Verify Your Certificate
info@pmexpertgroup.com
Cart
0
0.00 INR

Leading and Managing AI- Sample Paper 1

25/02/2026

Leading and Managing AI

This sample paper explores the core principles of leading and managing artificial intelligence initiatives in modern organizations. It introduces key concepts such as AI strategy, governance, ethics, and change management, helping leaders understand how to align AI projects with business goals. The content emphasizes cross‑functional collaboration between technical and non‑technical teams, risk management, and responsible data use. It is suitable for students, managers, and professionals who need a structured overview of how to plan, implement, and oversee AI solutions in a practical, results‑oriented way.

Within this sample, you will find example frameworks for AI project lifecycles, guidance on stakeholder engagement, and discussion prompts that encourage critical thinking about AI’s impact on people and processes. It also highlights leadership skills required to foster innovation while maintaining transparency and trust, including communication, ethical decision‑making, and continuous learning. Use this paper as a reference for assignments, workshops, or internal training sessions focused on AI leadership, and adapt the ideas to your specific industry, organizational culture, and maturity level in digital transformation.

CPMAI Full Simulator

CPMAI Full Practice Simulator

Q1. 1. What is the primary "Paradox" of AI projects described in the guide?

A. AI talent is increasing while computing costs are rising.
B. While technology races ahead, most AI projects still stall or fail.
C. AI is designed to replace humans but requires more humans to manage it.
D. Data is becoming more abundant but less useful for machine learning.
Answer: 2
Explanation: The guide identifies that the failure of AI projects is rarely due to the technology or talent, but rather the use of outdated project management models that cannot keep up with AI’s unique rhythm .
Q2. The PMI-CPMAI™ methodology was originally developed by which organization?

A. Gartner
B. McKinsey & Company
C. Cognilytica®
D. Project Management Institute (PMI)
Answer: 3
Explanation: Cognilytica® created the CPMAI methodology in 2017. PMI later acquired them in September 2024 to establish it as a global standard (p. 10).
Q3. Which legacy data process was identified as "outdated" because it lacked governance, ethics, and continuous monitoring?

A. Agile
B. Waterfall
C. CRISP-DM
D. Six Sigma
Answer: 3
Explanation: While CRISP-DM provided a data focus, it failed to address modern AI complexities like ethical oversight and ongoing model evaluation, which PMI-CPMAI™ now includes .
Q4. According to research cited in the guide, what is the failure rate of AI initiatives?

A. More than 40%
B. More than 60%
C. More than 80%
D. Exactly 50%
Answer: 3
Explanation: The guide notes that more than 80% of AI initiatives do not meet their goals, which is nearly double the failure rate of typical IT projects (p. 11).
Q5. Which of the following is NOT one of the "Three Ps of Intelligence" used to differentiate AI from traditional automation?

A. Perception
B. Processing
C. Prediction
D. Planning
Answer: 2
Explanation: The Three Ps are Perception (sensing the environment), Prediction (forecasting outcomes), and Planning (devising strategies). "Processing" is a standard computer function, not a specific marker of intelligence (pp. 19-20).
Q6. In the context of the "Three Ps," if a machine perceives, predicts, and plans, it is exhibiting:

A. Traditional Automation
B. Intelligent characteristics (AI)
C. Robotic Process Automation (RPA)
D. Deterministic logic
Answer: 2
Explanation: If a machine is responsible for one or more of these functions, it exhibits characteristics of intelligence. If a human does them and the machine just executes, it is simple automation (p. 20).
Q7. Which AI pattern focuses on identifying and distinguishing objects within unstructured data like images or audio?

A. Hyper-Personalization
B. Recognition
C. Goal-Driven Systems
D. Patterns and Anomalies
Answer: 2
Explanation: The Recognition pattern uses machine learning to label, segment, and classify existing content (like facial recognition or medical imaging) .
Q8. A project manager is working on a fraud detection system. Which AI pattern is most relevant?

A. Conversational
B. Predictive Analytics
C. Patterns and Anomalies
D. Autonomous Systems
Answer: 3
Explanation: The Patterns and Anomalies pattern is specifically designed to detect when data does not fit the norm, making it the primary pattern for fraud and cybersecurity (p. 18).
Q9. Which pattern relies heavily on "Reinforcement Learning" to find the best solution to a problem through trial and error?

A. Recognition
B. Hyper-Personalization
C. Goal-Driven Systems
D. Conversational and Human Interaction
Answer: 3
Explanation: Goal-Driven Systems focus on learning objectives and finding optimal solutions by studying the goals of a problem, often through reinforcement learning ).
Q10. What is "Agentic AI"?

