View Full PMI CPMAI Exam Dumps and Practice Test Dumps.

 

Question 141

Which activity should be performed when defining the scope of an AI project?

1. Clearly identify what the project will and will not deliver
2. Select the most complex algorithm available
3. Begin deployment before requirements are finalized
4. Remove all project constraints

Correct Answer: 1

Explanation

Defining project scope establishes the boundaries of the AI initiative by specifying what is included and excluded from the project. Scope may describe the intended use case, major deliverables, affected processes, data requirements, system capabilities, and known exclusions. Clear scope helps prevent uncontrolled expansion and provides a reference for managing stakeholder expectations. Selecting an algorithm or beginning deployment is premature when the project boundaries are not established. Constraints should also be documented rather than removed artificially. A well-defined scope enables the team to estimate resources, schedule work, identify risks, establish acceptance criteria, and maintain alignment with the approved business objectives.

Question 142

What is the main purpose of identifying assumptions during AI project planning?

1. To guarantee that assumptions will become facts
2. To document conditions believed to be true for planning purposes
3. To eliminate all project risks
4. To replace stakeholder requirements

Correct Answer: 2

Explanation

Project assumptions are conditions that the team considers likely to be true when developing plans, even though they may not yet be fully verified. Examples include assumptions about data availability, stakeholder participation, infrastructure capacity, vendor support, or access to subject-matter experts. Documenting assumptions helps teams recognize areas of uncertainty and identify items that may require validation. If an assumption proves incorrect, the project manager can assess its impact and adjust plans accordingly. Assumptions do not eliminate risks or replace requirements. Instead, they provide transparency about the conditions underlying project plans and support more realistic planning and risk management.

Question 143

Which characteristic of data indicates that information is available when it is needed for the intended purpose?

1. Accuracy
2. Timeliness
3. Uniqueness
4. Validity

Correct Answer: 2

Explanation

Timeliness refers to whether data is sufficiently current and available at the time it is required for a particular purpose. Data can be accurate but still have limited usefulness if it is too old for the business decision or AI application. For example, a real-time fraud detection system may require very recent transaction information, while a historical research model may tolerate older data. Timeliness should therefore be evaluated according to the intended use and operational requirements. Accuracy concerns correctness, uniqueness addresses duplication, and validity concerns adherence to defined rules or formats. Assessing timeliness helps determine whether data can support the required AI workflow.

Question 144

Which approach can help improve the quality of labels used to train a supervised learning model?

1. Allowing inconsistent labeling rules
2. Removing all quality checks
3. Defining clear annotation guidelines and reviewing labeling quality
4. Randomly changing labels between reviewers

Correct Answer: 3

Explanation

High-quality labels are essential because supervised learning models rely on labeled examples to learn relationships between inputs and target outcomes. Clear annotation guidelines help different reviewers apply consistent criteria when assigning labels. Teams can also use quality checks, reviewer training, sampling, adjudication procedures, and measures of agreement to identify inconsistencies. Poorly defined labeling rules can introduce noise and reduce model performance. Randomly changing labels or eliminating quality checks makes the training data less reliable. Establishing a structured labeling process helps ensure that labels accurately represent the intended target and provides a stronger foundation for model training and subsequent evaluation.

Question 145

What is the primary purpose of separating training, validation, and test data?

1. To ensure every record is used for training
2. To support model development while providing independent evaluation of generalization
3. To eliminate the need for monitoring
4. To guarantee that the model will be fair

Correct Answer: 2

Explanation

Separating datasets into training, validation, and test portions helps prevent overly optimistic estimates of model performance. Training data is used to learn model parameters, validation data can support model selection and tuning, and a separate test dataset can provide a final evaluation on data that was not used during those development decisions. The exact approach depends on the project and available data. Proper separation does not guarantee fairness or eliminate the need for monitoring after deployment. It instead provides a structured method for evaluating whether a model generalizes beyond the examples used during development and helps reduce the risk of evaluating a model on data it has effectively seen.

