View Full PMI CPMAI Exam Dumps and Practice Test Dumps.

 

Question 181

Which factor should be considered when evaluating the economic feasibility of an AI project?

1. The programming language used by developers
2. The number of model parameters
3. The expected benefits compared with implementation and operating costs
4. The color scheme of the user interface

Correct Answer: 3

Explanation

Economic feasibility examines whether the expected value of an AI initiative justifies the costs and resources required to implement and operate it. Costs may include software, infrastructure, data acquisition, development, integration, training, support, monitoring, maintenance, and vendor services. Benefits may include cost reduction, increased revenue, improved productivity, reduced errors, or other measurable outcomes. A technically feasible project may still be economically unattractive if its expected benefits are insufficient. Economic analysis should therefore consider both initial and ongoing costs as well as realistic benefits. This helps decision-makers determine whether an AI initiative represents a reasonable investment under the organization’s financial constraints.

Question 182

What is the primary purpose of establishing a data governance framework for an AI project?

1. To define how data is managed, protected, accessed, and used
2. To automatically select the best machine-learning algorithm
3. To eliminate the need for data quality checks
4. To guarantee that all AI predictions are correct

Correct Answer: 1

Explanation

A data governance framework establishes policies, roles, responsibilities, standards, and controls for managing data throughout its lifecycle. In AI projects, governance can address data ownership, access, quality, privacy, security, retention, classification, lineage, and appropriate use. Strong governance helps ensure that project teams use trustworthy data and handle sensitive information appropriately. It does not automatically select an algorithm or guarantee model accuracy. Data quality checks remain necessary because governance defines expectations and controls rather than correcting every data problem automatically. Effective data governance provides a foundation for responsible AI development by ensuring that data is managed consistently and according to organizational requirements.

Question 183

Which situation represents a model security risk rather than a typical data-quality issue?

1. A dataset contains missing values
2. Several records use inconsistent date formats
3. A model artifact is accessed by an unauthorized user
4. A dataset contains duplicate customer records

Correct Answer: 3

Explanation

Unauthorized access to a model artifact is a security risk because it involves a person or system accessing protected resources without appropriate authorization. Model artifacts may contain valuable intellectual property or information about how an AI system operates. Appropriate controls can include authentication, authorization, encryption, access logging, and least-privilege permissions. Missing values, inconsistent formats, and duplicate records are primarily data-quality issues that may require cleansing or preprocessing. Distinguishing security risks from data-quality problems helps project teams assign appropriate controls and responsibilities. AI security should cover not only training data but also model artifacts, deployment infrastructure, APIs, credentials, and operational environments.

Question 184

Which practice is most useful for determining whether an AI project remains aligned with its original business objectives?

1. Reviewing only the model’s source code
2. Comparing project outcomes and performance against approved business objectives and success criteria
3. Increasing the number of project meetings
4. Replacing the original requirements with technical preferences

Correct Answer: 2

Explanation

Alignment should be evaluated by comparing actual project outcomes with the business objectives and measurable success criteria established during planning. This can include reviewing business KPIs, expected benefits, operational improvements, user adoption, cost outcomes, and relevant model performance. Technical quality alone does not prove that the project is solving the intended business problem. Regular reviews can identify changes in organizational priorities or gaps between expected and actual benefits. Increasing meetings without meaningful evaluation does not establish alignment. Maintaining traceability between business objectives, requirements, deliverables, and outcomes helps ensure that technical development remains connected to the reason the AI initiative was undertaken.

Question 185

Which technique can help an AI team identify unusual observations that may require investigation before model training?

1. Outlier detection
2. User authentication
3. Model deployment
4. Access logging

Correct Answer: 1

Explanation

Outlier detection identifies observations that differ substantially from expected patterns within a dataset. Outliers may represent legitimate rare cases, measurement errors, data-entry problems, fraud, unusual events, or other conditions. They should not automatically be removed because unusual observations may contain valuable information. The project team should investigate their source and determine how they should be handled based on the intended use case. Authentication and access logging are security controls, while deployment is an operational activity. Identifying unusual observations during data analysis can improve understanding of the dataset and help prevent inappropriate assumptions from affecting model development and evaluation.