A. AI that only answers prompts from human agents.
B. An advanced approach where AI agents autonomously define, optimize, and iterate on workflows.
C. A system that requires manual human reprogramming for every new task.
D. A marketing term for standard Generative AI.
Answer: 2
Explanation: Agentic AI represents a shift from AI as a tool to AI as a collaborator that can independently improve its own operations with limited human interaction (p. 21).
Q11. According to the Trustworthy AI Framework, "Ethical AI" is defined as:

A. Making the AI code open source.
B. Ensuring accountability for AI outcomes.
C. Aligning AI with human values like fairness and inclusion.
D. Designing systems that provide understandable explanations.
Answer: 3
Explanation: Ethical AI is the layer of the framework that focuses on preventing discrimination and embedding ethical checks early in the project .
Q12. Which layer of Trustworthy AI ensures that stakeholders can see and understand the logic behind an AI’s decision?

A. Governed AI
B. Responsible AI
C. Transparent AI
D. Fair AI
Answer: 3
Explanation: Transparent AI focuses on making the logic understandable by documenting data sources and model reasoning (p. 23).
Q13. What is the primary purpose of Phase I (Business Understanding) in the PMI-CPMAI™ methodology?

A. To clean the data.
B. To determine if the business problem is a "right fit" for AI.
C. To select which GPU to purchase.
D. To write the final project report.
Answer: 2
Explanation: Phase I ensures AI is justified by pinpointing true needs and confirming that AI adds more value than traditional automation .
Q14. During Phase I, a "Red Light" in the Go/No-Go feasibility check means:

A. Proceed but manage risks.
B. The project is finished successfully.
C. Stop and fix issues before moving to Phase II.
D. The project needs more data scientists.
Answer: 3
Explanation: A Traffic Light approach is used: Green means go, Yellow means manage risks, and Red means stop and fix first .
Q15. All ROI goals in an AI project should be "SMART." What does the "S" stand for?

A. Sustainable
B. Strategic
C. Specific
D. Standardized
Answer: 3
Explanation: Goals must be Specific, Measurable, Achievable, Relevant, and Time-bound to be effective (p. 26).
Q16. Why is Phase II (Data Understanding) described as the "technical feasibility" step?

A. It is where the code is written.
B. It asks if the right data is in place to make the business goal work.
C. It determines the salary of the data engineer.
D. It is when the model is deployed.
Answer: 2
Explanation: Phase II tests if success is possible with the available data (quality, quantity, and accessibility) before investing in further work .
Q17. In Phase II, what does the "Veracity" of Big Data refer to?

A. How much data there is.
B. How fast the data moves.
C. Is the data reliable, consistent, and trusted?
D. What formats the data is in (CSV, JSON, etc.).
Answer: 3
Explanation: The "Four Vs" include Volume, Variety, Velocity, and Veracity (the reliability/trustworthiness of the data) .
Q18. "Data Preparation" (Phase III) sets the __________ for model performance.

A. Floor
B. Baseline
C. Ceiling
D. Budget
Answer: 3
Explanation: The guide emphasizes that no advanced algorithm can fix poor data preparation; the quality of prep sets the maximum limit of how well a model can perform .
Q19. Which task in Phase III ensures that "New York" and "NY" are treated the same by an AI model?

A. Outlier detection
B. Consistency standardization
C. Feature creation
D. Data labeling
Answer: 2
Explanation: Standardization removes variations in data formats so the AI sees consistent inputs for learning .
Q20. Why is "Bias Amplification" a risk in Phase III?

A. Because AI learns and exaggerates any bias present in the prepared data.
B. Because humans are naturally biased.
C. Because cleaning data removes the "truth."
D. Because computers cannot understand ethics.
Answer: 1
Explanation: AI doesn't just learn patterns; it amplifies them. If data preparation isn't rigorous, existing biases become even more dangerous in the final model .
Q21. A key decision in Phase IV is "Build-Versus-Buy." When is Custom Model Development considered the optimal choice?

A. When speed to market is the only priority.
B. When proprietary AI offers a strategic competitive advantage.
C. When the organization has no internal machine learning expertise.
D. When the AI pattern identified is widely available as a cloud API.
Answer: 2
Explanation: Custom builds are best when unique data patterns or proprietary IP provide a competitive edge that off-the-shelf tools cannot match (p. 35).
Q22. What is a primary risk of using third-party "Black Box" models?

A. They are too expensive to integrate.
B. They may conflict with transparency and explainability requirements.
C. They do not require any data preparation.
D. they are slower than custom-built models.
Answer: 2
Explanation: Because the inner logic of "black box" models is hidden, they can create legal or ethical risks if they fail to meet transparency standards required by regulators .
Q23. In Phase IV, "Layer Freezing" is a technique used specifically for:

A. Custom model training from scratch.
B. Pretrained model fine-tuning and transfer learning.
C. Writing prompts for LLMs.
D. Securing data encryption.
Answer: 2
Explanation: Layer freezing keeps early layers of a pretrained model (general knowledge) intact while retraining later layers for specialized organizational tasks .
Q24. For GenAI projects, what is the goal of "Prompt Engineering"?

