Bias testing that measures whether model outcomes differ: Among the planned testing activities, which type

Bias testing measures whether outcomes differ systematically across protected groups, which is the fairness evidence required.

The question

Preparing for AIGP? Take the free 5-min readiness quiz →

1. Performance testing that measures the model's overall accuracy, throughput and response time against clearly defined engineering service targets.

   Plausible but wrong: aggregate performance can look strong while masking disparities across groups, so it does not answer the fairness question.
2. Bias testing that measures whether model outcomes differ systematically across protected groups, alongside interpretability of those results. ✓

   Correct: bias testing directly measures disparate outcomes across groups, producing the evidence governance requires here.
3. Integration testing that confirms the model exchanges data correctly with the surrounding hiring application and downstream systems.

   Plausible but wrong: integration testing checks that components work together, not whether outcomes are equitable across groups.
4. Security testing that probes the model and its pipeline for vulnerabilities that an attacker could exploit to manipulate its behavior.

   Plausible but wrong: security testing addresses adversarial threats, not systematic disadvantage to protected groups.

How to remember it

Bias testing measures whether outcomes differ systematically across protected groups, which is the fairness evidence required.

More Understanding How to Govern AI Development questions

• Establishing data lineage and provenance records: Which data governance practice most directly provides this →
• Investigate the underrepresentation: Following good practice for managing issues and risks during training and →
• Recording datasets: Which practice best satisfies all three needs at once? →
• All 426 Understanding How to Govern AI Development questions →