How it works
Technology can create unequal outcomes through biased data, design assumptions, limited testing groups, or inaccessible interfaces.
Recommend broader testing, better data review, accessibility features, and ways to monitor outcomes across user groups.
Common error
A system is not neutral simply because it uses code or data.
Worked example
Review a speech interface tested only with one group of speakers.
Data / pseudocode
Test group: 20 speakers with similar accents
Deployment: broad school population
Keyboard alternative: absentThe test coverage may miss users with different speech patterns or access needs. Test with varied users, measure errors across relevant groups, and provide another usable input path. Code does not become neutral merely because its decision is automated.
Exercises
5.3.1
Give an accessibility improvement for a chart that conveys categories only by color.
Show solution
Add readable labels, patterns, or a text/table alternative so the distinction does not depend only on color perception.
5.3.2
Why can an overall average accuracy hide unequal outcomes?
Show solution
A large well-served group can dominate the average while smaller groups experience high error rates. Review disaggregated results with appropriate privacy safeguards.