Task 8 Flashcards
Automation
Technical developments make it possible to automate many aspects of a system
AUTOMATION
1. automatic control of the manufacture of a product through a number of successive stages
2. Automatic control in any branch of industry or science
3. use of electronical or mechanical devices to replace human labor
automatic control, which can be open loop as well as closed loop and can refer to electronic and mechanical action.
System design issue
which functions should be automated?
levels of automation
Four stages of human info processing has its equivalent system function that can be automated:
- possible to differing degrees or many levels
- ACQUISITION AUTOMATION: sensing and registration of input data (info acquisition and sensory processing)
a) low level: mechanically moving sensors in order to scan and observe. visual and haptic sensors could also be used
b) moderate: organize incoming data according to some criteria
c) more complex: certain items are selected and brought to operators attention. Highlighting and filtering can lead ti differing human performance consequences - ANALYSIS AUTOMATION: info analysis involves cognitive functions such as WM and inferential processes (perception)
a) higher level: integration of several input variables into a single value
b) more complex: info managers that provide context-dependent summaries of data to user - DECISION AUTOMATION: Selection from decision alternatives: Varying levels of augmentation or replacement of human selection of decision options with machine decision making
- expert systems: use conditional logic to chose
Levels of automation as defined by original taxonomy by Sheridan - no assistance by computer
- complete set of decision alternatives is offered
- narrows selection down
- suggests one alternative only
- Executes suggestion if human approves it
- human has restricted time to veto before automatic execution
- Executes automatically and informs human
- Executes automatically and only informs if asked
- Decides and informs only if itself decides to
- Decides everything without human
4 ACTION AUTOMATION: execution of action
- defined by relative amount of manual vs automatic activity in executing the response
ADAPTIVE AUTOMATION: levels of automation could be divided to vary depending on situational demands during operational use (context dependent)
Framework of automated design
Series of steps proposed to use it correctly:
- WHAT SHOULD BE AUTOMATED?
- WHAT LEVEL OF AUTOMATION SHOULD BE APPLIED WITHIN EACH FUNCTIONAL DOMAIN?
To evaluate these questions, one should consider:
- PRIMARY EVALUATIVE CRITERIA: evaluate level of automation by considering associated human-performance consequences
a) mental workload: automation can change human operator workload to a level that is appropriate for a system to be performed
- clumsy automation: a system in which automation is difficult to engage
b) situation awareness: automation of decision making may reduce operators awareness of system and certain dynamic features of work environment
operator: nable to sustain a good picture of info sources leading to a decision
c) complacency: if automation is highly but not perfectly reliable, operator may stop monitoring and miss occasional fails –> effect of over-trust
- automated cuing leads operator to pay less attention to on-cued areas
d) Skill degradation: when systems overtake a function, the skill of humans will decay and be forgotten - SECONDARY EVALUATIVE CRITERIA:
Include automation reliability and cost of decision/action consequences
- evaluates feasibility and appropriateness of a certain level of automation
a) automation reliability: reliability influences human trust, which is important (low trust: undermine benefit)
- operator needs access to raw data and be aware of level of unreliability (often overestimated)
b) Costs of decision/action outcome: system cary in cost that occur if systems are incorrect
-risk: cost of error x probability of that error
I. little risk = out of loop problems that don’t have much impact
When a system is highly automated, error trapping mechanisms can give an opportunity to stop action
PROBLEM: unclear how costs and benefits should be weighted
Humans and AI are joining forces (presentation article)
Firms achieve most significant performance improvement when humans and machines work together
Humans: leadership, teamwork, creativity, social
Machines: speed, scalability, quantitative skills
5 PRINCIPLES OF OPTIMIZING COLLABORATION BTW THEM:
- Re-image business processes
- Embrace experimentation/employee involvement
- actively direct AI strategy
- Responsibly collect data
- Re-design work to incorporate AI and cultivate skills of employees
HUMANS ASSISTING MACHINES: humans need to perform 3 crucial roles
