Cognitive Engineering PSYC 530 Attention and Multi-Tasking

1
Cognitive EngineeringPSYC 530Attention and
Multi-Tasking

• Raja Parasuraman

2
Overview

• Mechanisms of Time Sharing
• Multiple-Resource Theory
• Mental Workload Assessment

3
The Proliferation of Multi-Tasking

• Tower air traffic controller (visual search,
  speaking, listening)
• Driving with in-vehicle technologies
• Financial analysts (speaking and listening)
• Simultaneous language interpretation (listening
  and speaking)

4
Success and Failure in Dual-tasking

• Need baseline of single-task performance
• Some dual tasks can be performed concurrently as
  well as they can in isolation (e.g., reading
  music and playing in skilled pianists)
• Other dual-task pairings show a decrement
  compared to single-task performance
• Tasks with automatic processing can often be
  time-shared without decrement

5
Automatic Processing

• Consistent mapping between stimulus and its
  categorization and response (Shiffrin
  Schneider, 1977)
• Visual search with same targets and distractors
• a, e, u vowels c, g, k consonants
• driver seat in car on left side drive on right
  side of road
• Extensive practice
• Consistent mapping and practice can lead to
  automaticity

6
Automaticity and the Brain
Before consistently-mapped practiceresource
loading
After consistently-mapped practiceautomaticity
7
Automatic Processing

• Automatic tasks can be time-shared efficiently
  with other resource-demanding tasks
• Walking and decision making
• Driving on familiar route with little or no
  traffic and conversation with passenger
• However, not all tasks can become automatic

8
Mechanisms of Time-Sharing

• Resource demand
• Attention allocation and switching
• Multiple resources
• Confusion and similarity

9
Performance Resource Function (Norman Bobrow,
1975)
data-limited region
easy or practiced task
resource-limited region
Performance
difficult or unpracticed task
Resources applied to task
10
Time-Sharing and the PRF

• To the extent that two tasks A and B are
  resource-demanding, allocating more resources to
  A will improve performance on A but degrade
  performance on B because of the consequent
  withdrawal of resources from B
• If tasks A or B are data-limited for some region
  of the PRF, perfect time sharing is possible
• The optimal strategy for allocating resources
  between tasks may not always be used because of
  the negative utility of high effort hence
  heuristic strategies that conserve effort are
  often used (cf. decision making models)

11
Attention Allocation Strategies

• Sub-optimal allocation strategies can lead to
  poor dual-task performance (e.g., allocation of
  resources to an automatic task)
• Variable-priority better than fixed-priority
  training for dual-task performance (Gopher et
  al., 1989)
• Variable priority Task A 90, Task B 10 then
  Task A 10, Task B 90 finally 50/50
• Fixed priority Task A 100, Task B 0 then Task
  A 0, Task B 100 finally 50/50

12
Training Attention Allocation

• Variable-priority training leads to better
  transfer to a complex video game (Space Fortress)
• Similar training on the video game transfers to
  better performance in piloting fighter aircraft
  and better chance of selection by Israeli Air
  Force (Gopher et al., 1994)

13
Multiple Resources

• Single resource demand cannot always predict
  extent of dual-task decrement (e.g., driving
  reading a book vs. driving listening to a book)
• Stages of processing define one dimension of
  different resource type (perception/working
  memory vs. response selection and execution)
• Working memory research also suggests a dimension
  of verbal vs. spatial codes of processing
• Input/output modalities also define a third
  dimension of resources

14
Multiple-Resource Theory (Wickens, 1984)
15
Multiple Resources and Dual Task Performance

• Overlap between different resource dimensions
  determines degree of dual-task decrement
• Perfect time sharing is possible if there is no
  overlap, e.g., sight reading music and auditory
  shadowing (Allport et al., 1972)
• Reading music visual, manual
• Auditory shadowing auditory, verbal, vocal

16
Multiple Resources and Dual Task Performance

• However, lack of overlap does not guarantee
  perfect time sharing
• Difficulty of central processing or other factors
• Confusion
• Similarity)
• Central bottle neck response selection
  (Pashler, 1988)
• All of these may still lead to dual-task
  decrement

