Risk & governance
AI bias
Systematic differences in model behaviour that may create unfair outcomes.
Definition
AI bias can arise from data, labels, objectives, model design, deployment context or feedback loops and affect groups differently.
Why it matters
Banks must test outcomes across relevant populations, especially for credit, fraud, servicing, employment and customer treatment.
Related concepts
- Data provenance
The origin, ownership and transformation history of data.
- Algorithmic fairness
Assessing whether model outcomes treat relevant groups appropriately.
- Human baseline
Performance achieved by an appropriate group of people on the same task.