Time Sharing

1
Time Sharing

• chapter 11

2
Time Sharing

• Divided attention attention is divided between
  several tasks at once.
• Ability to succeed affected by
• Resource Demands
• Switching costs and Allocation skills
• Structural Limits (e.g. bottlenecks)
• Confusion

3
Automaticity and Resources

• Automaticity
• highly overlearned skills that require no
  attention
• riding a bike (assuming youve done it for years)
• a continuum from no automaticity to purely
  automatic
• Attentional Resources
• Mental resources (effort) are limited.
• Amount demanded affected by task difficulty and
  degree of automaticity.

4
Resource Theory

• What are attentional resources?
• Not a mental energy or a fuel
• An analogy
• sometimes time-slices of tissue use
• Other times, and epiphenomen of SDT
• e.g. one you pay attention to one visual object,
  there is only one source of signal
• when you pay attention to two items, one item now
  adds noise to the others SNR (d), and
  vice-versa
• As more and more items are attended, each item
  becomes noisier
• makes it appear as if attention is spread too
  thin.

5
Resource Theory

• What are attentional resources?
• Resources are a good analogy (heuristic)
• Do a good job of capture most attentional
  phenomenon
• However, what is happening at the implementation
  level is much different.

6
Performance Resource Function
high
low
high

• Easy tasks demand less resources than hard tasks
  for the same level of performance

7
Automaticity and Resources

• Data Limits (Ceiling Effect)
• Adding more resources to a task does not improve
  performance.
• The amount of data, not resources limits
  performance
• External data limit
• Trying to rehearse a 2 digit number is trivially
  easy
• identifying a gray car in the fog
• Internal data limit (e.g. knowledge)
• Trying to understand an unfamiliar language

8
Automaticity and Resources

• Resource Limits
• If adding resources to a task improves
  performance, then the task was said to be
  resource limited.
• Tasks that are more difficult demand more
  resources to reach the same level of performance
  as an easy task.

9
Data Resource Limits
Data Limited
Resource Limited
Performance
Resources
10
Hydraulic Model

• Views attention as a resources that can be
  allocated in any manner, but can not exceed 100.
• Arousal has been analogized as the total amount
  of attention available.
• Higher arousal rising waters raise all boats

11
Time Sharing and PRF

• If two tasks demand more resources than
  available
• then performance of at least 1 task will suffer
• Likewise, diverting resources to one task will
  cause a drop in resources available to the other
  task.
• As long as performance is not data-limited, an
  increase in performance in one task will lead to
  a drop in performance in the other task.

12
Time Sharing and PRF
13
Automaticity and Difficulty

• Highly automatic tasks will require little or no
  attention.

More Automatic
Performance
Less Automatic
Resources
14
Automaticity and Difficulty

• Highly automatic tasks will be less likely to
  interfere with a secondary task.

High Automaticity

• Likewise, tasks with low automaticity will demand
  more attention, leaving little left over for
  other tasks.

Performance
Resources
100
0
Low Automaticity
Performance
100
Resources
0
15
Automaticity and Difficulty

• Highly automatic tasks will be less likely to
  interfere with a secondary task.

Low Automaticity

• Likewise, tasks with low automaticity will demand
  more attention, leaving little left over for
  other tasks.

Performance
Resources
100
0
Low Automaticity
Performance
100
Resources
0
16
Costs of Multitasking

• Each time a task is switched, there is some
  switch-cost in loading a new task set.
• Therefore, the more switches that are made, the
  more times one will incur a task-switching cost.
• Rubinstein et al (2001)
• Trying to juggle two tasks at once is less
  efficient than finishing one task before starting
  another.

17
Structural Factors

• Bottlenecks
• In PRP, two tasks interfere with each other only
  if they attempt to use the same stage of
  processing.
• Multiple Resource
• The mind is made up of several independent units,
  each with its own resource pool.
• If two tasks need to access the same units, then
  they will have to share resources.
• Tasks that use separate units do not interfere
  with each other.

