Bias Mitigation: Identifying and reducing unwanted bias in data, models, and outcomes.

Equity: Ensuring that AI systems provide fair opportunities and outcomes for all, especially marginalized or historically disadvantaged groups.

Transparency in Impact: Making it clear how AI decisions affect different groups and individuals.

Inclusive Design: Involving diverse stakeholders in the design, testing, and deployment of AI systems.

Audit datasets for representativeness and historical bias before training models.

Use fairness-aware algorithms and metrics (e.g., demographic parity, equalized odds).

Test AI systems for disparate impact across demographic groups.

Engage with affected communities to understand real-world impacts and needs.

Document fairness considerations and trade-offs in model cards or datasheets.

Defining fairness can be context-dependent and value-laden.

Trade-offs may exist between different fairness criteria or between fairness and accuracy.

Historical data may reflect systemic inequalities that are difficult to fully correct.