[Microsoft] AI-900 - Azure AI Fundamentals Exam Dumps & Study Guide

Free [Microsoft] AI-900 - Azure AI Fundamentals Practice Questions Preview

• Question 1

  A company employs a team of customer service agents to provide telephone and email support to customers.
  The company develops a webchat bot to provide automated answers to common customer queries.
  Which business benefit should the company expect as a result of creating the webchat bot solution?
  

  • A. increased sales
  • B. a reduced workload for the customer service agents
  • C. improved product reliability

  Correct Answer: B

  Explanation:

  Answer: B (A reduced workload for the customer service agents)
  
  Reasoning: Webchat bots are designed to automate repetitive customer interactions, such as answering frequently asked questions (e.g., order status checks, account updates, or basic troubleshooting). By handling these routine inquiries, the bot reduces the volume of customer requests requiring human intervention. This directly decreases the workload for human agents, allowing them to prioritize complex or escalated issues. Industry sources like Microsoft's Azure Bot Service documentation emphasize that bots "free up human agents to focus on higher-value tasks" by resolving common queries autonomously^[1]. Similarly, IBM's Watson Assistant use cases highlight workload reduction as a key outcome, noting that chatbots can resolve up to 80% of routine inquiries, significantly alleviating agent burden^[2]. While increased sales (A) or improved product reliability (C) are potential indirect benefits, they are not the primary goal of a customer support chatbot. Sales growth would require integration with marketing strategies, and product reliability relates to engineering processes, neither of which is addressed by automating customer queries.
  
  Citations:
  [1] Microsoft Azure Bot Service Overview
  [2] IBM Watson Assistant Use Cases

• Question 2

  For a machine learning progress, how should you split data for training and evaluation?
  

  • A. Use features for training and labels for evaluation.
  • B. Randomly split the data into rows for training and rows for evaluation.
  • C. Use labels for training and features for evaluation.
  • D. Randomly split the data into columns for training and columns for evaluation.

  Correct Answer: B

  Explanation:

  Recommended Answer: B. Randomly split the data into rows for training and rows for evaluation.
  
  Reasoning:
  In machine learning, data is typically split row-wise (option B) to ensure both features (input variables) and labels (target variables) are available in both training and evaluation datasets. This preserves the integrity of individual data instances. Here’s why:
  - Training requires features and labels: Models learn patterns by mapping features to labels during training. Splitting by features/labels (options A/C) would leave incomplete data for either phase.
  - Column-wise splitting (option D) disrupts feature sets, making models incompatible with evaluation data structures.
  - Row-wise splitting (e.g., 70:30 or 80:20 ratios) is standard practice, as noted in Microsoft’s AI-900 documentation and frameworks like scikit-learn. For example, train_test_split in scikit-learn splits rows to maintain consistency.
  
  Citations:
  - Microsoft Learn: Evaluate Machine Learning Models
  - Scikit-learn Documentation: train_test_split

• Question 3

  HOTSPOT -
  You are developing a model to predict events by using classification.
  You have a confusion matrix for the model scored on test data as shown in the following exhibit.
  
  Use the drop-down menus to select the answer choice that completes each statement based on the information presented in the graphic.
  NOTE: Each correct selection is worth one point.
  Hot Area:
  
  

  Correct Answer: See interactive view.

  Explanation:

  Recommended Answer:
  - Correctly predicted positives (True Positives): 11
  - False Negatives: 1,033
  
  Reasoning:
  In a confusion matrix, True Positives (TP) represent cases where the model correctly predicts the positive class ("1"), while False Negatives (FN) occur when the model incorrectly predicts the negative class ("0") despite the actual value being positive ("1").
  
  The confusion matrix in this question reverses the standard axis convention used in Microsoft documentation (where rows typically represent actual values and columns represent predicted values). Here, the matrix likely follows a predicted (rows) vs. actual (columns) structure:
  - TP = 11: Predicted "1" and Actual "1" (top-right cell).
  - FN = 1,033: Predicted "0" but Actual "1" (bottom-right cell).
  
  This interpretation aligns with Microsoft’s guidance on evaluating classification models, where False Negatives are explicitly defined as "actual yes, predicted no" scenarios. Contributors in discussions emphasized verifying axis labels to avoid misclassification of metrics.
  
  Citations:
  - Microsoft Azure Machine Learning documentation on confusion matrices.
  - Model evaluation metrics in Azure ML Designer.

