Glossary

Algorithmic bias

Systematic, unfair skew in a model's outputs that disadvantages particular groups, usually traced to skewed training data or design choices.

Algorithmic bias is a systematic, unfair skew in a model's outputs that disadvantages particular groups, most often traced back to training data that under- or over-represents certain populations, to proxy variables that correlate with a protected characteristic even when that characteristic isn't used directly, or to design choices, such as which outcome a model is optimized for, that embed a particular set of assumptions.

Bias is distinct from a model simply being inaccurate: a biased model can be highly accurate on average while systematically performing worse, or making worse decisions, for a specific subgroup, which is why aggregate accuracy metrics alone can hide the problem. Detecting it typically requires evaluating model performance and outcomes separately across relevant groups, not just overall, and is closely tied to Explainable AI techniques used to understand why a model reached a given output.

Algorithmic bias matters wherever machine learning informs decisions that affect people, lending, hiring, pricing, content ranking, because a biased model can reproduce or amplify existing societal disparities at scale, and because it can create legal exposure under anti-discrimination law even without intent. It's a central concern within Responsible AI and data ethics practice. A common misreading is assuming that removing a protected attribute like race or gender from the model's inputs is sufficient to eliminate bias; correlated proxy variables can reproduce the same skew even when the attribute itself is absent.

Last reviewed September 22, 2026

In the index now

Related terms

Responsible AI A set of practices for developing and deploying AI systems that are fair, transparent, safe, and accountable to the people they affect. Explainable AI (XAI) Techniques and models that make a machine learning system's predictions understandable to humans, rather than a black box. Training data The dataset with known outcomes that a machine learning model learns patterns from before being used to predict new cases. Data ethics The study and practice of using data responsibly, addressing fairness, consent, transparency, and harm beyond what regulation strictly requires. Machine learning Building models that learn patterns from data to make predictions or decisions, rather than following hand-coded rules.