HF Ch 11+12
Secondary tasks
: if someone is performing a primary task, ask them to perform a secondary task as well. Performance on the secondary task indicates how much workload is being consumed by the primary task Note - secondary tasks degrading primary task performance is a potential problem
Reasons to Automate
A task is impossible for a human to perform Humans have performance limitations, so a machine would be better at it Humans are doing a task well enough, but performance could be augmented or workload saved by assisting them with automation Economical reasons: machines don't want a paycheck! Productivity: outside of maintenece/repairs, machines can work forever; like economical reasons, machines won't demand petty human requirements like "food" or "sleep" - amateurs!
Environmental Solutions
Basically, if we can change the environment to remove a stressor, go ahead and remove it!
Arousal Theory
An individual's level of activity, usually as a result of the influence of some stressor stimulus (or stimuli) We can measure this through many of the same physiological measures we talked about with workload This close association with workload makes sense: increased arousal "trying harder" to complete a task; this is essentially the same thing as workload
Effects of Stress Components
Attentional Narrowing Selective Attention; Distraction Working Memory Loss Perseveration Under high levels of stress, people will tend to continue on with their present course of action; this is seen as an impact of stress on decision-making, driving people toward confirmation bias Strategic Control: can have positive or negative outcomes
Subjective measures
Bedford, Modified Cooper-Harper These are totally dependent on self-reports which can be difficult to compare Multidimensional workload: NASA-TLX ("task load index")
Further into the future
Brain-computer interfaces Injured/Disabled Why not the healthy too!
Dynamic function allocation
Break it down - we talked about function allocation earlier in the chapter; now what happens when a system does it dynamically, in real time? That's what dynamic function allocation is!
Design Solutions
Can we provide the user with additional information? Can we automate something? Can we provide additional support in emergency situations? Can we determine what cognitive constructs are being most taxed and redesign to relieve that?
Variables impacting trust/dependence
Complexity of the automation process: the less a person understands or relates to what the machine's process of the automated actions are, the less they will trust the machine's automation Perfect example from "Office Space"
Measurement tools
EEG: Electroencephalogram ERP: Event-related potentials Cerebral blood flow ultrasounds fMRI Heart rate Pupil diameter Visual scanning: eye tracking, fixation duration
Complacency
over-trust - users blindly accept the machine's functioning. In its worst form, the users shirk their responsibilities under the assumption that automation will take care of it.
Behavioral measures
simply observing the actions of the worker Note - can't determine internal cognitive processes
Calibration curve
slide
Levels of Automation (not all or nothing)
slide
Mental workload
sometimes the task is just really difficult, and taxes our cognitive resources heavily We have limited resources to deal with all of the differing cognitive aspects of any task we're given to complete The level of mental workload is discussed in the same way we discuss physical workload: rather than demands on the energy and force our muscles are capable of producing, it's the demands on the processing power of our different cognitive processes
Training
How can we prepare people ahead of time to cope with stress? Can we teach them to move their thinking toward some of the more beneficial strategic control outcomes? Can additional exposure to stressors ahead of time train them for stress in the field?
Fitts' List (remember this guy?) (keep in mind it's from 1951)
Humans are better at: detecting minor sensory information, discerning patterns in sensory information, improvisation & flexibility, long-term memory store is lasting and able to recall relevant information quickly, inductive reasoning, exercising judgment Machines are better at: responding quickly to control signals, applying great amounts of force smoothly and precisely, performing repetitive/routine tasks, storing information briefly and deleting it entirely, deductive reasoning, complex reasoning/multitasking
The NASA Task Load Index (NASA TLX) is an example of a ________ method of mental workload assessment.
subjective
Dependence
the behavioral, objective actions that a person takes when interacting with an automated system
Trust
the cognitive, subjective evaluation of an automated system by a person
Automation
Imagine aviation or healthcare without it!
