Question 310
An agency's AI application is technically accurate but inaccessible to people using screen readers. Which next step is most defensible?
- Design accessible interfaces and test assistive-technology workflows.
- Provide clear AI involvement disclosures and meaningful ways to request human review.
- Use transparent data-use notices and lawful, purpose-limited feedback collection.
- Measure group-level error disparities and investigate sources before deployment.
Explanation
Correct Answer
A:Fair access involves interaction barriers as well as statistical model performance. Test accessible AI user experience across affected users and give them intelligible disclosures or correction channels where the impact warrants it.
Incorrect Answers
B:The alternative could reduce risk associated with honest disclosure and contestability, yet the described behavior requires accessible ai user experience. That outcome depends on the capability to design accessible interfaces and test assistive-technology workflows.
C:The proposed measure is aimed at ethical secondary use of interaction data, not accessible ai user experience. The architecture must instead design accessible interfaces and test assistive-technology workflows; a separate change is still needed for accessible ai user experience.
D:The relevant trade-off is not primarily evidence-based demographic fairness assessment. A successful fix needs accessible ai user experience; otherwise accessible ai user experience remains unverified even after this change.
Question 311
A hiring tool infers protected characteristics from names even though the business requirement needs only skills. Which approach should be recommended?
- Report relevant disaggregated harms, mitigations and limitations to decision makers.
- Remove unnecessary sensitive inference and audit for proxy effects.
- Align success measures with user welfare and task completion, not raw session length.
- Use transparent data-use notices and lawful, purpose-limited feedback collection.
Explanation
Correct Answer
B:Data minimization and feature scrutiny reduce opportunities for discriminatory recommendations. Test avoidance of sensitive proxy inference across affected users and give them intelligible disclosures or correction channels where the impact warrants it.
Incorrect Answers
A:Its main effect concerns meaningful fairness reporting. The case explicitly hinges on avoidance of sensitive proxy inference; to resolve it reliably, the implementation has to remove unnecessary sensitive inference and audit for proxy effects, which this choice does not establish.
C:The proposed action prioritizes user-benefiting product incentives. That can coexist with the right solution, but it cannot replace avoidance of sensitive proxy inference: the application must still remove unnecessary sensitive inference and audit for proxy effects.
D:The design concern is real but different: ethical secondary use of interaction data. It does not impose the specific requirement for avoidance of sensitive proxy inference; a suitable solution must remove unnecessary sensitive inference and audit for proxy effects and demonstrate the result.
Question 312
An automated complaints triage system has no mechanism for correcting a misclassified urgent request. What is the best production design response?
- Provide clear AI involvement disclosures and meaningful ways to request human review.
- Use transparent data-use notices and lawful, purpose-limited feedback collection.
- Add an appeal and correction channel with accountable escalation.
- Expand representative testing and collect consented feedback from underserved groups.
Explanation
Correct Answer
C:Users need practical recourse when model errors alter service priority. Test correctable consequential classifications across affected users and give them intelligible disclosures or correction channels where the impact warrants it.
Incorrect Answers
A:It emphasizes honest disclosure and contestability without closing the stated gap in correctable consequential classifications. By contrast, the remedy to add an appeal and correction channel with accountable escalation acts on the condition that makes this scenario fail.
B:That is primarily a technique for ethical secondary use of interaction data. It lacks the explicit measure for correctable consequential classifications that is necessary here, namely a system that will add an appeal and correction channel with accountable escalation.
D:This recommendation targets inclusive language and population coverage at a different point in the workflow. It leaves correctable consequential classifications outside the design contract and does not require the team to add an appeal and correction channel with accountable escalation.
Question 313
A public assistant logs personally identifying transcripts to improve prompts without informing affected users. Which change most directly satisfies this requirement?
- Use transparent data-use notices and lawful, purpose-limited feedback collection.
- Flag uncertainty and provide qualified review for high-impact terms.
- Expand representative testing and collect consented feedback from underserved groups.
- Present confidence limitations, supporting evidence and disputed assumptions clearly.
Explanation
Correct Answer
A:Model improvement objectives do not override consent and privacy expectations. Test ethical secondary use of interaction data across affected users and give them intelligible disclosures or correction channels where the impact warrants it.
Incorrect Answers
B:The control would be justified if transparent limits of machine translation were the limiting factor. In this scenario the decisive target is ethical secondary use of interaction data; success requires an architecture that will use transparent data-use notices and lawful, purpose-limited feedback collection.
C:The option belongs to the broader work on inclusive language and population coverage, but it cannot by itself deliver ethical secondary use of interaction data. A decision on this case needs the team to use transparent data-use notices and lawful, purpose-limited feedback collection and verify that outcome.
D:This shifts attention toward calibrated transparent risk communication when the relevant decision is about ethical secondary use of interaction data. Unless the architecture can use transparent data-use notices and lawful, purpose-limited feedback collection, the failure described in the stem remains unresolved.