• Question 4

  You build a machine learning model by using the automated machine learning user interface (UI).
  You need to ensure that the model meets the Microsoft transparency principle for responsible AI.
  What should you do?
  

  • A. Set Validation type to Auto.
  • B. Enable Explain best model.
  • C. Set Primary metric to accuracy.
  • D. Set Max concurrent iterations to 0.

  Correct Answer: B

  Explanation:

  Answer: B (Enable Explain best model)
  
  Reasoning: Microsoft's Responsible AI transparency principle mandates that AI systems be interpretable and their decision-making processes explainable. Enabling "Explain best model" in Azure Automated Machine Learning (AutoML) generates feature importance metrics (e.g., SHAP values) for the top-performing model, revealing which input variables most influenced predictions. This aligns with transparency goals by allowing developers and stakeholders to understand model behavior, validate fairness, and identify biases. Options A (validation type) and C (accuracy metric) address model training or performance but not interpretability. Option D (concurrent iterations) controls resource usage, which is unrelated to transparency.
  
  Citations:
  - Microsoft's Responsible AI principles emphasize transparency as a core requirement.
  - Azure AutoML documentation confirms enabling "Explain best model" generates model explanations.

• Question 5

  HOTSPOT -
  For each of the following statements, select Yes if the statement is true. Otherwise, select No.
  NOTE: Each correct selection is worth one point.
  Hot Area:
  
  

  Correct Answer: See interactive view.

  Explanation:

  Recommended Answer: No, Yes, No (N/Y/N)

  Reasoning:

  1. Scenario A (Predicting housing prices over time): This is a regression task, not anomaly detection. Regression predicts continuous numerical values (e.g., future prices), whereas anomaly detection identifies rare deviations from expected patterns. Microsoft Learn confirms regression applies to forecasting, such as sales or price trends (Azure ML Algorithm Cheat Sheet).
  2. Scenario B (Detecting suspicious sign-ins): This is a classic anomaly detection use case. It identifies unusual behavior (e.g., impossible geographic travel between logins) to flag potential fraud. Microsoft Security documentation highlights anomaly detection for threat identification in user activity (Azure Anomaly Detection Guide).
  3. Scenario C (Diabetes likelihood prediction): This is a classification problem (binary outcome: "diabetic" or "non-diabetic"), not anomaly detection. Medical predictions based on historical data are classification tasks under supervised learning. Microsoft Learn notes classification for categorical outcomes in healthcare (Healthcare ML Use Cases).

  Debates & Consensus: While there was initial uncertainty about Scenario C (due to potential overlaps with anomaly detection in rare medical cases), consensus emerged that classification aligns with its categorical outcome. The answer is reinforced by Microsoft’s official categorization and historical voting patterns in AI-900 discussions.

• Question 6

  HOTSPOT -
  To complete the sentence, select the appropriate option in the answer area.
  Hot Area:
  
  

  Correct Answer: See interactive view.

  Explanation:

  Recommended Answer: Reliability and Safety
  
  Reasoning:
  The correct answer aligns with Microsoft's Responsible AI principle of Reliability and Safety. This principle emphasizes that AI systems must operate consistently under diverse conditions, including edge cases, through rigorous testing, validation, and monitoring. Microsoft's documentation explicitly states that AI systems should be "resilient to manipulation" and maintain performance over time, even when encountering unusual or missing input values. Handling such scenarios ensures safe and predictable behavior, which is critical for real-world deployment.
  
  While "privacy and security" was debated in discussions, this principle focuses on data protection rather than system robustness. "Transparency" relates to explainability of AI decisions, and "inclusiveness" addresses accessibility and bias mitigation. Neither directly addresses edge-case safety.
  
  Citations:
  - Microsoft's Responsible AI Standard: https://www.microsoft.com/ai/responsible-ai
  - Microsoft's Reliability and Safety Documentation: https://learn.microsoft.com/en-us/azure/architecture/guide/responsible-ai/trusted-ai

• Question 7

  DRAG DROP -
  Match the types of AI workloads to the appropriate scenarios.
  To answer, drag the appropriate workload type from the column on the left to its scenario on the right. Each workload type may be used once, more than once, or not at all.
  NOTE: Each correct selection is worth one point.
  Select and Place:
  
  

  Correct Answer: See interactive view.

