AI Bias & Fairness

AI models reflect biases in their training data.

Well-documented issues:
- Racial bias — Face recognition errors, language assumptions
- Gender bias — Stereotyped associations, pronoun defaults
- Cultural bias — Western/English-centric worldviews
- Historical bias — Past inequities encoded in data

Why it happens:
- Training data reflects historical human biases
- Underrepresented groups have less training data
- Objective functions don't optimize for fairness
- Teams building AI lack diversity

What users should do:
- Be skeptical of AI decisions about people
- Check outputs for stereotyped content
- Report bias when you see it
- Don't automate high-stakes decisions without review

What companies should do: Diverse teams, bias testing, transparency, human oversight.