A. To rewrite the underlying model's code.
B. To design structured input templates that deliver consistent, reliable outputs.
C. To increase the number of tokens used per request.
D. To replace the need for a database.
Answer: 2
Explanation: Prompt engineering focuses on structured design and refinement of wording to ensure the model responds accurately and safely.
Q25. "Retrieval-Augmented Generation" (RAG) is used to:

A. Train a model from scratch using internal data.
B. Ground LLM responses in trusted, real-time internal data to prevent "hallucinations."
C. Encrypt the data pipeline.
D. Automatically delete outliers in the dataset.
Answer: 2
Explanation: RAG combines LLMs with real-time information retrieval from semantic search or vector databases to ensure outputs are based on facts .
Q26. What does "Hyper-parameter tuning" help a team achieve?

A. Cleaning raw data faster.
B. Finding the optimal internal settings for a model while avoiding overfitting.
C. Communicating with stakeholders.
D. Predicting the total project cost.
Answer: 2
Explanation: Tuning (using methods like grid search) identifies the best mathematical configuration for the model to maximize performance without just "memorizing" data (p. 38).
Q27. Why should "Explainability" tools (like LIME or SHAP) be implemented during Phase IV?

A. Because they improve the speed of the model.
B. To ensure that model decisions can be interpreted by stakeholders and regulators.
C. To reduce the cost of API calls.
D. Because they are required for data cleaning.
Answer: 2
Explanation: Trustworthy AI requires that decisions are not "black boxes"; explainability tools must be integrated early to support trust and compliance .
Q28. Which technique allows for AI development while minimizing data exposure for sensitive information?

A. Cross-validation.
B. Differential privacy or Federated learning.
C. Prompt chaining.
D. Feature extraction.
Answer: 2
Explanation: These privacy-preserving methods protect individual records during the development process .
Q29. Phase V (Model Evaluation) serves as the checkpoint to verify if the model is ready for:

A. Data cleaning.
B. Production/Operationalization.
C. Initial brainstorming.
D. Hiring a new data scientist.
Answer: 2
Explanation: Phase V bridges the gap between controlled development and live, operational environments .
Q30. In Phase V, "Business Value Validation" requires the model to:

A. Achieve 100% accuracy.
B. Prove it delivers measurable outcomes like cost savings or revenue gains.
C. Be written in a specific programming language.
D. Use the largest possible dataset.
Answer: 2
Explanation: Technical performance is not enough; the model must prove it achieves the ROI targets set during Phase I .
Q31. Which metric is most appropriate for evaluating the error rate of a "Prediction" model?

A. F1 Score.
B. Mean Absolute Error (MAE).
C. Silhouette Score.
D. Recall.
Answer: 2
Explanation: MAE (along with RMSE and MAPE) is a specialized metric for regression/prediction patterns (p. 43).
Q32. "Overfitting" occurs when a model:

A. Is too simple to learn the data patterns.
B. Memorizes specific training examples rather than learning generalizable patterns.
C. Runs too fast on the hardware.
D. Has too little data to work with.
Answer: 2
Explanation: Overfitting results in high accuracy on training data but poor performance on new, real-world data .
Q33. What is the purpose of "Adversarial Robustness Testing" in Phase V?

A. To see if the team can work under pressure.
B. To test the model against malicious data manipulation or unexpected inputs.
C. To compare the model against a competitor's AI.
D. To check if the model meets the budget.
Answer: 2
Explanation: This ensures the AI fails gracefully and remains secure when encountering "tricky" or bad-faith inputs (p. 43).
Q34. "Data Drift" (Distribution Shift) assessment checks if:

A. The data is moving too slowly through the pipeline.
B. The real-world data has changed compared to the data used during training.
C. The database is full.
D. The data scientist has left the project.
Answer: 2
Explanation: If the world changes (e.g., a shift in customer behavior), the model's training data becomes outdated, causing performance to drop .
Q35. "Throughput and Latency Validation" are part of which evaluation category?

A. Ethical Validation.
B. Operational Performance Requirements.
C. Feature Engineering.
D. Business ROI.
Answer: 2
Explanation: These metrics ensure the model can handle real-world production demands, such as response time and processing capacity (p. 43).
Q36. When evaluating a "Clustering" pattern, which metric is commonly used?

