Domain I · Support Responsible and Trustworthy AI Efforts · 15% of the exam
I.3 Conduct bias checks (e.g., model, data, algorithm)
Task 3 of Support Responsible and Trustworthy AI Efforts
May 9: Low Sept 6: High Study Hall misses: 1 Logged: 0
Enablers (verbatim, ECO)
PMI, ECO p. 5: enablers are "illustrative examples of the work associated with the task" and "not meant to be an exhaustive list but rather offer a few examples to help demonstrate what the task encompasses."
Ours: PMI's stems paraphrase one of these as a symptom; the right option is that enabler in PMI's words; the distractors are its siblings or a neighbouring task's enablers.
- Analyze training data for demographic and representation imbalances
- Perform fairness testing across different population groups
- Implement bias detection metrics and monitoring systems
- Review model outputs for discriminatory patterns
- Apply bias mitigation techniques during model development
The rule
'Trustworthy AI standards' or fairness across groups means bias and fairness testing, not prediction accuracy.
Study Hall misses on this task (1)
- Full-length Q51 · confidence Medium · 75sScenario: Recommender that optimises completion speed; wants assurance of trustworthy-AI standardsDecided by: 'Trustworthy AI standards' means fairness and bias testing across groupsChose: Validating prediction accuracy (technical performance)
Logged misses on this task
Log a missNothing logged yet.
What to read
- Trustworthy AI Framework p. 10, 13 — Bias & Discrimination; Bias Measurement & Mitigation
- Module 02 Lesson 32: What Is Informational Bias?
- Module 03 Lesson 37: Implementing Trustworthy AI in Data Preparation
- Leading and Managing AI Projects p. 62 — Scenario 1: A Biased Hiring Algorithm Is Creating a Legal Liability
- Study guide Lesson 14: ECO Task I.3 — Conduct bias checks (e.g., model, data, algorithm)