18
Multiple Resources (Wickens)

• 3 (mostly) orthogonal dimensions
• Stages
• Perception, WM, Cognition
• Responding
• Modality
• Auditory
• Visual
• Codes
• Verbal
• Spatial

19
Stages

• Perception, Working Memory, Cognition are
  different sides of the same coin
• Example
• Imagery/visualization is perception without
  sensory input.
• Solving a spatial task involves
  imagery/visualization
• How do I get that couch down the stairs?
• All of these tasks try to use the same tissue
  (e.g. visual cortex).
• Response Selection
• choosing and launching motor responses

20
Modalities

• Visual Auditory
• It is easier to perform two task when they use
  different modalities then when they use the same
  modalities
• Easy driving and listening to the radio
• Hard holding a conversation and listening to the
  radio

21
Codes

• Verbal vs. Spatial
• Verbal
• Reading text, listening to a conversation
• Spatial
• Object rotation, auditory spatial localization
• Easy (little overlap)
• reading sheet music and playing an instrument
• Difficult (great overlap)
• singing and playing an instrument

22
Perceptual Modality Codes

• Modalities Visual vs. auditory
• Codes Spatial vs. verbal

Codes
Modalities
23
Problems with MRS

• MRS adds artificial dichotomies
• Notice (fig 11.5)that the sides are not
  completely orthogonal.
• Good heuristic, but misses some nuances

24
Problems with MRS

• Just et al. (in press)
• Mental rotation auditory sentence comprehension
  tasks.
• Should not interfere with each other
• Performance in dual task should be as good as in
  isolation.
• The parts of the brain that light up in the
  single task should be unaffected in the dual task
• e.g. the dual task image should look like the sum
  of the two single task images.

25
Problems with MRS

• Just et al. (in press)
• Performance for both tasks was poorer when
  performed together than when performed alone.
• The pattern of activation for both areas of
  interest was substantially less than in the
  single tasks.
• i.e. amount of activation correlated with
  behavioral performance.
• Both overt and hemodynamic behaviors changed in
  the dual-task condition, suggesting that these
  two tasks, which should use separate pools of
  resources according to RMS, nonetheless interfere
  with each other.

26
(No Transcript)
27
Cell Phone Driving

• Strayer Johnstone (2001)
• Subjects either listened to radio or were engaged
  in a conversation.
• Radio Group
• did as well on the driving task as hey did
  without the radio.
• Phone Group
• more likely to miss a stop-light (7) than when
  they performed the driving task alone (3).
• How does MRS explain this?
• Central Executive?

28
Confusion

• Limitation not due to attention
• crosstalk interferes with performing the task
• semantic interference/priming
• Example Stroop Task BLUE
• lateral visual masking (crowding)
• Hard zxz
• Easy z x z

29
Conclusions

• Automaticity affects resources needed to perform
  at a certain level.
• Data limited adding attention/effort/resources
  has no effect on performance.
• Resource limited paying more attention/giving
  more effort increases performance.
• If two concurrent tasks demand more resources
  than there are available, then performance in at
  least 1 task will suffer.
• Multiple Resources the more similar the tasks,
  the more likely they are to interfere with each
  other.

30
Predicting Multi-task Performance

• Relative Predictions
• Which configuration will provide better
  multi-tasking performance?
• While driving, is it better to change the radio
  using manual controls or voice controls?
• Absolute Predictions
• What are the limits of human performance?
• For air traffic controllers, at what level of
  traffic does performance start to break down?

31
Predicting Multi-task Performance

• Time-Line Analysis
• Assumes that the proportion of time spent on a
  task is indicative of its workload.
• i.e. if you spend 100 of your time on 1 task,
  assumes that 100 of workload was on that task.

32
(No Transcript)
33
Time-Line Analysis

• Tells us
• Proportion of total time spent on a task
• i.e. checking rearview mirrors
• Potential bottlenecks, when time on two tasks
  overlaps.
• Can id when too much time is spent on a lesser
  task
• Problems
• If people have the chance to reschedule, then
  might not perform 2 tasks at same time.
• Only measures overt indicators.
• Not sensitive to resource demands of different
  tasks.
• Do overlapping tasks represent a potential
  bottleneck or are both easy enough to do at once?

