Bias in AI Systems
If training data is biased, AI will be biased; examples: facial recognition working better for some skin tones, translation assuming gender; where bias comes from and whether we can fix it
What a learner can do afterwards
- Explain what bias in AI means using a real-world example
- Describe how biased training data leads to biased AI results
- Suggest one way to reduce bias in an AI system (use more diverse data, test with different groups)
The lesson
You already know that AI learns from examples, not rules. If those examples leave out certain kinds of people or situations, the AI gets worse at handling them. That gap is called bias.
A face-recognition app was trained mostly on photos of light-skinned faces. It got very good at spotting those faces, but it often failed on people with darker skin, because it had not seen enough examples of them to learn from.
A translation tool turns 'the doctor' into a male word and 'the nurse' into a female word, even when no gender was ever mentioned. It copied that pattern from old text where doctors were usually written as men and nurses as women.
Bias can be reduced. Teams can add more examples from the groups that were missing, and test the AI separately on each group before trusting its answers.
AI copies the patterns in its training data, so any group missing from that data ends up with worse results.
Watch it
Where it sits
Where this leads
Jobs that lean on this skill. Follow one to see everything it is built on.
8 questions wait behind this lesson, each with its answer explained. Every answer feeds the sky: stars light as they are learned, and dim when it is time to come back.