Question 146

Which situation is most likely to indicate underfitting?

1. The model performs poorly on both training and validation data
2. The model performs perfectly on training data and well on unseen data
3. The model has too many monitoring dashboards
4. The dataset contains duplicate records

Correct Answer: 1

Explanation

Underfitting occurs when a model is too simple or insufficiently trained to capture important patterns in the data. A common indication is poor performance on both the training dataset and unseen validation data. In contrast, overfitting is generally associated with strong training performance combined with significantly weaker performance on unseen data. Addressing underfitting may involve improving feature representation, selecting a more suitable model, adjusting model complexity, or providing better training data. The appropriate response depends on the underlying cause. Identifying the relationship between training and validation performance helps the project team distinguish underfitting from overfitting during model evaluation.

Question 147

Which practice is most useful for managing experiments involving multiple AI model configurations?

1. Changing parameters without recording them
2. Tracking configurations, results, and relevant experiment metadata
3. Deleting unsuccessful experiments immediately
4. Using a single undocumented configuration

Correct Answer: 2

Explanation

Experiment tracking allows AI teams to systematically record model configurations, parameters, datasets, evaluation metrics, code versions, and other relevant information for each experiment. This makes it easier to compare approaches, reproduce successful results, investigate unexpected outcomes, and understand why one configuration performed differently from another. Without tracking, teams may repeat experiments unnecessarily or lose the ability to reconstruct important results. Deleting unsuccessful experiments can also remove useful information about approaches that were evaluated. A structured experiment-tracking process therefore improves collaboration, reproducibility, decision-making, and governance throughout model development.

Question 148

Why should an AI project establish clear model ownership before production deployment?

1. To ensure responsibility for maintenance, monitoring, and lifecycle decisions is defined
2. To allow only one person to use the model
3. To eliminate all operational risks
4. To prevent the model from being updated

Correct Answer: 1

Explanation

Clearly defined model ownership establishes accountability for important lifecycle activities after deployment. The owner or responsible team may oversee monitoring, maintenance, incident response, performance reviews, documentation, retraining decisions, approvals, and eventual retirement. Without clear ownership, problems may be overlooked because teams are uncertain about who should respond. Model ownership does not mean that only one person can use a system, nor does it eliminate operational risk. It also should not prevent legitimate updates. Instead, ownership ensures that responsibility for managing the model remains visible throughout its operational life and that appropriate decisions can be made when conditions or requirements change.

Question 149

Which approach is most appropriate for evaluating a third-party AI vendor?

1. Select the vendor solely because it has the lowest price
2. Evaluate capabilities, security, privacy, reliability, support, contractual terms, and risks
3. Avoid reviewing the vendor’s documentation
4. Assume the vendor manages every organizational responsibility

Correct Answer: 2

Explanation

Third-party AI vendors should be evaluated using criteria that reflect the organization’s technical, business, security, privacy, operational, and governance requirements. Relevant considerations can include model capabilities, performance evidence, data handling practices, security controls, availability, scalability, support arrangements, service-level commitments, costs, ownership terms, regulatory considerations, and exit options. Price alone does not establish suitability. Organizations should also understand which responsibilities remain with them rather than assuming the vendor manages everything. A structured vendor assessment helps identify dependencies and risks before adoption. Contractual requirements can then establish expectations around service levels, security, data handling, incident notification, and other important responsibilities.

Question 150

What is the primary objective of a proof of concept for an AI initiative?

1. To validate whether a proposed approach is technically and practically viable on a limited scale
2. To immediately replace the production system
3. To guarantee long-term return on investment
4. To eliminate the need for stakeholder approval

Correct Answer: 1

Explanation

A proof of concept, or POC, is a limited effort designed to test whether a proposed AI approach is technically and practically viable. It can help the team investigate data availability, model feasibility, integration challenges, performance expectations, or other critical assumptions before committing to a larger implementation. A POC does not guarantee long-term return on investment or automatically replace a production system. It also does not remove the need for governance and stakeholder approval. By testing important assumptions early, a POC can provide evidence that supports decisions about whether to continue, modify, or stop an AI initiative before substantial resources are invested.

