Fairness
The goal that AI systems treat all groups equitably and don't cause systematic discrimination.
Fairness in AI means equitable treatment of all groups. Different definitions (demographic parity, equalized odds) can conflict – no universal "fair".
Explanation
Fairness definitions: Demographic parity (equal rates), equalized odds (equal TPR/FPR), individual fairness (similar treated similarly). Problem: Definitions can conflict – not all achievable simultaneously.
Marketing Relevance
Marketing AI must be fair: Targeting without discrimination, pricing without group disadvantage, recommendations without exclusion.
Example
A credit scoring model is checked for fairness: Do different demographic groups have equal approval rates at the same risk level?
Common Pitfalls
Fairness definitions often conflict. "Fair" differs by stakeholder. Fairness optimization can cost accuracy.
Origin & History
Fairness research in ML exploded after 2016 (ProPublica COMPAS analysis). Google, IBM, and Microsoft released fairness toolkits. EU AI Act mandates bias tests for high-risk AI.
Comparisons & Differences
Fairness vs. Bias
Bias is the problem (distortion); Fairness is the goal (equal treatment). Bias mitigation is the path to fairness.
Fairness vs. Equity
Fairness can mean equal treatment; Equity means adjusted treatment to achieve equal outcomes.