Also known as: AI Bias or Machine Learning Bias or Algorithm Bias Β· Algorithmic bias Β· Bias (Algorithmic Bias) Β· bias (ethics/fairness)
Bias is a systematic tendency in data, models, measurements, design choices, or human processes that can affect an AI system's behavior or outcomes. In technical contexts, bias can refer to model assumptions or error patterns; in governance and fairness contexts, it often concerns unjustified differences in treatment or impact across people or groups. Bias may arise from training data, labels, deployment context, feedback loops, or human decisions and should be identified, measured, and mitigated where it creates risk or unfairness.
Bias allows AI systems to determine how to treat different situations accordingly, and is therefore fundamental to its adaptive capacity when minimised and justified (so as to avoid unfairness).
Bias in AI models typically arises from two sources: the design of models themselves and the training data they use. Models can sometimes reflect the assumptions of the developers coding them, which causes them to favour certain outcomes. Additionally, AI bias can develop due to the data used to train the AI.
AI systems can have bias embedded in them, which can manifest through various pathways including biased training datasets or biased decisions made by humans in the design of algorithms. See PN 708 and PN 633 for further details.