Bias in AI systems is one of the most well-documented and impactful ethical challenges. Understanding where bias comes from and how to address it is critical.

Sources of Bias

AI bias doesn't appear from nowhere — it's inherited from data and design choices:

• Training Data Bias — If the data reflects historical inequalities (and it usually does), the model will reproduce them. A hiring model trained on a company's past decisions will perpetuate any existing discrimination.
• Representation Bias — When certain groups are underrepresented in training data, the model performs worse for those groups. Facial recognition trained mostly on lighter-skinned faces fails more often on darker-skinned faces.
• Measurement Bias — When the thing being measured is a poor proxy for the actual goal. Using zip codes as a feature can introduce racial bias because of residential segregation.
• Aggregation Bias — When a model that works well on average fails for specific subgroups.

Types of Fairness

There's no single definition of "fair" — different contexts require different approaches:

• Demographic Parity — Equal outcomes across groups (e.g., same loan approval rate)
• Equal Opportunity — Equal true positive rates across groups
• Individual Fairness — Similar individuals receive similar outcomes
• Counterfactual Fairness — Outcome wouldn't change if a protected attribute were different

Importantly, these definitions can conflict with each other. Achieving one type of fairness may require sacrificing another.

Mitigation Strategies

• Audit training data for representation gaps
• Test model performance across demographic groups
• Use fairness-aware algorithms during training
• Implement human review for high-stakes decisions
• Publish bias reports and model cards documenting known limitations
• Involve diverse teams in model development and evaluation

What You Can Do

• Question AI recommendations, especially for consequential decisions
• Ask vendors about their bias testing and mitigation practices
• Report biased outputs when you encounter them
• Support organizations working on AI fairness and accountability