algorithmic bias

Systematic errors in AI systems that create unfair outcomes for certain groups. Often originates from biased training data, feedback loops, or flawed problem framing.

Syntax

ai-ethics

bias_audit: measure model outputs across demographic groups

Example

ai-ethics

// Types of algorithmic bias:

// 1. Historical bias: training data reflects past discrimination
//    E.g., resume screening trained on historically male-dominated field

// 2. Representation bias: underrepresented groups in training data
//    E.g., facial recognition less accurate for darker skin tones

// 3. Measurement bias: proxies that correlate with protected attributes
//    E.g., using zip code as credit proxy (correlates with race)

// Mitigation:
// - Audit training data for representation
// - Test model outputs across demographic groups
// - Use fairness-aware training objectives