COMPEL Glossary / algorithmic-bias
Algorithmic Bias
Algorithmic bias is systematic and unfair discrimination in AI system outputs, often arising from biased training data, flawed model design, unrepresentative data samples, or proxy variables that encode protected characteristics.
What this means in practice
A hiring algorithm trained on a decade of recruitment data may learn to penalize resumes from women's colleges. A credit scoring model may deny loans disproportionately in certain geographic areas, recreating historical redlining patterns. Algorithmic bias is not merely a technical problem -- it is an organizational, ethical, and legal concern with regulatory consequences under the EU AI Act and sector-specific regulations. The COMPEL framework addresses bias through multiple mechanisms: training data auditing during the Model stage, disparate impact testing during Evaluate, ongoing monitoring in production, and the ethical review processes embedded in governance artifacts.
Why it matters
Algorithmic bias can systematically discriminate against vulnerable populations at scale and speed, recreating historical prejudices in automated systems that appear objective. The consequences include regulatory penalties, legal liability, reputational damage, and genuine harm to individuals denied fair treatment. Addressing bias requires understanding that it is not merely a technical deficiency but an organizational, ethical, and legal challenge demanding cross-functional governance.
How COMPEL uses it
COMPEL addresses algorithmic bias through multiple stages and pillars. During Model, training data auditing processes and fairness metrics are designed within the Governance pillar. During Produce, bias testing is built into the deployment pipeline. The Evaluate stage conducts disparate impact testing and ongoing monitoring of production decisions across demographic groups. The Learn stage analyzes bias patterns to improve data collection, model design, and governance controls in subsequent cycles.
Related Terms
Other glossary terms mentioned in this entry's definition and context.