A. Precision.
B. Silhouette Score.
C. Mean Absolute Percentage Error (MAPE).
D. Perplexity.
Answer: 2
Explanation: Silhouette scores (and the Davies-Bouldin index) assess the quality and stability of unsupervised learning clusters .
Q37. "Stakeholder Acceptance" in Phase V is typically achieved through:

A. Showing the stakeholders the raw code.
B. Pilot tests and user acceptance testing (UAT).
C. Sending a weekly email.
D. Increasing the project budget.
Answer: 2
Explanation: Pilots confirm that end users trust and can effectively use the system in their daily workflows .
Q38. What should a project team do if a model shows high bias during Phase V?

A. Deploy it anyway and fix it later.
B. Iteratively loop back to Phase III (Data Preparation) or Phase II to fix the data.
C. Delete the biased results from the report.
D. Change the business goals to match the bias.
Answer: 2
Explanation: The PMI-CPMAI™ phases are iterative by design; discovery of a major issue in evaluation requires returning to earlier steps to fix the foundation (p. 24).
Q39. "Fail-safes" and "Circuit Breakers" are specifically critical for which AI approach?

A. Static rule-based automation.
B. AI Agents and Agentic AI.
C. Basic spreadsheets.
D. Project documentation.
Answer: 2
Explanation: Because agents act autonomously, fail-safes are needed to stop them if they begin behaving in risky or unintended ways .
Q40. The transition from Phase V to Phase VI (Operationalization) requires:

A. Only a technical sign-off from the lead developer.
B. Evidence and confidence that the model meets technical, business, and ethical criteria.
C. The project to be under budget.
D. All data to be deleted for privacy.
Answer: 2
Explanation: Phase V provides the documented proof needed to move to full, safe deployment .
Q41. What is the primary focus of Phase VI: Model Operationalization?

A. Writing the initial model code.
B. Deploying the validated model into real-world use and establishing value delivery.
C. Cleaning the data one final time.
D. Shutting down the project team.
Answer: 2
Explanation: Phase VI is the "inference" phase where the model is put into production to solve the business problem defined in Phase I.
Q42. Which activity is a core part of Model Operationalization?

A. Selecting the best algorithm.
B. Setting up monitoring and performance management for data drift.
C. Labelling data for training.
D. Defining the ROI for the first time.
Answer: 2
Explanation: Operationalization isn't just deployment; it requires continuous monitoring to ensure the model doesn't degrade over time.
Q43. What is "Model Drift" in the context of Phase VI?

A. The model moving to a different server.
B. A decline in prediction accuracy as real-world data changes.
C. The project team changing its goals.
D. The model being rewritten in a different language.
Answer: 2
Explanation: Model drift (or performance degradation) occurs when the model's accuracy drops because the live data no longer matches the patterns it was trained on.
Q44. Your team needs to decide on a deployment approach for a system requiring instant predictions on machinery. Which is most appropriate?

A. Batch prediction.
B. Real-time prediction.
C. Monthly manual updates.
D. Proof-of-concept lab.
Answer: 2
Explanation: For systems requiring immediate response (like machinery monitoring), Real-time prediction is the standard operational approach.
Q45. What is the role of "Retraining Pipelines" in Phase VI?

A. To train new employees on how to use AI.
B. To automatically update models with new data to maintain accuracy.
C. To delete old data from the server.
D. To rewrite the project scope.
Answer: 2
Explanation: Retraining pipelines are essential to combat drift by allowing the model to "learn" from new data regularly.
Q46. According to the guide, when is the AI project lifecycle officially "finished"?

A. After the model is deployed.
B. When the budget is spent.
C. Never; it is an iterative process that loops back for continuous improvement.
D. When the Data Scientist leaves.
Answer: 3
Explanation: The PMI-CPMAI™ framework is iterative; after deployment, teams often return to Phase I to reassess or expand the solution.
Q47. In a real-world scenario, a "Biased Hiring Algorithm" creates what kind of risk?

A. High compute cost risk.
B. Legal liability and ethical risk.
C. Data storage risk.
D. Hardware failure risk.
Answer: 2
Explanation: Bias in high-stakes domains like hiring or finance leads directly to regulatory and ethical consequences.
Q48. If an AI diagnostic tool makes a dangerous misdiagnosis in healthcare, what is the best mitigation tactic?

A. Hide the results from the patient.
B. Ensure strong "Responsible AI" checks and human-in-the-loop oversight.
C. Speed up the model.
D. Switch to a different cloud provider.
Answer: 2
Explanation: High-risk systems require accountability and continuous monitoring to maintain safety standards.
Q49. To prevent "Market Risk" in Financial AI, what must be monitored?

A. Only the server's CPU usage.
B. Systemic behavior and model performance under market volatility.
C. The number of users logged in.
D. The project's social media presence.
Answer: 2
Explanation: Financial models must be tested for stability during periods of high change or volatility.
Q50. A customer service AI violates privacy regulations. What should have been the first step to prevent this?

