Support Responsible and Trustworthy AI Efforts
Free PMI Certified Professional in Managing AI practice — 6 questions on Support Responsible and Trustworthy AI Efforts, with explanations. No sign-up. Full 12-question mixed test →
A healthcare organization is planning to deploy an AI system that analyzes patient imaging data to predict disease risk. The system will process large volumes of PII and protected health information. According to responsible AI practices, at which point in the AI project lifecycle should the Privacy Impact Assessment (PIA) be conducted?
Privacy by design requires assessing privacy risk from the earliest business understanding phase, and because AI projects are iterative, the PIA must be revisited whenever data sources, features, or use cases change throughout the lifecycle.
A financial institution builds an AI model to predict loan default risk. During bias testing, the team finds that overall accuracy is nearly identical across two demographic groups, but the false negative rate (approving loans that later default) is significantly higher for Group A, while the false positive rate (denying loans that would not have defaulted) is significantly higher for Group B. Which fairness metric should the team prioritize to address this specific disparity?
Equalized odds directly targets parity in true positive and false positive rates between groups, which is exactly the disparity described in the scenario.
An AI project team is designing access controls for a shared training data repository containing sensitive customer records. Data scientists need to query and transform data, data engineers need to modify pipeline configurations, business analysts need only aggregate reports, and external auditors need read-only access to audit logs but not raw data. Which access control configuration BEST supports least privilege while enabling project execution?
RBAC with permissions scoped to each role's actual needs, plus audit logging, directly implements least privilege and supports accountability across the AI lifecycle.
A team must choose an interpretability approach for a complex deep learning fraud-detection model. Regulators require an explanation for each individual denied transaction, identifying which specific features drove that particular decision, without requiring a full explanation of the model's overall internal logic. Which type of interpretability method BEST satisfies this requirement?
Local interpretability methods like LIME and SHAP are designed specifically to explain individual predictions by attributing influence to specific input features, matching the per-decision explanation requirement.
A multinational company deploys an AI-based credit scoring system used by customers in both the European Union and California. A rejected EU applicant requests details on how the automated decision was made and asks for human review of the decision. Which regulatory requirement specifically compels the company to provide this right?
GDPR Article 22 specifically addresses automated decision-making with legal or similarly significant effects, granting data subjects the right to obtain human intervention, express their view, and receive an explanation of the decision.
During the data preparation phase of an AI project, the team makes multiple transformations to the training dataset, including removing outliers, imputing missing values, and re-labeling records after discovering annotation errors. Before proceeding to the model development go/no-go gate, which documentation practice is MOST critical for maintaining accountability and supporting future audits?
A version-controlled, traceable record of every transformation with rationale and ownership preserves chain of custody and data lineage, which is essential for reproducibility, audits, and accountability at the go/no-go gate.
Ready for the real thing?
The full course: two full-length practice tests, video lessons for every exam domain, hands-on labs and detailed explanations.
$109.99 $34.99 with code FREETEST33 — valid through September 14.
Get my $34.99 deal →