What is Algorithmic Bias

Definition

Algorithmic bias is systematic, unfair discrimination in the outputs of an AI system, arising when training data, design choices, or objectives cause the model to treat certain groups inequitably, producing skewed or harmful decisions.

« Back to Glossary Index

Identifying bias enables fairer, more equitable AI decisions
Awareness drives diverse data and fairness testing in development
Detecting it reduces legal, ethical, and reputational risk
Measuring bias supports compliance with anti-discrimination requirements

Real World Example

A hiring model trained on historical decisions is found to score one demographic lower for the same qualifications; recognising this algorithmic bias, the team rebalances the data and adds fairness checks before redeploying.

FAQs

What causes algorithmic bias?

Biased or unrepresentative training data, flawed design choices, and objectives that ignore fairness can all introduce systematic bias.

How is algorithmic bias detected?

By testing model outcomes across demographic groups and measuring disparities using fairness metrics.

How can bias be mitigated?

Through more representative data, fairness-aware modelling, bias testing, and ongoing monitoring of outcomes after deployment.