Question 151

Which security threat involves deliberately manipulating training data to negatively affect an AI model?

1. Model monitoring
2. Data poisoning
3. Data normalization
4. Feature scaling

Correct Answer: 2

Explanation

Data poisoning occurs when an attacker intentionally introduces manipulated, incorrect, or malicious data into a dataset used to train an AI model. The objective may be to reduce model accuracy, create targeted incorrect behavior, or otherwise influence the resulting system. Protecting against data poisoning can involve controlling data sources, validating data quality, monitoring unexpected changes, restricting write access, maintaining provenance, and reviewing suspicious records. Data normalization and feature scaling are legitimate preprocessing techniques and are unrelated to this security threat. Understanding AI-specific threats such as data poisoning is important when designing security controls for the data and model development lifecycle.

Question 152

Which principle requires an AI system to have clearly defined responsibility for important decisions and outcomes?

1. Accountability
2. Compression
3. Normalization
4. Dimensionality reduction

Correct Answer: 1

Explanation

Accountability means that responsibility for an AI system’s development, operation, decisions, and impacts is clearly assigned. Organizations should identify appropriate owners and decision-makers and establish processes for oversight, review, incident handling, and corrective action. Accountability is particularly important when AI systems influence consequential decisions because stakeholders need to know who is responsible for governance and response. Technical practices such as normalization or dimensionality reduction do not establish organizational responsibility. Clear accountability also supports transparency because stakeholders can understand who is responsible for ensuring that the system operates according to approved requirements, policies, and controls throughout its lifecycle.

Question 153

Why is explainability important for some AI applications?

1. It guarantees that predictions are always correct
2. It can help stakeholders understand factors contributing to model outputs
3. It eliminates the need for model validation
4. It guarantees compliance with every regulation

Correct Answer: 2

Explanation

Explainability can help stakeholders understand how or why an AI system produced a particular output. This may support user trust, debugging, governance, auditing, risk assessment, and decision review, especially when model outputs influence important decisions. The appropriate level and method of explanation depend on the use case and the model. Explainability does not guarantee prediction accuracy or automatically establish regulatory compliance. It also does not replace validation. Instead, it provides additional insight into model behavior and can help identify unexpected patterns, inappropriate dependencies, or situations in which users need to exercise additional judgment before acting on an AI-generated recommendation.

Question 154

Which approach is most appropriate when an AI model’s performance begins to decline after deployment?

1. Ignore the change if the model previously passed testing
2. Investigate monitoring results and determine whether recalibration, retraining, or another response is appropriate
3. Delete all production data immediately
4. Increase model complexity without investigation

Correct Answer: 2

Explanation

Performance degradation after deployment should trigger investigation rather than an automatic technical response. The team should examine monitoring results, input data changes, concept drift, data quality, infrastructure conditions, user behavior, and other factors that may explain the decline. Depending on the findings, possible responses include recalibration, retraining, adjusting thresholds, correcting data pipelines, changing the model, or rolling back to a previous version. Increasing complexity without understanding the cause may fail to address the actual problem. Historical testing alone is not sufficient because production conditions can change. Continuous monitoring therefore supports timely identification and appropriate management of model degradation.

Question 155

Which metric combines precision and recall into a single measure?

1. F1 score
2. Mean absolute error
3. R-squared
4. Mean squared error

Correct Answer: 1

Explanation

The F1 score combines precision and recall into a single metric by calculating their harmonic mean. It can be useful when both false positives and false negatives matter and a balance between precision and recall is desired. F1 is commonly used for classification problems, particularly when class distributions are uneven and accuracy alone may not provide sufficient information. It should not automatically replace other metrics because the appropriate evaluation depends on the business context and consequences of errors. MAE, MSE, and R-squared are commonly associated with regression evaluation. Teams should select metrics that correspond to the specific objectives and risks of the AI application.

Question 156

What is the main purpose of a stakeholder communication plan for an AI project?