Question 186

Which approach is appropriate when an AI model must operate with highly sensitive personal information?

1. Collect as much personal information as possible
2. Apply privacy-by-design principles, minimize unnecessary data, and use appropriate protections
3. Share the information with all project members
4. Disable access controls during development

Correct Answer: 2

Explanation

Privacy-by-design means incorporating privacy considerations into the AI system from the beginning rather than attempting to address them only after implementation. For sensitive personal information, teams should determine what data is genuinely necessary, minimize unnecessary collection, apply suitable access controls, protect data through appropriate security mechanisms, and establish retention and deletion practices. Depending on the use case, de-identification or pseudonymization may also be appropriate. Broadly sharing sensitive information increases exposure and risk. Privacy requirements should be considered throughout acquisition, preparation, training, deployment, monitoring, and retirement. This approach supports responsible data use while reducing unnecessary privacy risks.

Question 187

What is the primary purpose of hyperparameter tuning during model development?

1. To change the business objective
2. To modify project governance policies
3. To find suitable model configuration settings that improve performance on validation data
4. To remove the need for test data

Correct Answer: 3

Explanation

Hyperparameter tuning involves evaluating different configuration settings that control how a machine-learning algorithm learns. Examples may include learning rate, tree depth, regularization strength, number of estimators, or other algorithm-specific settings. The goal is to identify configurations that provide suitable performance while avoiding excessive complexity or overfitting. Validation data or an appropriate cross-validation strategy is commonly used during tuning, while a separate test set may be reserved for final evaluation. Hyperparameter tuning does not change the business objective or governance policies. It is a technical optimization activity that should remain connected to the project’s broader performance, risk, and operational requirements.

Question 188

Why is model calibration useful for certain classification applications?

1. It helps ensure predicted probabilities better reflect observed outcome frequencies
2. It guarantees perfect classification accuracy
3. It eliminates false positives completely
4. It removes the need for threshold selection

Correct Answer: 1

Explanation

Model calibration concerns how well predicted probabilities correspond to actual outcome frequencies. For example, among cases assigned a predicted probability of approximately 0.7, a well-calibrated model would ideally produce the relevant outcome around 70 percent of the time over a sufficiently large and appropriate sample. Calibration can be important when organizations use predicted probabilities to support decisions, prioritize cases, or allocate resources. It does not guarantee perfect classification accuracy or eliminate false positives. Threshold selection may still be necessary because converting probabilities into classes requires a decision rule. Calibration should therefore be considered alongside other evaluation measures appropriate to the use case.

Question 189

Which deployment strategy runs a new AI model alongside the existing system without immediately using its outputs for production decisions?

1. Canary deployment
2. Blue-green deployment
3. Shadow deployment
4. Full replacement deployment

Correct Answer: 3

Explanation

Shadow deployment allows a new model to process production-like or actual production inputs while the existing model continues to make the operational decisions. The new model’s outputs can then be compared with the current system without exposing users or business processes to the new model’s decisions immediately. This provides an opportunity to evaluate behavior under realistic conditions, including performance, latency, and differences in predictions. Canary deployment instead exposes a limited portion of traffic to the new version, while blue-green deployment generally maintains two environments and switches traffic between them. Shadow deployment is therefore useful when teams want realistic evaluation with limited decision-making risk.

Question 190

Which metric is commonly used to measure the proportion of predicted positive cases that are actually positive?

1. Recall
2. Precision
3. Accuracy
4. Mean absolute error

Correct Answer: 2

Explanation

Precision measures the proportion of predicted positive cases that are actually positive. It is calculated as true positives divided by the sum of true positives and false positives. Precision is particularly relevant when false-positive predictions have meaningful costs, such as unnecessary investigations, customer alerts, or manual reviews. Recall measures the proportion of actual positive cases that are successfully identified, so it addresses a different aspect of classification performance. Accuracy considers overall correct predictions, while mean absolute error is primarily used for regression. Selecting the appropriate metric depends on the consequences of different prediction errors and the specific objectives of the AI application.