Question 314
A user-support service optimizes for session length instead of resolved user needs. Its product-recommendation flow also quietly ranks paid affiliates over genuinely better user matches without telling users. Which TWO responsible-design actions should be taken? Choose TWO.
- Disclose material affiliate incentives and independently check recommendations for conflicts with documented user interests.
- Provide clear AI involvement disclosures and meaningful ways to request human review.
- Flag uncertainty and provide qualified review for high-impact terms.
- Align success measures with user welfare and task completion, not raw session length.
- Report relevant disaggregated harms, mitigations and limitations to decision makers.
Explanation
Correct Answers
A:Undisclosed commercial preferences can distort advice even when the answer is fluent. Clear disclosure and ranking checks make the conflict visible and let user benefit, rather than commission, drive decision quality.
D:Optimizing session duration can reward unhelpful or manipulative exchanges. Resolution quality and user benefit should drive the success measure, while independent affiliate-disclosure and ranking controls address the separate commercial conflict.
Incorrect Answers
B:This addresses honest disclosure and contestability; however, the problem asks for user-benefiting product incentives and transparent control of affiliate-ranking conflicts. Without the safeguard to align success measures with user welfare and task completion, not raw session length, the original failure mode is still possible.
C:That choice would be responsive to transparent limits of machine translation, but the reported defect is different. It leaves user-benefiting product incentives and transparent control of affiliate-ranking conflicts unproved because there is no commitment to disclose material affiliate incentives and independently check recommendations for conflicts with documented user interests.
E:Consider its actual operating scope: meaningful fairness reporting. The requested assurance is user-benefiting product incentives and transparent control of affiliate-ranking conflicts, which calls for an implementation that can disclose material affiliate incentives and independently check recommendations for conflicts with documented user interests, not merely for a related improvement.
Question 315
An automated translator handles government notices but does not signal low-confidence legal terminology. What design decision would best fit these facts?
- Expand representative testing and collect consented feedback from underserved groups.
- Flag uncertainty and provide qualified review for high-impact terms.
- Use transparent data-use notices and lawful, purpose-limited feedback collection.
- Present confidence limitations, supporting evidence and disputed assumptions clearly.
- Report relevant disaggregated harms, mitigations and limitations to decision makers.
Explanation
Correct Answer
B:Responsible use requires acknowledging limitations where linguistic errors can affect rights. Test transparent limits of machine translation across affected users and give them intelligible disclosures or correction channels where the impact warrants it.
Incorrect Answers
A:This addresses inclusive language and population coverage; however, the problem asks for transparent limits of machine translation. Without the safeguard to flag uncertainty and provide qualified review for high-impact terms, the original failure mode is still possible.
C:This would make sense as an additional project activity for ethical secondary use of interaction data. The priority required here is transparent limits of machine translation, so the workflow still needs a control that will flag uncertainty and provide qualified review for high-impact terms.
D:By itself, this improves calibrated transparent risk communication, but the case is not asking the architect to optimize that dimension. The requirement is transparent limits of machine translation, supported by an approach to flag uncertainty and provide qualified review for high-impact terms.
E:The proposed measure is aimed at meaningful fairness reporting, not transparent limits of machine translation. The architecture must instead flag uncertainty and provide qualified review for high-impact terms; a separate change is still needed for transparent limits of machine translation.
Question 316
An audit finds that harmful outputs were unevenly distributed, but the development team publishes only mean accuracy. What should the solution architect prioritize?
- Measure group-level error disparities and investigate sources before deployment.
- Flag uncertainty and provide qualified review for high-impact terms.
- Remove unnecessary sensitive inference and audit for proxy effects.
- Report relevant disaggregated harms, mitigations and limitations to decision makers.
Explanation
Correct Answer
D:Responsible governance depends on visibility into who bears errors, not just averages. Test meaningful fairness reporting across affected users and give them intelligible disclosures or correction channels where the impact warrants it.
Incorrect Answers
A:This shifts attention toward evidence-based demographic fairness assessment when the relevant decision is about meaningful fairness reporting. Unless the architecture can report relevant disaggregated harms, mitigations and limitations to decision makers, the failure described in the stem remains unresolved.
B:For a workflow whose primary weakness was transparent limits of machine translation, that proposal could be reasonable. Here the unmet obligation is meaningful fairness reporting, and that requires the system to report relevant disaggregated harms, mitigations and limitations to decision makers.
C:Although avoidance of sensitive proxy inference can be valuable, it is not the acceptance condition in the question. The missing provision is meaningful fairness reporting, addressed by the control to report relevant disaggregated harms, mitigations and limitations to decision makers.
Question 317
A CEO asks for a 'smart enterprise agent' but has not named users, decisions or integration owners. Which design choice best matches the stated constraint?
- Record required locales, specialist terminology and per-language acceptance metrics.
- Reconcile desired business outcomes with existing tool permissions and identify required approvals.
- Map workflow steps, decision rights and irreversible side effects during requirements discovery.