  Explanation:

  Box 3: Natural Language Processing (NLP)

  Reasoning:
  The scenarios listed in Box 3 (sentiment analysis, key phrase extraction, document categorization, language detection, etc.) are core NLP tasks. Microsoft’s Azure Cognitive Services explicitly categorizes these under NLP capabilities via services like Text Analytics (for sentiment, key phrases), Translator (language detection), and Language Understanding (LUIS) (intent recognition and categorization). While Conversational AI (e.g., chatbots) often uses NLP techniques, the listed tasks are foundational NLP workflows. Computer Vision applies to image/video analysis, which is irrelevant here. Microsoft’s AI-900 exam framework emphasizes NLP for text-based analysis, making this the most accurate answer.

  Citations:
  - Azure Language Service (Microsoft Learn)
  - Azure AI Services Overview (Microsoft Learn)

• Question 8

  You are designing an AI system that empowers everyone, including people who have hearing, visual, and other impairments.
  This is an example of which Microsoft guiding principle for responsible AI?
  

  • A. fairness
  • B. inclusiveness
  • C. reliability and safety
  • D. accountability

  Correct Answer: B

  Explanation:

  Recommended Answer: B (inclusiveness)
  
  Reasoning: Microsoft's responsible AI guiding principle of inclusiveness explicitly focuses on ensuring AI systems are accessible and beneficial to people of all abilities, backgrounds, and experiences. Designing AI to empower individuals with hearing, visual, or other impairments aligns directly with this principle. Microsoft emphasizes inclusive design practices to address accessibility barriers, as seen in their AI for Accessibility initiative and official documentation. For example, Microsoft's Responsible AI Standard states that inclusiveness requires systems to "address the needs of people with disabilities" and "reflect universal design principles" (Microsoft, 2023).
  
  Citations:
  - Microsoft's Responsible AI Principles: https://www.microsoft.com/ai/responsible-ai
  - AI for Accessibility Program: https://www.microsoft.com/ai/ai-for-accessibility
  - Inclusive Design Guidelines: https://learn.microsoft.com/en-us/style-guide/inclusive-software

• Question 9

  DRAG DROP -
  Match the Microsoft guiding principles for responsible AI to the appropriate descriptions.
  To answer, drag the appropriate principle from the column on the left to its description on the right. Each principle may be used once, more than once, or not at all.
  NOTE: Each correct selection is worth one point.
  Select and Place:
  
  

  Correct Answer: See interactive view.

  Explanation:

  Box 1: Reliability and Safety
  Box 2: Accountability
  Box 3: Privacy and Security
  
  Reasoning:
  Microsoft's responsible AI framework outlines six guiding principles. For Box 1, "AI systems must operate reliably, safely, and resist manipulation" directly aligns with the Reliability and Safety principle, which emphasizes system dependability and resistance to harmful interference (Microsoft Responsible AI, 2023).
  
  Box 2 matches Accountability, as the requirement for human oversight and responsibility for AI decisions reflects Microsoft's focus on human governance. The discussion summary highlights this with references to Azure documentation on auditability and human override capabilities.
  
  Box 3 corresponds to Privacy and Security. The mention of "transparency and user controls for data" relates to Microsoft's commitment to ethical data practices, including GDPR compliance and explicit user consent, as detailed in the Microsoft AI Principles (Azure Trust Center). While "transparency" is a standalone principle, the emphasis on data controls here ties it to privacy protections.
  
  Citations:
  - Microsoft Responsible AI Principles: https://www.microsoft.com/ai/responsible-ai
  - Azure AI Security & Privacy: https://azure.microsoft.com/en-us/explain/ai/privacy/

• Question 10

  HOTSPOT -
  To complete the sentence, select the appropriate option in the answer area.
  Hot Area:
  
  

  Correct Answer: See interactive view.

  Explanation:

  Recommended Answer: Reliability and Safety
  
  Reasoning: The correct answer aligns with Microsoft's Trusted AI principles, which emphasize that AI systems must operate reliably and safely under both normal and unexpected conditions. This includes resistance to harmful manipulation and consistent performance as designed. The discussion summary highlights broad consensus on this point, supported by Microsoft's framework requiring rigorous testing, edge-case validation, and system resilience. These criteria directly map to the "Reliability and Safety" pillar of responsible AI, as outlined in Microsoft's documentation.
  
  Citations:
  - Microsoft Responsible AI Principles: https://www.microsoft.com/ai/responsible-ai
  - Trusted AI Framework (Testing & Resilience): https://learn.microsoft.com/en-us/azure/architecture/guide/responsible-ai/