Learning Algorithmic Bias and Fairness

Definition and scope of algorithmic bias

Algorithmic bias refers to systematic and repeatable harmful tendencies in computerized sociotechnical systems that produce “unfair” outcomes—such as privileging one group over another—even when that result may differ from the algorithm’s intended function. Bias can be intentional (e.g., design choices reflecting discriminatory goals) or unintentional, arising from how data is coded, collected, selected, labeled, and used to train or operate models, as well as from design constraints and technical limitations. The scope of algorithmic bias extends beyond purely technical errors: it includes how algorithms are embedded in real-world institutions and can shape decisions in areas like search, hiring, healthcare, criminal justice, advertising, and elections. Bias may also be emergent—appearing when correlations are mistaken for causal signals, when algorithms are used in unanticipated contexts or by different audiences, or when feedback loops cause model outputs to influence future data (e.g., predictive policing or recommender systems that reinforce “filter bubbles”). Because algorithms are often treated as neutral and authoritative, they can displace human responsibility and amplify existing social inequities under a veneer of objectivity.

0/2 modes complete

How algorithms produce outputs from data

0/2 modes complete