- Facilitate discovery of user journeys, success criteria, constraints and accountable stakeholders.
Explanation
Correct Answer
D:Architecture decisions cannot be justified until the intended business action and ownership are clear. Capture specific user and outcome requirements as a stakeholder-owned acceptance criterion early enough to influence integration, staffing and risk scope.
Incorrect Answers
A:There is a legitimate use case for measurable localization coverage, but it does not supply the test or enforcement needed for specific user and outcome requirements. The design needs to facilitate discovery of user journeys, success criteria, constraints and accountable stakeholders.
B:Consider its actual operating scope: feasible autonomy within approved capabilities. The requested assurance is specific user and outcome requirements, which calls for an implementation that can facilitate discovery of user journeys, success criteria, constraints and accountable stakeholders, not merely for a related improvement.
C:The alternative could reduce risk associated with clear action ownership and boundaries, yet the described behavior requires specific user and outcome requirements. That outcome depends on the capability to facilitate discovery of user journeys, success criteria, constraints and accountable stakeholders.
Question 318
A project sponsor says the assistant must be 'accurate' while compliance expects near-zero critical omissions. Which architecture adjustment is most appropriate?
- Convert abstract quality claims into measurable metrics, test cases and severity thresholds.
- Record required locales, specialist terminology and per-language acceptance metrics.
- Gather jurisdiction, customer-segment and data-classification requirements separately.
- Facilitate discovery of user journeys, success criteria, constraints and accountable stakeholders.
Explanation
Correct Answer
A:Negotiated acceptance criteria prevent subjective disagreement at launch. Capture testable quality requirements as a stakeholder-owned acceptance criterion early enough to influence integration, staffing and risk scope.
Incorrect Answers
B:Its main effect concerns measurable localization coverage. The case explicitly hinges on testable quality requirements; to resolve it reliably, the implementation has to convert abstract quality claims into measurable metrics, test cases and severity thresholds, which this choice does not establish.
C:The option might improve country-specific regulatory needs; that remains insufficient for testable quality requirements. The case is resolved by testable quality requirements, which depends on a design that will convert abstract quality claims into measurable metrics, test cases and severity thresholds.
D:The proposed action prioritizes specific user and outcome requirements. That can coexist with the right solution, but it cannot replace testable quality requirements: the application must still convert abstract quality claims into measurable metrics, test cases and severity thresholds.
Question 319
A support team wants to automate email replies but cannot distinguish drafting from sending on behalf of employees. Which next step is most defensible?
- Facilitate discovery of user journeys, success criteria, constraints and accountable stakeholders.
- Interview end users and sample actual workflows before finalizing scope.
- Map workflow steps, decision rights and irreversible side effects during requirements discovery.
- Gather jurisdiction, customer-segment and data-classification requirements separately.
Explanation
Correct Answer
C:The authorization model follows the real process, not a generic chatbot feature list. Capture clear action ownership and boundaries as a stakeholder-owned acceptance criterion early enough to influence integration, staffing and risk scope.
Incorrect Answers
A:It is aimed at a different outcome, specific user and outcome requirements. No evidence is given that it provides clear action ownership and boundaries; the needed mechanism is to map workflow steps, decision rights and irreversible side effects during requirements discovery at the appropriate boundary.
B:It emphasizes representative frontline workflow discovery without closing the stated gap in clear action ownership and boundaries. By contrast, the remedy to map workflow steps, decision rights and irreversible side effects during requirements discovery acts on the condition that makes this scenario fail.
D:Improving country-specific regulatory needs does not imply clear action ownership and boundaries. The selected design must map workflow steps, decision rights and irreversible side effects during requirements discovery, whereas this choice omits that direct protection or verification step.
Question 320
A regional retailer plans one assistant for countries with conflicting data retention rules. Which approach should be recommended?
- Gather jurisdiction, customer-segment and data-classification requirements separately.
- Include control owners early in discovery and maintain a decision register.
- Convert abstract quality claims into measurable metrics, test cases and severity thresholds.
- Validate interaction expectations and define separate latency budgets for quick and deep tasks.
Explanation
Correct Answer
A:Local compliance variation may force different storage and deployment paths. Capture country-specific regulatory needs as a stakeholder-owned acceptance criterion early enough to influence integration, staffing and risk scope.
Incorrect Answers
B:The relevant trade-off is not primarily early cross-functional control input. A successful fix needs country-specific regulatory needs; otherwise country-specific regulatory needs remains unverified even after this change.
C:The control would be justified if testable quality requirements were the limiting factor. In this scenario the decisive target is country-specific regulatory needs; success requires an architecture that will gather jurisdiction, customer-segment and data-classification requirements separately.
D:The option belongs to the broader work on realistic task-specific latency expectations, but it cannot by itself deliver country-specific regulatory needs. A decision on this case needs the team to gather jurisdiction, customer-segment and data-classification requirements separately and verify that outcome.