Dernière mise à jour :
Bias that occurs when an algorithm produces results that are systemically prejudiced due to erroneous assumptions in the machine learning process.
Unfair or disproportionate treatment resulting from algorithmic logic - amplified by biased inputs, skewed objectives, or flawed feature selection - requiring continuous auditing and correction.
Exemple concret :
A university’s admissions-screening AI overweights “legacy applicant” status, leading to disproportionately high acceptance for alumni children. The admissions office removes that feature, retrains the model, and tracks acceptance rates to confirm fairness across demographics.