Question 191

Which activity is most appropriate for identifying the root cause of an unexpected AI system failure?

1. Immediately replacing the entire system
2. Ignoring the incident if the model previously performed well
3. Conducting structured incident investigation and root cause analysis
4. Removing all monitoring records

Correct Answer: 3

Explanation

Root cause analysis helps determine why an incident or failure occurred rather than simply addressing its immediate symptoms. For an AI system, the investigation may examine data pipelines, model versions, infrastructure, dependencies, configuration changes, access events, deployment history, monitoring results, and user actions. Establishing the root cause can help the team determine appropriate corrective and preventive actions. Immediately replacing the entire system may be unnecessary if the underlying problem can be identified and resolved. Removing monitoring records would also make investigation more difficult. A structured incident process supports accountability, learning, system improvement, and prevention of similar failures in the future.

Question 192

What is a key benefit of using automated testing in an AI deployment pipeline?

1. It guarantees that no production issue will ever occur
2. It can consistently verify defined technical conditions before release
3. It eliminates the need for human review in all cases
4. It guarantees that model predictions are unbiased

Correct Answer: 2

Explanation

Automated testing can consistently verify predefined conditions before an AI model or software change is released. Tests may cover data validation, model functionality, API behavior, performance thresholds, security checks, integration behavior, or other technical requirements. Automation improves consistency and can identify certain problems earlier in the deployment process. However, automated tests cannot guarantee that no production issue will occur or that a model is unbiased. Some decisions still require human judgment, domain expertise, governance review, and real-world monitoring. Automated testing is therefore an important component of a controlled AI delivery process rather than a complete replacement for validation and oversight.

Question 193

Which factor should be considered when determining whether an AI model is appropriate for real-time inference?

1. Required latency and available computational resources
2. The number of pages in the project documentation
3. The number of stakeholders in the organization
4. The age of the project manager

Correct Answer: 1

Explanation

Real-time inference requires predictions to be generated within an acceptable response time. Therefore, teams should evaluate model latency, computational requirements, infrastructure capacity, network overhead, concurrency, and expected workload. A highly sophisticated model may provide strong predictive performance but still be unsuitable if it cannot meet the required response time or resource constraints. Real-time requirements should be established before final solution selection and verified through realistic performance testing. Documentation volume and stakeholder count do not directly determine inference feasibility. Balancing predictive performance with latency, scalability, reliability, and infrastructure costs helps ensure that the selected model can function effectively in its intended operational environment.

Question 194

Which practice can help reduce the risk of unauthorized changes to production AI models?

1. Allowing all users write access
2. Sharing administrator credentials
3. Using controlled deployment permissions and approval workflows
4. Removing model version identifiers

Correct Answer: 3

Explanation

Controlled deployment permissions and approval workflows can reduce the likelihood of unauthorized or accidental changes to production AI models. Access should be granted according to role and least-privilege principles, while production releases can require appropriate testing, review, and authorization. Version identifiers and deployment records should also be maintained so that changes remain traceable. Sharing administrator credentials or giving every user write access increases security and operational risk. Removing version information makes it harder to determine which model is active and complicates incident investigation. Production model protection should therefore combine access controls, authentication, approvals, version management, monitoring, and audit logging.

Question 195

What is the main purpose of conducting user acceptance testing for an AI solution?

1. To verify that the solution satisfies user and business requirements in its intended context
2. To determine the programming language used to build the model
3. To guarantee that the model will never drift
4. To replace security testing

Correct Answer: 1

Explanation

User acceptance testing evaluates whether an AI solution meets practical user and business requirements in the environment where it is intended to operate. Users can assess workflows, usability, outputs, integration, expected functionality, and whether the system supports the intended business process. Technical model validation remains important, but strong technical metrics alone do not guarantee that users can effectively use the solution. User acceptance testing does not prevent model drift or replace security testing. Instead, it provides evidence that the delivered system is suitable from an operational and user perspective and can reveal issues that may not have been identified through purely technical testing.

