Yvonne Wrn Tema Kommunikation

1
Yvonne WærnTema Kommunikation

• Information Processing -
• In humans and machines
• April 24, 2001

2
Information Processing Psychology - IPP

• Revolution and opposition against behaviorism
• Behaviorism characterised psychology in general
  from about 1900 to about 1970.

3
What was behaviorism?

• Theory to describe (explain and predict)
  behaviour in observable terms only Stimulus -
  Response
• Pavlov conditioned reflex (dogs salivating)
• Skinner instrumental conditioning (pigeons
  playing table-tennis)
• Watson conditioned behaviour (Child afraid of
  rabbits)
• Nowadays Behavioural therapy (diverse phobias)
• Some computer supported learning

4
Information Processing Psychology -

• One of several approaches to model cognitive and
  mental processes - in opposition to behaviorism
• (but still some behavioristic traits)
• Other approaches with similar aim
• Gestalt Psychology, (contemporary with beh.)
• Piaget (contemporary with behaviorism)
• Bruner (inspired by Piaget)

5
Information Processing Psychology - ingredients

• Model from the computer -
• In contrast to previous cognitive models that
  were often statistical
• A modelling language - production rules
• In contrast to verbal descriptions
• A qualitative method to derive information
  processes
• In contrast to quantitative methods

6
IP- model from the computer (skipping psychology)

• Model from the computer - (1956!)
• Information processing
• Transformation of knowledge states from start
  to goal operations on symbols
• Success AI in form of The Logical Theorist
  derived all theorems (and some more) in Russel
  Whiteheads volume on Logics.

7
IP- model from the computer

• Was this human information processing?
• No
• people have bounded rationality
• Use heuristics
• Use smart ways of representing problems
• Are restricted by their information processing
  apparatus

8
IP- modelling language

• A modelling language - production rules
• In contrast to verbal descriptions
• If-then rules. The current state is matchted
  towards the system of rules. The first rule that
  matches the current state is fired.
• Then a new state results, that is matched
• What does this remind us of?

9
Information Processing methods

• A qualitative method to derive information
  processes
• The think-aloud protocol was used to elicit data
  on sequential problem solving.
• Hypotheses people expressed (parts of) that what
  existed in their working memory - i.e. part of
  the current knowledge state.

10
IPP - prerequisites-psychology reintroduced

• Since people are not computers, we have to use
  reverse engineering to understand the mechanisms
  by which they proceed
• Define problem
• Identify process
• Derive specific strategy from process
• Derive general cognitive architecture from
  several studies

11
Define problem

• A problem exists when you have a goal and an
  initial state that does not correspond to the
  goal and you do not know how to get from the
  initial state to the goal

12
Define problem space
A problem space consists of the hypothetical
states that a problem solver goes through in its
processing/transformation of the initial state to
the goal state. Ex. Problem space intitial
state operations required to reach goal state
13
Example of problem- Tower of Hanoi
You have three disks on a peg (A) as in the
figure. These should be moved to the right peg
(C). You are only allowed to move one disk at a
time. You can only place a smaller disk on top of
a bigger one.
A B
C
14
Think aloud protocol-example Tower of Hanoi
First I put the smallest one here (on C) Then I
put the nextsmallest here (on B) Then I take the
biggest one - O no, that is not allowed, OK I
move the smallest back to A And the next smallest
to C Then I take the smallest to B And the next
smallest to - where should it go...
15
A think-aloud protocol can be regarded as the
top of the iceberg

• Toppen av isberget bild

16
Some production rules that may produce the think
aloud protocol
IF goal achieved THEN end If disc1 free THEN
move disc1 If move disc1 THEN check if C is
possible IF C possible THEN move disc1 to C IF C
is not possible THEN move disc1 to A If disc2
free THEN move disc2 If move disc2 THEN check if
B is possible If B empty, THEN move disc2 to
B IF disc3 free THEN move disc3 IF move disc 3
THEN check if C is possible
17
What production rules may produce the shortest
path?

• Can production rules only solve this problem?