A. Buying faster servers.
B. Conducting a Privacy Impact Assessment (PIA).
C. Hiring more marketing staff.
D. Increasing the accuracy of the chatbot.
Answer: 2
Explanation: A PIA is a foundational risk control that identifies privacy risks before any sensitive data is processed.
Q51. What is "Data Lineage" and why is it important for governance?

A. A list of names of everyone on the project.
B. The ability to trace data back to its source and see every transformation step.
C. The speed at which data travels.
D. The backup location of the data.
Answer: 2
Explanation: Data lineage provides the traceability required for audits, proving that AI decisions are based on well-governed data.
Q52. In a "Gradual Phased Rollout," what is the benefit for the organization?

A. It is cheaper than a full rollout.
B. It builds trust and reduces employee anxiety while allowing for adjustments.
C. It requires no training.
D. It eliminates the need for monitoring.
Answer: 2
Explanation: Phased rollouts are a core change management tactic that supports iterative adoption and better stakeholder buy-in.
Q53. What should a Project Manager do FIRST if a model overwrites the original and performance drops?

A. Delete all data.
B. Implement versioning to allow rolling back to the original model.
C. Buy more cloud storage.
D. Ask for more budget.
Answer: 2
Explanation: Model versioning is a critical operational control to ensure business continuity if an update fails.
Q54. Which role bridges the gap between development and production by managing CI/CD and monitoring?

A. Project Manager.
B. MLOps/DevOps Engineer.
C. Business Analyst.
D. Data Scientist.
Answer: 2
Explanation: MLOps engineers are specifically responsible for the reliable deployment and monitoring of machine learning models.
Q55. "Audit Trails" are a requirement of which Trustworthy AI layer?

A. Ethical AI.
B. Responsible AI.
C. Hyper-Personalization.
D. Patterns and Anomalies.
Answer: 2
Explanation: Responsible AI ensures accountability through mechanisms like audit trails and escalation paths.
Q56. Why is a "Fact Checklist" used during AI data governance?

A. To check if the team is happy.
B. To systematically verify data sources, integrity, and provenance.
C. To write a creative story about the data.
D. To increase the volume of data.
Answer: 2
Explanation: A checklist ensures that data has not been tampered with and conforms to regulatory rules.
Q57. If an AI system is used for "Route Optimization," which pattern is it typically following?

A. Conversational.
B. Recognition.
C. Predictive Analytics.
D. Autonomous Systems.
Answer: 3
Explanation: Optimizing routes requires forecasting future travel times and traffic conditions based on data.
Q58. What is the "highest risk" when operationalizing a decentralized data storage system for AI?

A. It is too slow.
B. Unauthorized access due to complex entry points and weak endpoints.
C. It costs too much.
D. It requires too many people.
Answer: 2
Explanation: Decentralization makes access control and identity management much more challenging for security.
Q59. What should be considered "continuous" throughout all six phases of the project?

A. Coding.
B. Ethics and trustworthy practices.
C. Spending the budget.
D. Hiring new staff.
Answer: 2
Explanation: Ethics is not a one-time step; a "continuous ethics thread" must run through every phase of the project.
Q60. "Vision Expansion" during iterations involves:

A. Identifying new opportunities to apply AI across the organization.
B. Looking at the screen for too long.
C. Making the model bigger for no reason.
D. Increasing the project manager's salary.
Answer: 1
Explanation: After a successful project, teams leverage operational learning to expand AI's value to other business areas
Q61. Which role is primarily responsible for ensuring that the AI project delivers measurable business ROI and remains strategically aligned?

A. Data Scientist
B. Executive Sponsor
C. MLOps Engineer
D. Project Manager
Answer: 2
Explanation: The Executive Sponsor secures cross-functional buy-in and ensures the project is not just a "technical experiment" but a strategic business initiative (p. 6).
Q62. Why is the "Business Analyst / Domain Expert" critical during Phase V (Model Evaluation)?

A. To write the deployment scripts.
B. To confirm that the model’s outputs make sense in the real-world business context.
C. To label the data.
D. To manage the budget.
Answer: 2
Explanation: AI may find statistical patterns that are technically accurate but business-irrelevant. Domain Experts validate that results are actionable and logically sound (p. 18).
Q63. According to the guide, what is a primary shift for Project Managers in the age of AI?

A. They must become expert Python coders.
B. They should shift toward higher-value work like coaching and stakeholder management.
C. They must personally label 20% of the training data.
D. They should focus exclusively on hardware procurement.
Answer: 2
Explanation: As AI automates routine tasks like documentation and reporting, PMs must pivot to soft skills and strategic leadership (p. 12).
Q64. Which role bridges the gap between development and production by managing CI/CD and monitoring?