34
Assessing mental workload

• How busy is the operator?
• How complex are the tasks?
• Will the operator be able to respond to
  unexpected events?

35
Importance of Workload

• Workload Prediction
• Assessment of workload imposed by equipment
• Goal optimize system to minimize workload
• Operator Assessment
• Choosing between operators
• Can the person keep up with the workload?
• Lightly stressed operator preferred over highly
  stressed.

36
Importance of Workload

• What is Workload, anyway?
• Workload is a function of the amount of capacity
  demanded to reach an adequate level of
  performance.
• A skilled worker will require fewer resources to
  reach the same performance level, and therefore
  has a lighter workload.
• Thus, workload is a function of operator-capacity
  and task-resource demands.

37
Importance of Workload
Skilled

• A skilled worker will require fewer resources to
  reach the same performance level, and therefore
  has a lighter workload.

Unskilled
Performance
Resources demanded
38
Importance of Workload
Performance
Reserve Capacity
Resources Supplied
Resources demanded
39
Importance of Workload
Demand lower than Capacity
Performance
Reserve Capacity
Resources Supplied
Resources demanded
40
Importance of Workload
Demand higher than capacity-poor performance
Performance
Reserve Capacity
Resources Supplied
Resources demanded
41
Criteria for Workload Assessment

• Sensitivity
• Index must be sensitive to changes in task
  difficulty or resource demand
• No Ceiling or Floor effects.
• Diagnosticity
• Must indicate what aspect of the task increased
  resource demand
• Selectivity
• Should not be measuring other effects, such as
  physical load or emotional distress.
• i.e. should not be correlated with other measures
• Obtrusiveness
• Reliability

42
Criteria for Workload Assessment

• Sensitivity
• Diagnosticity
• Selectivity
• Obtrusiveness
• The index should not affect performance, which
  would in turn affect what was being measured.
• Reliability
• Results should be repeatable.

43
Primary Task Measures

• Ideally, we would measure people on the primary
  task
• However, that is not always workable
• A primary task might lie in the underload region
  of the supply-demand space
• i.e. Floor Effect.
• Since both tasks are at floor, measurement can
  not discriminate between the two.

44
Primary Task Measures

• Ideally, we would measure people on the primary
  task
• However, that is not always workable
• A primary task might lie in the underload region
  of the supply-demand space
• Two primary tasks might be measured differently,
  making comparisons difficult
• e.g. one measured in accuracy, the other only in
  speed.
• e.g. Accuracy in both tasks needs to be similar
  in order for RT comparisons to be meaningful
  (speed-accuracy trade-off).

45
Primary Task Measures

• Ideally, we would measure people on the primary
  task
• However, that is not always workable
• A primary task might lie in the underload region
  of the supply-demand space
• Two primary tasks might be measured differently,
  making comparisons difficult
• Sometimes it is impossible to obtain good
  measures of primary-task performance
• e.g. vigilance task, which by its nature
  provides very sparse data, yet might have a high
  workload.

46
Primary Task Measures

• Ideally, we would measure people on the primary
  task
• However, that is not always workable
• A primary task might lie in the underload region
  of the supply-demand space
• Two primary tasks might be measured differently,
  making comparisons difficult
• Some times it is impossible to obtain good
  measures of primary-task performance
• Two primary tasks might differ in performance,
  but not because the resource demands were
  different, but because of data limits.
• One task might be ridiculously easy and the other
  ridiculously hard
• In both instances, performance is unaffected by
  the amount of resources devoted to the task at
  hand.

47
Secondary Task Measures

• PRP
• used the second task to measure performance in
  first task.
• delays in responding to T2 were due to processing
  of T1.

48
Secondary Task Measures

• Benefits
• High Face Validity
• Measures the amount of residual
  attention/performance when performing the primary
  task, which is usually what we are interested in.
• That is, it measures excess capacity, or the
  amount left over from performing the primary
  task.
• The same measure can be used for different
  primary tasks.
• Can measure resource differences in the primary
  tasks even when the primary tasks are near
  ceiling.

