2. error and bias

0.0(0)
Studied by 0 people
call kaiCall Kai
learnLearn
examPractice Test
spaced repetitionSpaced Repetition
heart puzzleMatch
flashcardsFlashcards
GameKnowt Play
Card Sorting

1/44

encourage image

There's no tags or description

Looks like no tags are added yet.

45 Terms

1
New cards

error

deviation from an intended goal or target

2
New cards

2 ways to get error

  • noise

  • bias


3
New cards

outcomes are accurate when they are…

consistent and correct

4
New cards

outcomes are noisy when they are…

inconsistent

5
New cards

outcomes are biased when they are…

consistent but incorrect

6
New cards

outcomes are biased and noisy when they…

are consistently incorrect in one direction with some random scatter

7
New cards

3 steps for designing for error

  1. assume errors will be made by humans interacting with it

  2. detection/audit

  3. remedy


8
New cards

what 2 questions does the detection/audit step of designing for error address?

  • how detectable are the errors?

  • who has the resources to detect errors?


9
New cards

what question does the remedy step of designing for error address?

who fixes the detected problem

10
New cards

3 types of human errors

  • slips

  • mistakes

  • value bias


11
New cards

slips are errors that occur due to… (3 things)

  • inattention

  • impulses

  • accidents


12
New cards

3 types of slips

  • capture slip

  • description slip

  • memory-lease slip


13
New cards

capture slip

performing more habitual/familiar activies instead of the intended ones

14
New cards

description slip

performing the right action on the wrong object

15
New cards

memory-lapse slip

unconsciously forgetting to do something or forgetting a step

16
New cards

are slips noisy or biased and why?

noisy, because they are situational errors

17
New cards

the amount of slips increase with… (2 things)

  • stress

  • speed


18
New cards

what is one way of preventing slips with design?

forcing functions

19
New cards

forcing functions

logical/physical constraint that prevents from proceeding until a previous step is completed properly

20
New cards

do humans or machines slip more and why?

humans, because machines are more consistent and have less noise

21
New cards

what do mistakes do?

produce the wrong outcome

22
New cards

how do mistakes produce the wrong outcome?

when the goal or plan is inappropriate

23
New cards

what are 2 ways to prevent mistakes with design?

  • warnings

  • attestations


24
New cards

value bias

preference for an unjust outcome

25
New cards

what are 2 types of value biases?

  • intentional

  • unexamined uninentional


26
New cards

what 2 things cause humans to make mistakes?

  • biased intentions

  • biased cognition


27
New cards

are humans making mistakes biased, noisy, or both?

both biased and noisy

28
New cards

what 2 things cause machines to make mistakes?

  • training

  • inputs


29
New cards

are machines making mistakes biased, noisy, or both?

biased only

30
New cards

are humans or machines more reliable and why?

machines, because they inherit mistakes and bias from humans

31
New cards

correctability

how difficult it is to correct slips or mistakes

32
New cards

is it easier to correct humans or machines and why?

machines, because only one correction is needed for them to update reliably

33
New cards

what is a limitation of correcting machines?

only owners and developers of the software or system are able to correct it

34
New cards

detectability

how easy and through what means errors can be identified

35
New cards

what does system bias require to be detected?

a statistical audit

36
New cards

how is a statistical audit performed?

by gathering a large number of cases and comparing directly to a truthful standard

37
New cards

what are 3 implications of statistical audits?

  • require access to a large number of cases

  • require computing power and analytical skill

  • anecdotal evidence is irrelevant or unconvincing


38
New cards

who does system bias work against and who does it work for? why?

  • works against affected users

  • works for system provider

  • because there is information asymmetry between those who have control over the system and those who don’t


39
New cards

how can human bias be assessed?

via judgments of moral character

40
New cards

moral character

a person’s inner workings that lead them to make consistent judgments and perform consistent behaviors

41
New cards

why can an individual’s character be inferred from their behaviors?

humans are consistent in their virtues and flaws

42
New cards

why do people tend to infer others’ characters from their behaviors?

to maintain trust

43
New cards

whose character are humans not able to make inferences about? (2 things)

  • algorithms

  • AI agents


44
New cards

why can’t you infer the character of algorithms or AI?

their actions provide no information for detectability or on their consistency

45
New cards

what are 2 ways designs can be changed to reduce errors?

  • make products more usable

  • make systems less biased