Question 196

Which characteristic of an AI system refers to the ability to continue functioning appropriately when exposed to expected variations or disruptions?

1. Robustness
2. Compression
3. Labeling
4. Duplication

Correct Answer: 1

Explanation

Robustness describes an AI system’s ability to maintain appropriate behavior when exposed to expected variations, noise, changes, or disruptions within the conditions relevant to its intended use. A robust system should not fail unexpectedly because of minor variations that it should reasonably be able to handle. Robustness can be evaluated through testing with varied inputs, environmental conditions, edge cases, or other relevant scenarios. It does not mean that the system will perform correctly under every possible condition. Defining the expected operating range is therefore important. Robustness contributes to reliability and helps organizations understand how an AI solution behaves outside ideal development conditions.

Question 197

Which approach is most appropriate for handling a model requirement that conflicts with a regulatory or organizational policy?

1. Ignore the policy because the model provides business value
2. Document the conflict and escalate it through the appropriate governance process
3. Hide the requirement from stakeholders
4. Deploy the model first and address the issue later

Correct Answer: 2

Explanation

A conflict between an AI model requirement and an applicable regulatory or organizational policy should be documented and addressed through the established governance process. Relevant stakeholders can assess the requirement, determine the nature of the conflict, evaluate alternatives, and decide whether the requirement should be modified or the use case reconsidered. Ignoring or hiding the conflict creates unnecessary legal, compliance, operational, and reputational risks. Deploying first and resolving the issue later can also expose the organization to avoidable problems. Governance processes provide a structured mechanism for resolving such conflicts while maintaining traceability and ensuring that important requirements receive appropriate review.

Question 198

Which activity helps determine whether an AI solution continues to provide value after deployment?

1. Comparing ongoing business outcomes with defined benefit and performance measures
2. Increasing model complexity every month
3. Removing user feedback channels
4. Reviewing only the original training dataset

Correct Answer: 1

Explanation

Post-deployment value should be evaluated by comparing actual business outcomes with the benefits and performance measures established during project planning. These measures might include cost savings, productivity improvements, service quality, revenue impact, processing time, customer outcomes, or other relevant indicators. Reviewing only the original training dataset does not reveal whether the deployed solution is delivering business value in real operations. Similarly, increasing model complexity without evidence may add cost without improving outcomes. Ongoing measurement helps organizations determine whether the AI system remains useful and whether changes to the model, workflow, data, or operational process are needed to maintain or improve value.

Question 199

What is the primary purpose of maintaining a model change log?

1. To record significant model-related changes for traceability and review
2. To prevent all future changes
3. To replace model testing
4. To eliminate the need for access controls

Correct Answer: 1

Explanation

A model change log records significant changes made to an AI model or its associated configuration. Entries may include the model version, date, nature of the change, reason for the change, responsible person or team, approval information, and relevant validation results. This record supports traceability, auditing, troubleshooting, reproducibility, and lifecycle governance. A change log does not prevent future changes and cannot replace technical testing or security controls. Instead, it provides historical visibility into how the system evolved. Maintaining accurate change records is particularly important for production AI systems where changes can affect performance, risk, compliance, or business outcomes.

Question 200

Which activity should occur when formally closing an AI project?

1. Delete all project records immediately
2. Continue development indefinitely
3. Confirm deliverables, document lessons learned, transfer ownership, and close outstanding project activities
4. Disable all stakeholders’ access without review

Correct Answer: 3

Explanation

Formal project closure confirms that agreed deliverables have been completed or appropriately transitioned, outstanding activities have been addressed, and operational ownership has been transferred where necessary. The team should document lessons learned, finalize relevant documentation, close contracts or resources where applicable, archive required records, and communicate the project’s completion. Lessons learned can improve future AI initiatives by capturing what worked well and what should be changed. Simply deleting records or continuing development indefinitely does not constitute proper closure. A structured closure process ensures that technical, operational, administrative, governance, and knowledge-transfer activities are completed before the project is formally concluded.