is the time?
What is the time?
What is the time?
21
The Impact of Studies of Human Perception on
Design

• Differences among graphics elements should be
  recognizable
• Always try to encode differences with the highest
  ranked encoding on Cleveland McGill's scale
• Text should be legible
• Colors should be distinguishable

22
A Memory Test

• Try to remember the following numbers (there will
  be a quiz)
• 3, 12, 6, 20, 9, 4, 0, 1, 19, 8, 97, 13, 84

23
A Memory Test (cont.)

• Now quickly write down as many of the numbers as
  you can remember

24
Another memory test

• Try to remember as many of the following as you
  can (there will be a quiz)
• Split belt, fern crackers, banana laser, printer
  cream, cheddar tree, rain duckling, hot rock,
  fluffy crackers, cold music

25
A Memory Test (cont.)

• Now quickly write down as many of the items as
  you can remember

26
George Miller Knows How Many Items You Remembered!

• Miller (1956) We can hold 7 or 2 chunks in
  short term (working) memory
• A chunk is a unit of information, e.g., a number,
  a word
• Chunks can be combined and remembered as a unit
  (consider the second memory test you just took)
• What implications does this result have for user
  interface design?

27
Short Term vs. Long Term Memory

• Short-term memory (STM)
• Working or temporary memory of the present
• Can hold 7 ?2 items (Miller), or up to 10-12 with
  rehearsal
• May be effortlessly stored to and retrieved from,
  but is highly volatile
• Long-term memory (LTM)
• Memory of the past
• Enormous size (100 million items)
• Takes time and effort to commit items to LTM, and
  to retrieve from LTM
• Easier to store to and retrieve from if the item
  fits into what is already known

28
Conceptual vs. Procedural Memory

• Conceptual memory
• What Objects, attributes, facts
• Relations, e.g., cause-effect, nouns-verbs-objects
  
• Example Boise is capital of Idaho
• Procedural memory
• How Memorized steps linked to a goal
  (algorithm!)
• Results from practice
• Can become automatic and sometimes unconscious,
  yet difficult to change and error-prone
• Example Procedure for brushing teeth

29
Visual vs. Auditory (Textual) Memory

• Paivios (1971) dual-coding theory
• Pictures and words are stored in separate areas
  of memory
• Picture memory codes for an item can become
  connected to word memory codes for same item
  (dual coding)
• Pictures are more likely than words to be dually
  coded
• Implication We tend to remember pictures better
  than words
• Dozens of empirical studies corroborate this

30
A Penny For Your Thoughts
31
Memory in Head vs. World (Norman)

• Knowledge in Head
• Short term and long term memory stores
• Knowledge in World
• Great precision is not required for most
  decisions we just need to select from
  alternatives
• We can recognize far better than we can recall
• E.g., money, streets, cars
• Natural constraints are present
• E.g., assembly of object, rhyming words
• Cultural constraints are present
• E.g., face forward in elevator, show up late, but
  not too late

32
Human Memory Design Implications

• Recall is better than recognition ?
• When possible, put knowledge in the world, i.e.,
  in the interface
• GUI as opposed to a command-line interface
• Short term memory can store only 7 ? 2 items ?
• Dont make users remember items from screen to
  screen
• Automatically propagate essential information
  dont make user re-enter it
• Pictures are remembered better than words ?
• Where practical, provide pictorial and textual
  representations for items people will
  dually-code the representations and ultimately be
  able to remember the pictures better
• Procedural memory is error-prone ?
• Design should anticipate errors (see upcoming
  slides... )

33
What is Learning?

• Performance improvement
• Power Law of Practice Performance of task
  improves with time
• Affects perception, motor behavior, cognition
• Knowledge acquisition
• What is learned interacts with what is already
  known
• Transfer of training
• Metaphor/analogy
• Misconceptions Incongruities between current
  situation and what is already known

34
The Learning Curve

• The learning curve

Time to Perform Task
Number of Repetitions
Problem-Solving(Steps Uncertain)
Cognitive Skill(Steps Routinized)
NOVICE
CASUAL
EXPERT
35
Power Law of Practice

• Tn T1n-a
• where T1 is the time of the first trial and a is
  typically in range 0.2 to 0.6 (Plots as
  hyperbolic curve)
• Alternate version
• log Tn logT1 a log n
• (Plots as straight line good for linear
  regression)

36
Learning (cont.)

• Design Implications
• Need to provide variety of methods by which users
  can accomplish tasks
• Highly visible but relatively inefficient (for
  novices)
• Invisible but efficient (for experts)
• Example
• Novices are more likely to use menus to
  accomplish tasks, whereas experts migrate to
  keystroke shortcuts
• 80-20 rule Users will use 20 of a systems
  functionality 80 of the time.
• Know what the most frequently performed tasks
  are, and make sure that the full spectrum of
  users can access them
• The other 80 tasks dont need to be as
  accessible, as theyre most often performed only
  by experts

37
Human Limits of Motor Behavior

• There are limits with respect to how quickly
  humans can move
• One relevant limitation has to do with moving the
  mouse pointer to a target
• Fitts Law predicts this time
• T k log2 (D/S 0.5)
• where
• T time to move to target
• D distance between hand and target
• S size of target
• k 100 msec
• Lets test this out http//www.tele-actor.net/fit
  ts/
• Also take this quiz http//www.asktog.com/columns
  /022DesignedToGiveFitts.html

38
Errors (see Norman ch. 5)

• Humans routinely make errors
• Slips Errors resulting from automatic behavior
• Mistakes Errors resulting from conscious
  processing

Form Goal
Map goal to intention
Determine system is in desired state?
Map intentions to actions
Interpret system state?
Perform action
Perceive system state?
THE WORLD
39
Errors (cont.)

