Free Certified Responsible AI Governance & Ethics (C|RAGE) practice test: 30 questions with answers

Machine learning is a subset of AI that learns patterns from data rather than relying only on explicit rules. The other statements reverse or misstate the relationship.

Transparency requires disclosure of AI use and the basis of decisions affecting people. Efficiency, scalability and portability are engineering qualities rather than ethical disclosure duties.

Problem definition sets purpose, stakeholders and acceptable use before any technical work. Training, monitoring and retirement happen later in the lifecycle.

Fairness addresses unjust disadvantage to groups, which is the concern described. Accuracy, availability and latency are performance qualities that do not by themselves ensure fair outcomes.

A model card documents intended use, limitations, training data and evaluation results to support appropriate use. Purchase orders, network diagrams and bug entries serve other purposes.

Human oversight keeps people able to understand, challenge and override AI decisions. Automation bias is the risk that oversight is missing, drift is a performance change, and augmentation is a data technique.

An AI impact assessment considers affected people, potential harms and mitigations before deployment. Server counts, colour palettes and feature counts do not assess harm.

Comparing error rates and outcomes across groups with agreed fairness measures shows differential impact. Overall averages can hide group disparities, and login counts and speed do not measure fairness.

Documenting sources and checking representativeness reduces the risk of bias from the data. Easy-access data, removing documentation and deleting minority data each increase or hide bias.

Explainability makes the basis of a decision understandable to affected people and reviewers. Obfuscation hides logic, encryption protects data, and replication copies models.

Clear roles and decision rights for the full AI lifecycle make governance effective. A single decision-maker, informal agreements and constantly changing structures each leave gaps.

An approval record documents the decision, the risks accepted, the conditions and the accountable approver, which makes governance auditable. The other artefacts are unrelated to approval.

Aligning AI objectives with business goals and reviewing them through governance keeps strategy and oversight connected. Vendor-led choices, static strategies and uncoordinated teams weaken the link.

A change impact assessment re-evaluates risk, fairness and compliance for significant changes. Announcements alone, skipped reviews and removed monitoring each leave the change unassessed.

The share of systems with completed risk assessments and reviews shows governance coverage and effectiveness. Conference counts, word counts and chat activity do not measure governance.

A named business owner, supported by risk, legal and technical staff and reporting to governance, provides clear accountability. The system, no one and users alone cannot hold that accountability.

An AI-specific incident response plan with agreed notification, remediation and review steps allows a prompt and accountable response. Denial, renaming and passing responsibility to customers each undermine accountability.

Policies, roles and escalation, and monitoring and reporting are core governance framework elements. Furniture and drink menus are unrelated to governance.

Design review before build and decommissioning at end of life are governance lifecycle stages. Office moves, canteen menus and birthday events are not AI lifecycle stages.

An AI governance committee with risk, legal, ethics and business representation reviews high-impact proposals. The other options lack independence, accountability or relevance.

The EU AI Act uses a risk-based classification with obligations that depend on the risk tier. The other regulations and standards govern different subjects.

Data protection law such as GDPR gives individuals protection against solely automated decisions with significant effects, subject to exceptions and safeguards. The other rights do not exist in that form.

Due diligence, contractual terms, audit rights and compliance evidence allow the organisation to show its own compliance. Marketing claims, assumed transfer of responsibility and price-first buying each leave the organisation exposed.

Copyright and other intellectual property rights govern whether training on material is permitted. Tax, employment and building rules are unrelated to that question.

A compliance evidence pack maps each requirement to controls and test results, which is what a regulator examines. Brochures, posts and personal blogs do not provide verifiable evidence.

Clear data use, ownership and deletion terms, and audit rights with compliance evidence, protect the buyer. Open data use, removed responsibility and unwritten promises each weaken protection.

Informing customers, giving meaningful information and an appeal route meets transparency expectations. Hiding AI use, misleading descriptions and revealing personal data each break transparency or privacy.

Record-keeping and technical documentation obligations for high-risk systems allow compliance to be demonstrated. Tax, safety inspections and parking permits are unrelated.

Reproducing protected works without permission may infringe copyright. Trademark registration, insurance claims and payroll taxes do not arise from that reproduction.

Independent audits, certifications and contractual commitments provide verifiable evidence. Advertisements, verbal assurances and testimonials are not verification.