Deep dive · module 83 of 100
Recognising bias and fairness
Recognise when AI output may be biased or unfair.
After this lesson
You can recognise signs of bias and know how to watch out for them.
In view · Side by side
The same bar for different cases
View results for different groups and perspectives.
Watch for stereotypes, missing voices, and differences in treatment.
Involve different people and adjust your approach to skewed patterns.
Part 1 of 3 · Vary examples
Your preference for less movement is on. Choose each part yourself.
What is bias in AI?
The model systematically favours certain groups, perspectives, or outcomes.
Signs
Stereotypical descriptions, missing perspectives, or unequal treatment in examples.
What to do?
Ask for multiple perspectives, check representativeness, and involve people from different backgrounds.
New practice example
From explanation to a usable result.
Assess fairness with relevant comparisons and attention to who suffers a disadvantage.
Fictional practice situationA fictitious language experiment uses two identical customer queries with only a different first name. One receives a warm explanation, the other a short and curt answer.
View a possible solution
This is how it can be done
Keep task, source, and criteria identical. Repeat with more fictitious names and compare tone, completeness, and offered assistance. Document the pattern and have someone assess the potential impact; do not explain that the system is fair based on only two answers.
Why this works
A single example does not demonstrate a general effect or cause. A fixed yardstick makes differences discussable.
Make it slightly more difficult
Write three criteria for an equivalent customer response and determine which other variables you must keep constant in the test.
Extra practice, at your own pace. You do not need to submit anything. This does not count towards an automatic assessment.
Your workplace · learning by doing
Make something. See what changes.
Checking the AI connection… You can already view the written example.
Present your assignment to Ende.
Read the answer, change one thing, and compare.
Previously held
New version
You remain in control
What do you want to improve in your next attempt?
On this device only. AI answers remain only in this page memory. This is self-review, not a grade.
Why this approach works
A single example does not demonstrate a general effect or cause. A fixed yardstick makes differences discussable.
A step further
Write three criteria for an equivalent customer response and determine which other variables you must keep constant in the test.
A prompt to get you started
Give three ways bias can appear in AI-generated texts about people or target groups.
Apply it
A AI creates example characters. One group always receives the leadership role and another group receives supporting roles.
Which approach best fits this situation?
- Assume the pattern is objective because AI produced it.
- Only change the names without looking at the role distribution.
- Investigate the pattern and assess the examples using clear criteria.
With JavaScript, you can check your approach and receive immediate feedback.
Test your insight
What is a practical way to reduce bias?
- Asking for multiple perspectives and having the output checked by different people.
- Do nothing, AI is neutral.
- Always use the first version.
Extra practice
Note one situation in which you would be extra alert to bias in AI output.
When generating example characters or when making recommendations about target groups.