49
Secondary Task Measures

• An easy primary task will use fewer resources
  than a difficult primary task.

50
Secondary Task Measures

• leaving more resources for the secondary task
• leading to better secondary task performance.

51
Secondary Task Measures

• Example Cell phones and driving
• Primary tasks
• Cell phone conversation
• Listening to the radio
• Secondary task
• driving
• Design
• compare driving measures (lane drift, braking RT,
  etc.) for both primary tasks.
• Any differences in 2ndary task measure must be
  due to primary tasks degree of interference
  w/2ndary

52
Secondary Task Measures

• Costs of Secondary Tasks
• Must use the same kind of resource
• If the question is do these 2 tasks interfere
  with each other (i.e. driving cell phones),
  then this is not a concern.
• If the question is how much reserve capacity is
  left after the primary, then this is a concern.
• The secondary task can interfere with performance
  of the first.

53
Secondary Task Measures

• Costs of Secondary Tasks
• Must use the same kind of resource
• If the question is do these 2 tasks interfere
  with each other (i.e. driving cell phones),
  then this is not a concern.
• If the question is how much reserve capacity is
  left after the primary, then this is a concern.
• The secondary task can interfere with performance
  of the first.
• If the secondary task shows signs of interfering
  with the primary task, then this makes
  interpreting the secondary task data difficult.

54
Physiological Measures

• Heart Rate
• Appear to measure total demand rather than demand
  on a single task.
• That is can not tell which of two concurrent task
  lead to the increase in heart rate
• Contaminated by increased HR due to physical
  effort.
• Pupillary Dilation
• Eye movements
• EEG/ERPs
• fMRI

55
Physiological Measures

• Heart Rate
• Pupillary Dilation
• Similar problems to heart rate measure
• More intrusive head restraint or head mounted
  cameras.
• Eye movements
• EEG/ERPs
• fMRI

56
Physiological Measures

• Heart Rate
• Pupillary Dilation
• Eye movements
• Longer dwells
• a sign that processing is less efficient.
• Fixation location
• when paired with dwell times, can tell us which
  tasks people are spending time on.
• EEG/ERPs
• fMRI

57
Physiological Measures

• Heart Rate
• Pupillary Dilation
• Eye movements
• EEG/ERPs
• EEG - disagreements over meanings off alpha,
  beta, etc. waves.
• ERP - Event related components require MANY
  trials to get a clean signal.
• Electrodes are intrusive, and entail long set-up
  time
• Not very portable, easiest done in lab
• fMRI

58
Physiological Measures

• Heart Rate
• Pupillary Dilation
• Eye movements
• EEG/ERPs
• fMRI
• Must be done in MRI machine (claustrophobia)
• Task often are greatly simplified and might not
  generalize to complex real-world tasks
• Expensive!

59
Physiological Measures

• Benefits
• continuous recording over the course of the task
• Relatively unobtrusive, as far as task demands
• e.g. often does not change the task, as a
  secondary task might
• Costs
• Sometimes physically intrusive
• electrodes, fMRI tube, chin rest, etc.
• Are not a direct measure of interference
• whereas a secondary task directly measures how
  the primary task interferes with 2ndary

60
Subjective Measures

• Example NASA TLX
• Benefits
• Quick, easy, cheap
• Since administrated after task is over, can not
  interfere with task
• Costs
• Subjective introspection - perception of
  workload is imperfectly correlated with actual
  task demands.
• Might reflect other biases, such as physical
  demands or dislike of the task.

61
Workload Conclusions

• Workload is amount of resources demanded by a
  task to reach an acceptable level of performance.
• Goal is to minimize workload change task, change
  interface, or change operator.
• Secondary tasks provide best measure of excess
  capacity.
• If secondary task interferes with primary, data
  might be difficult to interpret.
• Physiological can be less obtrusive than using a
  secondary task.
• Subjective measures even less obtrusive, but are
  introspective in nature.