Bias

Systematic and unfair errors in AI model outputs that arise from biased training data or modelling assumptions. Model bias in AI refers to systematic, unfair discrepancies in model predictions or outputs that disadvantage certain groups, typically correlated with protected attributes such as race, gender, age, or national origin.

Bias is introduced through unrepresentative training data, proxy variables that correlate with protected characteristics, or objectives that optimise for majority-group performance at the expense of minorities.

Model Bias is a foundational concept in AI Safety. Systematic and unfair errors in AI model outputs that arise from biased training data or modelling assumptions.