A. Data Engineer
B. MLOps/DevOps Engineer
C. Business Owner
D. Data Labeler
Answer: 2
Explanation: MLOps Engineers are responsible for the operational stability, versioning, and performance monitoring of deployed models (p. 6).
Q65. What is the primary goal of Phase VI in the PMI-CPMAI™ methodology?

A. To clean the raw data.
B. To deploy the validated model and establish a cycle of continuous value delivery.
C. To choose the best algorithm.
D. To write the project charter.
Answer: 2
Explanation: Phase VI is where the model is put into real-world production and its performance is monitored for long-term value (p. 5).
Q66. What is "Model Drift" (Performance Degradation)?

A. When the server hardware fails.
B. When a model's accuracy drops because real-world data patterns have shifted since training.
C. When the team loses focus on the project goals.
D. When the project budget is exceeded.
Answer: 2
Explanation: Drift occurs when the "live" environment no longer matches the training environment (e.g., a sudden change in consumer behavior) (p. 43).
Q67. "Batch Prediction" is most suitable for which type of scenario?

A. A self-driving car making split-second decisions.
B. A scheduled nightly report that calculates credit scores for 10,000 applicants.
C. A real-time customer service chatbot.
D. An emergency alert system for factory failures.
Answer: 2
Explanation: Batch processing handles large-scale data on a schedule (e.g., nightly/weekly) where immediate latency is not required (p. 5).
Q68. Why is "Model Versioning" critical in Phase VI?

A. To keep track of the Project Manager's salary.
B. To allow "rollbacks" to a previous stable model if a new update fails in production.
C. To increase the cost of the project.
D. To ensure only one person can edit the model.
Answer: 2
Explanation: Versioning is an essential risk mitigation tactic that ensures business continuity by letting teams revert to a known good state if an update behaves poorly (p. 5).
Q69. What differentiates "Agentic AI" from standard AI systems?

A. It is slower and requires more prompts.
B. It can autonomously define, optimize, and iterate on workflows with limited human interaction.
C. It only works for text generation.
D. It does not require any data for training.
Answer: 2
Explanation: Agentic AI represents a shift from "AI as a tool" to "AI as an intelligent collaborator" that can refine its own processes (p. 21).
Q70. A "Circuit Breaker" in Agentic AI is designed to:

A. Increase the speed of the agent.
B. Automatically stop an agent if it behaves in a risky or unintended way.
C. Delete the agent's memory.
D. Help the agent learn faster.
Answer: 2
Explanation: Fail-safes/Circuit breakers are vital for autonomous agents to prevent them from drifting into risky behaviors or exceeding cost limits (p. 22).
Q71. In multi-agent systems, "Orchestration" refers to:

A. Deleting agents that are too slow.
B. Managing how specialized agents work together to solve complex, multifaceted problems.
C. Training all agents on the same data.
D. Making agents compete against each other.
Answer: 2
Explanation: Orchestration ensures different specialized agents contribute their unique intelligence to a single, coordinated workflow (p. 20).
Q72. Why should a team use a "Simulation Sandbox" for AI Agents?

A. To let the team play games.
B. To safely test agent behavior in complex scenarios before full production rollout.
C. To store the final project results.
D. To encrypt the training data.
Answer: 2
Explanation: Sandboxes provide a controlled environment to catch "emergent behaviors" that could be dangerous in a live system.
Q73. In the "Biased Hiring Algorithm" scenario, what is the recommended mitigation?

A. Hide the hiring data.
B. Implement fairness metrics and "debiasing" during the training/preparation phases.
C. Only use AI for part-time hiring.
D. Let the AI finish its training before checking for bias.
Answer: 2
Explanation: Trustworthy AI requires systematic bias assessment and mitigation techniques like fair representation learning (p. 40).
Q74. Which layer of the Trustworthy AI Framework ensures that AI logic is documented and its decisions are interpretable?

A. Ethical AI
B. Transparent AI
C. Governed AI
D. Responsible AI
Answer: 2
Explanation: Transparent AI focuses on making the "black box" understandable by documenting data sources and implementation reasoning (p. 23).
Q75. "Federated Learning" is a privacy-preserving method because:

A. It deletes data after 24 hours.
B. It allows models to learn from decentralized data without the raw data ever leaving the original device.
C. It is only used by the government.
D. It requires no internet.
Answer: 2
Explanation: This allows for privacy-compliant development where sensitive raw data stays on local devices while the central model learns general patterns .
Q76. What is a "Privacy Impact Assessment" (PIA)?

