What is Bias?
Bias in AI = Systematic errors that unfairly disadvantage certain groups.
Types of Bias
Representation Bias: Training data lacks Ethiopian faces, names, contexts
Historical Bias: Past data reflects old inequalities
Measurement Bias: What you measure doesn't capture reality
Algorithmic Bias: The math itself creates unfair outcomes
Language Bias: AI performs worse on non-English languages
Case Study: Facial Recognition
Accuracy by skin tone in major systems:
- Light-skinned males: 99.2%
- Light-skinned females: 98.3%
- Dark-skinned males: 88.0%
- Dark-skinned females: 79.2%
Why? Training data was 80%+ light-skinned faces.
What Can YOU Do?
- Test AI on diverse inputs
- Report biased results
- Include diverse data in your projects
- Demand transparency from AI companies