18
No Rules are not sufficient!
We need a system to interpret the rules! What can
the system perceive? How should the objects be
represented? In what order are the productions
tested? How will the actions performed be
remembered?
19
A cognitive architecture

• Defines how rules are interpreted
• In what order they are taken
• What conditions prevail for how the rules may be
  written (for instance how many conditions and
  actions are possible for one rule)
• How the results of actions are stored

20
From IP to HIP (Psychology has been changed to
human)

• Human beings differ from computers in several
  ways.
• Therefore, we have to define a mechanism that
  processes information in a similar way as a human
  being, a HIP
• Human Information Processor

21
A cognitive architecture for Human Information
Processing (HIP)

• Must comply with knowledge about human beings.
• Knowledge from various sources
• Senso-motoric
• Attention
• Perception
• Memory
• Metacognition

22
Visual rendering of a Human Cognitive
architecture (EPICS) (CHI 2001, p 130)
Long-term memory Productions
Cognitive processor
Task Environ ment
Production rule interpreter
Auditory input
Auditory proc.
Visual input
Working memory
Visual proc.
Vocal motor
Manual motor
23
Important HIP characteristics to be considered

• Perceptual capacities
• (time for writing, time for retrieving)
• Motor capacities
• Eye and hand movements, time
• Long-term memory (productions)
• Time for writing, time for retrieving, type of
  productions
• Working memory
• (restricts amount of material on which
  productions may work)

24
Important HIP characteristics to be considered

• Working memory
• (restricts amount of material on which
  productions may work)
• 5/- 2 chunks
• What is a chunk?
• A meaningful unit
• What is that?

Chunk på Zoo
25
Important HIP characteristics to be considered

• Long-term memory (productions)
• Time for writing, time for retrieving, type of
  productions
• Long-term memory (declarative)
• Semantic networds
• Schemata

Som Choklad-pudding
26
Characteristics of ingredients in the human
information processor
From Newell Simon, 1972 rendered by Card, Moran
Newell, 1983
Bild from C,MN
27
Applications of IP and HIP ideas

• Rule-based systems
• Knowledge base systems
• Intelligent tutoring
• User modelling
• HCI
• Analytic models
• Simulation models
• Quasi-empirical approaches

28
HIP applied to HCI

• Analytic model
• TAG Task Action Grammar
• Takes related tasks in a system, derives how many
  rules that have to be used to perform these
  tasks. The less rules, the easier to learn.

29
HIP applied to HCI

• Simulation models
• Different cognitive architectures
• ACT
• SOAR
• Input data are processed through simulated user
  model
• Results reaction times

30
HIP applied to HCI

• Quasi-empirical approach
• GOMS
• Analyses a task from an experts actions
• Goals, Operations, Methods and Selection rules
• Further applications of GOMS
• Cognitive walkthrough - what will a user find
  difficult in the system? (Goals, operations,
  methods analysed with respect to the designers
  knowledge about the user)

31
HIP applied to HCI

• Further applications of GOMS
• Keystroke level calculations How long will it
  take to perform a task with the system?
• Has been used to compare different system
  solutions, for instance for telephone operators
  asking callers questions.
• A small change in the time taken may mean much
  when many small tasks are performed by many
  persons.

32
Key-stroke level av GOMS

• Task Copy a word and position it at some place
  at the text
• Method Get the operations from the menu
• 1. Time to identify the word
• 2. Time to mark the word
• 3. Time to move to the menu and find the word
  copy
• 4. Time to click on copy
• 5. Time to go to the position in the text were
  the word should be placed
• 6. Time to click in order to move the cursor to
  this place
• 7. Time to move to the menu and get the command
  paste.
• 8. Time to click for placing the word.
• 9. Time for checking that the result is OK
• The time for the handmovements is calculated
  according to Fitts law

33
HIP applied to HCI

• Learning -
• The effect of prior knowledge
• Positive (can use old rules)
• Cognitive Complexity Theory (CCT)
• Negative (interference with old rules)
• Learning by doing
• A problem solving approach is possible

En tjusig morgon på kontoret
34
Mismatching models
Search!

• Conceptual models versus device models versus the
  users model of the system
• Designers have one particular (conceptual) model
  in their mind about how a system should work
• The system is implemented to show a model (device
  model) that may not be the same.
• When users work with the system they may
  construct yet another model of the system

35
Summary There is a gulf between perception and
execution. We may calculate the effects.

Goals
Intentions
Evaluation
Execution
Perception
System
(Norman, 1986)
36
Recent uses of IPP-models(from CHI 2001)

• Out of 69 papers, eight use some kind of theory.
• HIP theory is used in all eight cases.

37
Recent uses of HIP-models(Examples from CHI 2001)

• Ignoring Perfect Knowledge-in-the-world for
  Imperfect Knowledge-in-the head Implications of
  rational analysis for Interface Design
• Predicting the Effects of In-Car Interfaces on
  Driver Behavior Using a Cognitive Architecture
• Towards Demystification of Direct Manipulation
  Cognitive Modeling Charts the Gulf of Execution
• Beyond Command Knowledge Identifying and
  Teaching Strategic Knowledge for Using Complex
  Computer Applications

38
Conclusions

• HIP-models have a narrow range of application
• Within this range, they are surprisingly
  successful
• More so than any other models or theories within
  HCI.

How is success defined?
How do we know which applications?