- Training: machine learning algorithms must be taught the work they need to do
- how to best interact with humans (sympathy)
- trained by humans - Explaining: AI reaching solutions that are hard to understand: humans are needed to explain that
- Sustaining: employees who work to ensure that AI systems are working properly, safe and responsible
- ethical norms
- user privacy
MACHINES ASSISTING HUMANS:
- Amplify: AI boost analytic and decision making skills by providing right info at right time
- can heighten creativity - interacting: enables interaction in novel ways btw employees and customers (transcribe a meeting)
- Embodying: embody human skills to extend our physical capability
5 CHARACTERISTICS OF BUSINESS PROCESSES THAT COMPANIES USUALLY WANT TO IMPROVE:
- Flexibility: more flexible process (e.g. new level of customization of mercedes due to AI)
- Speed: e.g. credit card fraud detection
- Scale: poor scalability is often main obstacle in improvement. Particular true of processes with least atomization (e.g. hiring of new employees)
- Decision making: providing employees with tailored info and guidance
- Personalization: providing customers with individually tailored brand experience
NEED FOR NEW ROLE AND TALENT: commitment to developing employees with fusion skills that enable them to work efficiently with AI
- people must learn to delegate tasks
- knows how to combine human skills with ai
- teach agents new skills and undergo training
organizations will then be organized around different skills rather than rigid job titles
Trust
Is an important factor in Human machine interaction (HRI) especially in dangerous conflict situations (military).
- it also might create challenges
- affects willingness to accept robot-produced info or follow robot suggestions
BOTH DISTRUST AND OVER-RELIANCE CAN UNDERMINE THE VALUE OF HRI
NEGLECT TOLERANCE: decline of robot performance if human attention is directed elsewhere as complexity increases
- too much neglect: difficult to regain situational awareness
- too little: not attending to own task
TRUST CAN BE DYNAMICALLY INFLUENCED BY FACTORS:
- Human related: attentional capacity, expertise, personality
- Robot related
a) performance based: behavior, dependability, levels automation, predictability
b) attribute based: personality, type… - Environmental: Team collaboration and task
RESULTS
- human and environment is not strongly linked to trust
- Robot performance is strongly associated with trust development
But in experimental studies: environmental factors are moderately associated
- no human factor influence
Human error
Whenever there are humans, there will be accidents and incidents
- results of worker behavior: some workers have more accidents than others
After an accident, blame is placed on human error because:
a) humans are in control of the system
b) economic reasons
Risk perception and behavior
RISK PERCEPTION AND BEHAVIOR:
Accident behavior: precedes accidents and is a number of unsafe actions
Risk homeostasis theory: people act in order to maintain a certain level of risk all the time
RISK= cost x probability
- people generally underestimate risks in familiar tasks and overestimate it in unfamiliar tasks
Risk taking as part of decision making: diagnose of risky situation and generation of alternatives that are evaluated
Dynamics of accident causation model
A number of latent and active failures com together and produce the ‘impossible accident’
- the person being responsible for the last mistake is blamed, while the other mistakes persist
Multi causality- accidents are the result of many little events/mistakes
Defining error (types of errors)
a) reversible vs irreversible
b) random, systematic (always the same error), sporadic (a surprising error)
Slips, lapses and mistakes:
a) execution failure b) planning failure
Generic error modeling system:
a) skill-based: precede problem detection and are associated with monitoring failure (skill failure)
b) rule based: wrong rule used
c) knowledge based: problem solving failures
Errors of omission: failure to do something
Errors of commission: Failure to perform an act correctly
Relate to tasks involving human-machine interaction:
a) mode errors: action performed in an inappropriate mode
b) misperception errors: perceptual cues are missed
c) capture errors: performed in wrong situations
d) sequence errors: actions out of right order
e) timing error: wrong timing
Managing human error
Personell approach: select and train those workers suited to operations of machines and equipment needed to perform the Job –> education & training needed
Design approach: design equipment, procedures and environment that reduce likelihood of errors or its consequences
- when frequency errors occur –> design approach