17
Cell Phone Use During Driving

• Strayer Johnston (2001) Experiment 1
  Pursuit tracking combined with
• cell phone conversation
• radio control (listen to broadcast)
• Respond to red light by braking
• Single task or dual task

18
Cell Phone Use During Driving

• Experiment 2 Pursuit tracking combined with
• shadowing (repeating words)
• generation (produce word from last letter of
  heard word)
• Easy or difficult driving
• Single task or dual task

19
Some Practical Implications of Multiple-Resource
Theory

• User-computer interfaces
• Control of of cursor for text-reading tasks
  (Martin, 1989)
• keyboard
• mouse
• voice control

20
Some Practical Implications of Multiple-Resource
Theory

• In-vehicle navigation systems Visual vs.
  auditory navigational commands to driver (Parkes
  Coleman, 1990)
• Piloting and control of cockpit automated
  systems Voice vs. manual input to system
• depends on whether pilot is
• actively flying (spatial)
• or talking to co-pilot or ATC (verbal) (Wickens
  et al., 1993)

21
Confusion and Similarity

• Cooperative sharing of common property between
  tasks enhances time sharing
• Objects in close proximity (lateral and vertical
  flight controls in aircraft)
• Manual and vocal responses mapped to a single
  stimulus
• Confusion similarity between stimuli and
  responses between tasks degrades time sharing
• two spelling tasks vs. spelling and arithmetic
  tasks
• cross talk between tasks

22
Mental Workload

• Effects of task difficulty
• Effort involved in maintaining performance
• Workload is an intervening variable between task
  and environmental demands and operator
  performance
• Workload defined by relationship between task
  demand and resource supply

23
Workload and Performance
Resources Demanded
24
Workload Measurement Criteria

• Sensitivity
• Diagnosticity
• Selectivity
• Obtrusiveness
• Bandwith and Reliability

25
Workload Measurement Methods

• Primary Task Measures
• Secondary Task Measures
• Physiological (Neuroergonomic) Measures
• Subjective Scales

26
Primary Task Measures

• Accuracy, Speed, Errors, etc. on task of interest
• Can be related directly to system design issues
• However, primary task measures can be insensitive
  to differences in workload
• two tasks with same performance but in different
  parts of the underload regiondifferent reserve
  capacity
• performance may be affected by factors other than
  task demand or operator capacity

27
Secondary Task Measures

• Operator asked to perform primary task as well as
  possible
• Also asked to perform secondary task to best or
  his/her abilitya measure of reserve capacity
• Examples
• Simple/choice reaction time to tones
• Random number generation
• Sternberg memory search task
• Time estimation/production
• Embedded secondary taskslower priority tasks
  forming part of overall operator job (e.g.,
  monitoring flight strips in air traffic control)

28
Physiological/Neuroergonomic Measures

• Heart rate and heart rate variability (0.1 Hz
  component)
• Pupil size
• Visual scanning
• EEG alpha and theta
• Event-related potentials (ERPs) P300 component
• Transcranial Doppler Sonography (TCD) cerebral
  blood flow velocity
• Functional magnetic resonance imaging (fMRI)
  blood flow in prefrontal cortex
• Near infra-red spectroscopy (NIRS) optical
  imaging of cerebral blood flow

29
Subjective Measures

• NASA Task Load Index (NASA-TLX)
• Subjective Workload Assessment Techniques (SWAT)
• Boles Subjective Scale

30
Workload and Performance

• Workload is not the same as performance
• Sometimes performance is correlated with
  performance

31
Workload and Performance

• But workload can also be dissociated from
  performance
• Two people performing the same task can have
  identical performance, yet one may do so with
  many spare attentional resources left to allocate
  to concurrent tasks, and the other not so
• Poor performance may not necessarily be due to
  excessive workload but other factors
• poor interface design
• unstable control device

32
Workload/Performance Dissocations

• Performance may be stable even with higher task
  load and increased perceived workload
• Sperandio (1971) study of air traffic control
  Controllers maintained performance as more
  aircraft entered sector by
• spending less time talking to pilots
• processing fewer aircraft variable
• Driving workload can be a better predictor of
  future performance than current performance
• Steering, lane, and speed deviations may be
  currently normal in overloaded driver
• At a later time, however, overload leads to
  performance decrement