1. To define how relevant information will be communicated to stakeholders throughout the project
2. To prevent stakeholders from asking questions
3. To replace technical documentation
4. To guarantee that all stakeholders will agree on every decision

Correct Answer: 1

Explanation

A stakeholder communication plan defines what information should be communicated, to whom, when, through which channels, and by whom. AI projects often involve stakeholders with different levels of technical knowledge and different interests, including executives, users, data specialists, security teams, legal representatives, and operational personnel. Tailoring communication helps ensure that relevant risks, progress, decisions, changes, and results are understood by the appropriate people. A communication plan does not guarantee agreement or replace technical documentation. Instead, it provides a structured approach for keeping stakeholders informed and engaged, reducing misunderstandings and supporting timely decisions throughout the AI project lifecycle.

Question 157

Which practice best supports protection of sensitive AI training data?

1. Making the dataset publicly accessible
2. Applying appropriate access controls and encryption
3. Sharing credentials among all project members
4. Keeping sensitive information indefinitely without retention rules

Correct Answer: 2

Explanation

Sensitive AI training data should be protected through a combination of technical and organizational controls. Access controls restrict data access to authorized individuals based on legitimate responsibilities, while encryption can protect information during storage and transmission. Additional measures may include data minimization, secure key management, logging, retention policies, monitoring, and appropriate de-identification. Publicly exposing sensitive data or sharing credentials increases security risk. Keeping information indefinitely can also increase exposure if the data is no longer needed. Security controls should be appropriate to the sensitivity of the information and integrated into the data lifecycle rather than added only after model development has begun.

Question 158

Which activity helps determine whether an AI model meets predefined technical performance requirements before production deployment?

1. Model validation
2. Marketing analysis
3. Employee onboarding
4. Contract renewal

Correct Answer: 1

Explanation

Model validation evaluates whether an AI model satisfies predefined technical and, where applicable, business requirements using appropriate evaluation data and metrics. Validation may examine accuracy, precision, recall, F1 score, regression error, latency, robustness, subgroup performance, or other criteria depending on the use case. It helps determine whether the model is ready for the next lifecycle stage and can reveal issues that require further development. Model validation should be based on suitable datasets and predefined acceptance criteria rather than informal impressions. Marketing analysis, employee onboarding, and contract renewal may be relevant to broader project activities but do not establish whether the model meets technical requirements.

Question 159

What is the main benefit of using a structured risk response strategy for AI project risks?

1. It ensures every risk will occur
2. It provides planned actions for reducing, transferring, avoiding, or accepting identified risks
3. It removes the need to monitor risks
4. It guarantees that the project budget will never change

Correct Answer: 2

Explanation

A structured risk response strategy helps the project team decide how to address identified risks based on their likelihood, impact, context, and organizational risk tolerance. Depending on the situation, responses may include avoiding the risk, reducing its probability or impact, transferring responsibility, or accepting it with appropriate monitoring and contingency planning. A response strategy does not guarantee that risks will occur or disappear, and it does not eliminate the need for ongoing monitoring. AI projects may face changing data, security, model, vendor, compliance, and operational risks. Planned responses provide a more consistent framework for managing these uncertainties throughout the project.

Question 160

Which activity is most appropriate when an AI system reaches the end of its useful operational life?

1. Continue operating it indefinitely without review
2. Remove all historical documentation
3. Retire the system through a controlled process that addresses data, access, dependencies, and records
4. Disable monitoring before retirement

Correct Answer: 3

Explanation

AI model retirement should be handled through a controlled lifecycle process. The organization should determine whether the system can be safely decommissioned, identify dependent applications and users, revoke or modify access, preserve required records, address data-retention obligations, and communicate the retirement to relevant stakeholders. In some situations, a replacement model or process may need to be deployed before the old system is removed. Simply disabling monitoring or deleting documentation can create operational and governance problems. Controlled retirement ensures that the AI system is removed safely and that important dependencies, records, security considerations, and organizational requirements are addressed before decommissioning.