Stress
the nature of a task itself can be stressful (environmental conditions, temporal pressure, consequences), and we can bring our own personal stresses from our everyday lives into whatever we're doing
Is higher workload "bad?"
this can be subjective to some degree, much like our discussion of what makes for "good" decision making. Like that, there are some general guidelines to determine whether a given level of workload may be "good" or "bad."
The exponential rate of changes over the years in the speed and capacity of computers is known as:
Moore's Law
One of the most critical factors influencing trust in automation is:
automation reliability
Not without its flaws
automation-induced complacency
Current research
big focus on genetic differences in cognition Expect this data to become big in the near future as genetic analysis becomes more commonplace!
Remediating the negative effects of stress on performance should focus on:
both design and training
How do we adapt to increased workload?
People allow their task performance to degrade People find a way to perform tasks more efficiently, using fewer resources People eliminate tasks perceived to be unnecessary Optimal vs. non-optimal
Individual differences
People have varying levels of skill at different tasks - it is almost inevitable that you'll be a low performer on some things, and a high performer on others
Strategic Control
People take on a set of strategies to consciously attempt to cope with added levels of stress Recruitment of more resources: "try harder" Remove the stressor Change the goals of the task: adaptation Do nothing
Three main manifestations of stressors in people
Phenomenological experience: emotions, episodic memory Physiological arousal Change in performance characteristics/information processing
Automated systems that give frequent warnings that turn out to be false reduce operator trust and dependence, so that when a true alarm does occur, operators may ignore it. This phenomenon is known as:
cry wolf effect
Human-centered automation
designing automation for JOINT human-machine work, not a machine taking over a human's work
In order to avoid the problem of disruption of primary task performance when using the secondary task method of workload measurement, investigators have used:
embedded secondary tasks
Moore's Law
exponential rather than linear growth of computational power
Automation is often implemented in the work place because of:
human performance limitations economic factors safety factors
Yerkes Dodson Law
The level of performance on a task improves as arousal increases, but only to a limited threshold. Beyond that threshold, performance drops sharply.
Absolute workload
The overall "total" workload - composed of the combined smaller parts of mental workload, as well as other forms of workload (e.g. physical) - as you might expect, difficult to quantify
Neuroergonomics
The study of the human brain in relation to performance at work and in everyday settings
Transactional/Cognitive Appraisal Theories
These approaches focus instead on the reaction elicited by the stressor More of an "individual differences"-flavored approach Essentially, what we can study as an outcome of stress is an interaction between the environmental (stimulus) impact and the individual's cognitive appraisal of the situation
Which areas of data/research are most prone to this? (Individual differences)
Time-sharing/multi-tasking Working memory
"Cry wolf" effect
under-trust - users ignore automated warnings/indicators because they occur too frequently or at too low a threshold. In its worst form, the users will allow potentially dangerous/deadly events to occur because they fail to follow up on an automated warning.
Major concerns
What to adapt: pretty much the same considerations as regular automation When to adapt: environmental triggers vs. performance BEWARE: system lag can throw this off! Who decides to adapt: does the change occur because the human requests it? Or does the machine step in when it detects a need for it? Both?
Embedded secondary tasks
keep these related to the primary task so as to prevent them from being intrusive
A computerized diagnosis system that offers a complete set of decision/action alternatives to the human decision maker is an example of automation at a relatively ________ level.
low
Relative workload
must consider that our "mental resources" are not one monolithic structure: it's composed of smaller parts, and different parts can be taxed at different rates Think: Automaticity!
Stages of Automation
While our levels of automation are about the amount of control yielded to the machine by the human, stages of automation are the types of control yielded to the machine by the human Closely mimics the human information processing model Stages: Information Acquisition Information Analysis Decision Making and Action Selection Action Implementation
Automation reliability
Who counts cash when you take it out of an ATM? 100% reliability is rare unless it's extremely simple automation
The inverted-U relationship between performance and arousal is known as:
Yerkes Dodson Law
Stressor
any stimulus which causes stress Environment: noise, vibration, heat, dim lighting, high acceleration, etc. Psychological: anxiety, fatigue, frustration, anger