A. A check to see if the project is under budget.
B. A structured assessment of risks to individual data privacy throughout the project.
C. A test for server speed.
D. A survey for employees.
Answer: 2
Explanation: A PIA is a core governance tool used to identify and mitigate privacy liabilities before processing sensitive data .
Q77. The "F1 Score" is most valuable for evaluation when:

A. The data is perfectly balanced.
B. There is an uneven distribution of data classes (e.g., in fraud detection).
C. The model is very slow.
D. Only accuracy matters.
Answer: 2
Explanation: The F1 score balances Precision and Recall, making it better for "unbalanced" data where accuracy alone might be misleading (p. 42).
Q78. "Throughput and Latency" are metrics used to measure:

A. Ethical bias.
B. Operational performance and system capacity.
C. The number of lines of code.
D. Stakeholder happiness.
Answer: 2
Explanation: These technical metrics ensure the model can handle production-level traffic and response time requirements .
Q79. What does "Explainable AI" (XAI) tools like SHAP or LIME help achieve?

A. Faster model training.
B. Better understanding of which features most influenced a model’s specific decision.
C. More data volume.
D. Lower API costs.
Answer: 2
Explanation: Explainability tools provide a rationale for "black box" decisions, which is critical for trust and regulatory audits .
Q80. "Adversarial Robustness Testing" involves:

A. Testing the team's ability to work under stress.
B. Testing a model against malicious data manipulation or "tricky" edge cases.
C. Comparing the model against a competitor's model.
D. Checking if the model meets the budget.
Answer: 2
Explanation: This testing ensures the model fails gracefully and isn't easily tricked into making incorrect or unsafe decisions (p. 43).
Q81. What is "Data Lineage," and why is it critical for AI governance?

A. A list of all data scientists who worked on the project.
B. The ability to trace data back to its source through every transformation step.
C. The physical location where the backup tapes are stored.
D. The speed at which data travels from the server to the user.
Answer: 2
Explanation: Data lineage provides traceability and auditability. It allows organizations to prove that the data used to train an AI was sourced ethically, handled legally, and not tampered with during the preparation phase.
Q82. In the "Healthcare AI" risk scenario, what is the primary cause of dangerous misdiagnoses?

A. Lack of enough cloud storage.
B. Training models on non-representative data or failing to account for edge cases.
C. Using Python instead of Java.
D. The AI system running too fast.
Answer: 2
Explanation: High-risk scenarios like healthcare require rigorous validation across diverse populations to ensure the AI doesn't fail when encountering patient types it hasn't seen before.
Q83. Which task in Phase III (Data Preparation) involves converting "Customer Names" or "Cities" into a numerical format algorithms can understand?

A. Outlier detection.
B. Encoding and normalization.
C. Data versioning.
D. Privacy impact assessment.
Answer: 2
Explanation: Algorithms require mathematical inputs. Encoding converts categorical data (text) into numbers, while normalization ensures different features (like age and income) are on the same scale so one doesn't unfairly skew the learning.
Q84. "Model Inversion" is a specific security risk where an attacker:

A. Turns the model's logic upside down.
B. Attempts to reconstruct sensitive training data by analyzing the model's outputs.
C. Deletes the model from the server.
D. Changes the project's ROI targets.
Answer: 2
Explanation: This is a privacy attack. If a model is too "revealing," an attacker can reverse-engineer it to find out who was in the training set, violating confidentiality.
Q85. Why is "Model Versioning" considered a critical operational control in Phase VI?

A. It allows the team to change the project name.
B. It enables a "rollback" to a previous stable model if a new update performs poorly.
C. it is required to pay the software vendors.
D. It increases the accuracy of the model automatically.
Answer: 2
Explanation: In production, a new update might behave unexpectedly. Versioning ensures business continuity by letting the team revert to the last known "good" version immediately.
Q86. What is the primary purpose of "Temporal Feature Engineering"?

A. Predicting when the project will be finished.
B. Creating indicators like rolling averages or lags to capture patterns over time.
C. Measuring the team's coffee breaks.
D. Deleting data that is older than 5 years.
Answer: 2
Explanation: For time-series data (like sales or fraud detection), the AI needs to see trends (e.g., "is spending increasing compared to last month?") rather than just a single data point.
Q87. In the context of Agentic AI, a "Circuit Breaker" is used to:

A. Save electricity in the data center.
B. Automatically stop an autonomous agent if it exceeds safety or cost thresholds.
C. Speed up the agent's decision-making.
D. Reset the model's password.
Answer: 2
Explanation: Because agents act with autonomy, fail-safes (circuit breakers) are required to stop them if they begin looping infinitely or taking risky actions without human approval.
Q88. Which "Pattern" of AI would be used for a system that optimizes a city's "Smart Grid" energy flow?

A. Conversational.
B. Recognition.
C. Goal-Driven Systems.
D. Patterns and Anomalies.
Answer: 3
Explanation: Energy grids require optimization to find the best flow to achieve a goal (efficiency). This often involves reinforcement learning, a hallmark of the Goal-Driven pattern.
Q89. "Explainability" (XAI) should be implemented during which phase to be most effective?

