Algorithmic bias is when a computer-based decision-making process produces systematically unfair outcomes for certain groups of people. The bias can enter at any step, data collection, feature selection, model training, or deployment. When an algorithm consistently favors one demographic over another, it reflects hidden assumptions, historical inequities, or technical shortcuts in the pipeline. This matters because many high-stakes decisions now run on automated scores. Loan approvals, medical diagnoses, job shortlists, parole recommendations. When bias enters those scores, it denies credit to qualified borrowers, misclassifies patients, excludes capable candidates, or extends prison sentences unjustly. The damage compounds. Biased algorithms reinforce systemic inequality and erode public trust. Fixing it requires better data practices, transparent model design, and regular audits across diverse subpopulations. Developers must ask whether their training data reflects the real world and whether their metrics capture fairness alongside accuracy. Policymakers are beginning to require bias assessments before deployment.
Back to Glossary