Selection Bias
A bias introduced when the training data is not representative of the target population, leading to models that perform unevenly across different groups or scenarios.
In plain language
When your training data doesn't represent the real world. If you train a medical AI only on data from young adults, it may perform poorly on elderly patients.
Why this matters
Selection bias is a fundamental risk that your AI governance framework must address at the data collection stage. If training data does not represent your actual user population, the AI will not serve all users fairly and may expose your organisation to fairness and discrimination risks.
Relevance
GovernanceGovernance processes must require documented analysis of population representativeness before training any model, with explicit fairness validation across demographic groups before deployment.
Related terms
Putting selection bias into practice in your organisation?
Ready to transform your AI strategy?
Partner with Australia's AI strategy and governance specialists. From adoption roadmaps to ISO 42001 audit readiness.