Why AI can be unfair
If AI learns from one-sided examples, it can treat some people unfairly. Learn how this happens and why speaking up matters.
In 60 seconds
Why AI can be unfair
If AI learns from one-sided examples, it can treat some people unfairly. Learn how this happens and why speaking up matters.
How it happens
| If the examples were mostly… | The AI might… |
|---|---|
| Photos of one kind of face or skin colour | Recognise other faces less well |
| Voices with city accents | Misunderstand people from villages or with other accents |
| Text in English | Work less well in Indian languages |
| Pictures of men as doctors and women as nurses | Think only men are doctors |
Why it matters
- Examples from only one group
- Testing only on people like the builders
- Ignoring complaints
- Examples from many kinds of people, places and languages
- Testing with different groups
- Listening and fixing problems
You can speak up
- If an app doesn't understand your accent or language, it's not your fault. It's a gap in the app.
- Use the app's feedback or report button to say what went wrong.
- Notice AI pictures or answers that repeat unfair ideas, and talk about them.
Watch and read more
Lab
A "fair or unfair?" check of an example set you design on paper.
The problem
Pen and paper.
- Imagine you are building an AI that recognises "a cricketer" from photos.
- List 10 photos you would collect. Describe each one (who, where, what kind of cricket).
- Check your list: are there men and women? Different ages? Different places, like a city ground and a village field? Players with disabilities?
- Rewrite your list so it's fairer, and write 2 sentences on what you changed and why.
You are done when
Hard questions
Try to answer before you reveal. If you can answer these, you understood the lesson.
Q1If the real world already has unfair patterns, why is it a problem for AI to copy them?Reveal
Because AI can repeat those unfair patterns quickly, for many people, and make them look like computer-checked "facts". Instead of helping change unfair ideas, it can make them stronger. Builders need to notice and correct this.
Q2Why might the builders of an app not notice it's unfair?Reveal
If they test mostly on people like themselves, everything seems to work fine for them. Problems only show up for people who were not included. That's why testing with many different people, and listening to feedback, matters.
Questions people ask
Is the AI being mean on purpose?
No. It has no feelings or intentions. It copies patterns from its examples. The people who build and use it are responsible for checking it's fair.
Can bias be fixed?
It can often be reduced, by using more varied examples, testing with many groups and fixing problems. It's hard to remove completely, so people keep checking.
What can a kid do about it?
A lot! Notice unfair results, talk about them, give feedback in apps, and maybe one day build fairer AI yourself.
Lesson test
5 questions. Get 3 right (60%) to pass and complete this lesson.