A. Phase VI (after deployment).
B. Phase IV (Model Development).
C. Only if a user complains.
D. At the very end of the project.
Answer: 2
Explanation: Explainability cannot be "bolted on" later. It must be designed into the model architecture during Phase IV to ensure that decisions can be audited and understood.
Q90. What is "Alert Fatigue," and how is it managed in MLOps?

A. The team getting tired of long meetings.
B. Monitoring systems sending too many false-positive notifications.
C. The server running out of battery.
D. The model being too accurate.
Answer: 2
Explanation: If a monitor is too sensitive, it creates "noise." PMs must ensure thresholds are optimized so the team only receives alerts for significant performance drops.
Q91. A "Held-out" dataset is specifically used to:

A. Store data that was too dirty to use.
B. Provide a final, unbiased evaluation of the model in Phase V.
C. Train the model's first iteration.
D. Hide data from regulators.
Answer: 2
Explanation: To get a "true" measure of accuracy, you must test the model on data it has never seen before. This is the "held-out" set.
Q92. Which "V" of Big Data asks whether the data is structured or unstructured?

A. Volume.
B. Variety.
C. Velocity.
D. Veracity.
Answer: 2
Explanation: Variety refers to the different formats (text, images, SQL tables, etc.) that the project manager must account for in the preparation pipeline.
Q93. What is "Statistical Profile Verification"?

A. Predicting the project's profit.
B. Comparing prepared datasets against business logic (e.g., ensuring "Price" is not negative).
C. Checking the team's social media profiles.
D. Measuring the model's speed.
Answer: 2
Explanation: This is a quality check in Phase III to ensure the data makes sense before it is fed into the model.
Q94. The "Cold Start" problem is a major challenge for which AI pattern?

A. Recognition.
B. Hyper-Personalization.
C. Autonomous Systems.
D. Patterns and Anomalies.
Answer: 2
Explanation: Personalization requires history. If a user is new, the system has a "cold start" because it doesn't know their preferences yet to make a recommendation.
Q95. "Data Augmentation" is a technique used to:

A. Delete duplicate data.
B. Artificially increase dataset variety by slightly modifying existing data.
C. Increase the project's budget.
D. Hire more data labelers.
Answer: 2
Explanation: In image recognition, for example, you might flip or rotate an image of a car to create "new" training data, helping the model learn to recognize it from different angles.
Q96. Why is "Cross-Border Data Consideration" critical for international AI projects?

A. To ensure the AI speaks different languages.
B. To comply with local privacy laws (like GDPR) that limit how data can be moved or stored.
C. To save money on internet costs.
D. To hire developers from different countries.
Answer: 2
Explanation: Different countries have different sovereignty laws. Moving sensitive data across borders without a plan can lead to massive legal fines.
Q97. "Underfitting" is evaluated in Phase V by:

A. Checking if the model is too fast.
B. Comparing performance against a simpler baseline model.
C. Deleting the model and starting over.
D. Increasing the data scientist's salary.
Answer: 2
Explanation: If a model is too simple to learn the data, it is "underfitting." Comparing it to a basic baseline helps determine if the model needs more complexity.
Q98. What is the role of a "Simulation Sandbox" in Agentic AI?

A. To let the team play games during lunch.
B. To test agent behavior in complex, risk-free environments before live deployment.
C. To store the project's final code.
D. To encrypt the database.
Answer: 2
Explanation: Sandboxes allow for safe testing of emergent behaviors. It's where you find out if your autonomous agent will do something unexpected or dangerous.
Q99. "Vision Expansion" typically occurs during which stage?

A. At the very beginning of the first project.
B. During subsequent iterations after a successful rollout.
C. Only if the project fails.
D. During data preparation.
Answer: 2
Explanation: Once a model is operational and successful, the team leverages that operational learning to identify new business areas where AI can add value.
Q100. What does the "CPMAI" in PMI-CPMAI™ stand for?

A. Certified Project Manager in AI.
B. Certified Professional in Managing AI.
C. Cost Per Model Artificial Intelligence.
D. Cloud Project Management for AI.
Answer: 2
Explanation: Practitioners who hold this certification have demonstrated they have the critical skills necessary to manage AI projects, which is the core focus of the guide.
Q101. The ultimate success of a PMI-CPMAI™ project is defined by:

A. Deploying the largest possible neural network.
B. Delivering sustainable business value through a trustworthy, ethical AI system.
C. Spending exactly 100% of the allocated budget.
D. Having the most "mentions" on social media.
Answer: 2
Explanation: The guide emphasizes that AI projects are not just technical experiments; they are business transformations that must be ethical, reliable, and value-driven.
Share