• Types of slips
• Capture error
• Youre doing one activity, but then a similar
  activity takes over
• E.g, sing one tune, but then you begin singing
  another
• Description error
• You perform a correct action on a wrong object
• E.g., pour orange juice on cereal
• Data-driven error
• You see data immediately at hand, instead of
  correct data
• E.g, dial a number in view, instead of correct
  number

40
Errors (cont.)

• Types of slips (cont.)
• Associative Activation error
• Internal association causes you to say or respond
  inappropriately to event
• Tee kettle rings you open front door
• Freudian slips
• Loss-of-Activation error
• You begin activity, but then forget what you were
  doing
• Walk to bedroom, but cant remember why
• Mode error
• You perform an action that normally satisfies
  goal, but you get unexpected results because you
  werent in right mode
• Try to select a word in word processor when
  Search dialog box is up

41
Errors (cont.)

• Design implications
• Prevent slips
• Make it difficult to perform inappropriate action
• Dont allow oil to go into gas tank (physical
  constraint)
• Allow disk to fit in disk drive in only one way
• Require confirmation of destructive actions
• Enable easy detection/correction of slips
• Provide good feedback
• Allow actions to be reversed well after the fact
• Recycle bin bin must be explicitly emptied

42
Part II Models of Cognition Mental
ModelsInformation ProcessingExternal Cognition
43
Mental Models

• Internal constructions of some aspect of the
  external world, e.g., computer systems (Craik,
  1943)
• People run mental models to make predictions
  about system behavior
• People develop core mental models, and apply
  them to explain how other things work
• Not always appropriate!
• People can have deep or shallow models
• e.g. how to drive a car or how it works

44
Mental Models (cont.)

• Example Mental model of thermostat
• You arrive home on a cold winters night to a
  cold house. How do you get the house to warm up
  as quickly as possible? Set the thermostat to be
  at its highest or to the desired temperature?
• You arrive home starving hungry. You look in the
  fridge and find all that is left is an uncooked
  pizza. You have an electric oven. Do you warm it
  up to 375 degrees first and then put it in (as
  specified by the instructions) or turn the oven
  up higher to try to warm it up quicker?

45
How did you fare?

• Your mental model
• How accurate?
• How similar?
• How shallow?
• Payne (1991) did a similar study and found that
  people frequently resort to analogies to explain
  how they work
• Peoples accounts varied greatly and were often
  ad hoc

46
Mental Models (cont.)

• Many people have erroneous mental models
  (Kempton, 1996)
• Thermostats are particularly problematic
• Peoples mental models tend to be based on
  general valve theory, where more is more
  principle is generalized to different settings
  (e.g. gas pedal, gas cooker, tap, radio volume)
• However, thermostats are based on model of on-off
  switch

47
Information Processing Model

• Analogy drawn between the human mind and a
  computer
• Just like a computer, human information
  processors have various hardware components
• Memory
• Working (storage capacity 3 chunks access time
  70 msec)
• Long term
• Perceptual processor (clock speed 100 msec)
• Cognitive processor (clock speed (70 msec)
• Motor processor (clock speed 70 msec)

48
Information Processing (cont.)

• The Model Human Processor (MHP) is seen as an
  approximation of human behavior
• Good enough for purposes of prediction
• In fact, in pioneering studies by Card, Moran,
  and Newell, the MHP was able to predict human
  performance with an error rate of only 10-20
• However, the MHP is highly limited
• Predictions only good under artificially closed
  conditions
• Doesnt take into account the distractions of a
  typical environment
• Predictions limited mainly to expert performance

49
External Cognition

• Concerned with explaining how we interact with
  external representations (e.g. maps, notes,
  diagrams)
• Key questions
• What are the cognitive benefits?
• What processes are involved?
• How do they extend our cognition?
• What computer-based representations can we
  develop to help even more?

50
External Cognition (cont.)

• Humans externalize to reduce memory load
• Diaries, reminders,calendars, notes, shopping
  lists, to-do lists
• Remind us that we need to do something (e.g. to
  buy something for mothers day)
• Remind us of what to do (e.g. buy a card)
• Remind us when to do something (e.g. send a card
  by a certain date)
• Post-its, piles, marked emails
• Where placed indicates priority
• Replaces cognitive task with perceptual one

51
External Cognition (cont.)

• Computational offloading
• When a tool is used in conjunction with an
  external representation to carry out a
  computation (e.g. pen and paper)
• E.g., try doing the two sums below (a) in your
  head, (b) on a piece of paper and (c) with a
  calculator.
• 234 x 456 ??
• CCXXXIIII x CCCCXXXXXVI ???
• Which is easiest and why? Both are identical sums
  
• Replaces cognitive task with perceptual one

52
External Cognition (cont.)

• Annotation
• Involves modifying existing representations
  through making marks
• e.g. crossing off, ticking, underlining
• Replaces cognitive task with perceptual one
• Cognitive tracing
• involves externally manipulating items into
  different orders or structures
• e.g., moving scrabble tiles on rack
• e.g., moving around cards in hand
• Replaces cognitive task with perceptual one

53
External Cognition (cont.)

• Design Implication
• Carefully-designed external representations at
  the interface can improve task performance by
  reducing memory load and facilitating
  computational offloading

e.g. Information visualizations potentially allow
people to make sense of large amounts of data
54
Summary Points

• Empirical data suggests limits on human
  performance and cognition
• We can use these data to help us design computer
  systems that are easy to use and maximize human
  performance
• In particular, these data provide
• Design principles and concepts
• Design guidelines
• They also serve as the basis for analytic tools
• Predictive models of human performance
• GOMS, Keystroke-Level Model (to be covered later)
• Walkthrough methods for predicting performance
• Cognitive walkthrough (to be covered later)