Free Microsoft AI-900 Practice Exam with Questions & Answers

Microsoft’s Responsible AI principles are central to the AI-900 curriculum and consist of six key tenets:

Fairness – AI systems should treat all people fairly.

Reliability and safety – AI systems should perform reliably and safely.

Privacy and security – AI systems should be secure and respect user privacy.

Inclusiveness – AI systems should empower everyone.

Transparency – AI systems should be understandable.

Accountability – People should be accountable for AI outcomes.

The statement “AI systems should be secure and respect privacy” reflects the Privacy and Security principle, which ensures AI solutions protect personal data and operate within compliance frameworks. Microsoft’s responsible AI framework emphasizes building trust by safeguarding sensitive data used in AI applications.

The other options do not align with official responsible AI principles; for example, AI systems need not “be in the public domain,” nor are they meant to prioritize developers’ interests or expose personal details. Hence, the correct and Microsoft-verified answer is C. AI systems should be secure and respect privacy.

This question evaluates understanding of fundamental machine learning concepts as covered in the Microsoft Azure AI Fundamentals (AI-900) official study guide and Microsoft Learn module “Explore the machine learning process.” These statements relate to data labeling, model evaluation practices, and performance metrics—three essential parts of building and assessing a machine learning model.

Labelling is the process of tagging training data with known values → YesAccording to Microsoft Learn, “Labeling is the process of tagging data with the correct output value so the model can learn relationships between inputs and outputs.” This is essential for supervised learning, where models require historical data with known outcomes. For example, if training a model to recognize fruit images, each image is labeled as “apple,” “banana,” or “orange.” Hence, this statement is true.

You should evaluate a model by using the same data used to train the model → NoThe AI-900 guide stresses that using the same data for both training and evaluation can cause overfitting, where the model performs well on training data but poorly on unseen data. Instead, the dataset is split into training and testing (or validation) subsets. Evaluation must use test data that the model has never seen before to ensure an unbiased measure of performance. Therefore, this statement is false.

Accuracy is always the primary metric used to measure a model’s performance → NoMicrosoft Learn emphasizes that accuracy is only one metric and not always the best choice. Depending on the problem type, other metrics such as precision, recall, F1-score, or AUC (Area Under the Curve) may be more appropriate—especially in cases with imbalanced datasets. For example, in fraud detection, recall may be more important than accuracy. Thus, this statement is false.

According to the Microsoft Azure AI Fundamentals (AI-900) official study guide and Microsoft Learn modules on machine learning concepts, ensuring that the accuracy of a predictive model can be proven requires data partitioning—specifically splitting the available data into training and testing datasets. This is a foundational concept in supervised machine learning.

When you split the data, typically about 70–80% of the dataset is used for training the model, while the remaining 20–30% is used for testing (or validation). The reason behind this approach is to ensure that the model’s performance metrics—such as accuracy, precision, recall, and F1-score—are evaluated on data the model has never seen before. This prevents overfitting and allows you to demonstrate that the model generalizes well to new, unseen data.

In the AI-900 Microsoft Learn content under “Describe the machine learning process”, it is explained that after cleaning and transforming the data, the next essential step is data splitting to “evaluate model performance objectively.” By keeping training and testing data separate, you can prove the reliability and accuracy of the model’s predictions, which is particularly crucial in sensitive domains like clinical or healthcare analytics, where decision transparency and validation are vital.

Option A (Train the model by using the clinical data) is incorrect because you should not train and evaluate on the same data—it would lead to biased results.

Option C (Train the model using automated ML) is incorrect because automated ML is a method for training and tuning, but it doesn’t inherently prove accuracy.

Option D (Validate the model by using the clinical data) is also incorrect if you use the same dataset for validation and training—it would not prove true accuracy.

Therefore, per Microsoft’s official AI-900 study content, the verified correct answer is B. Split the clinical data into two datasets.

The Azure OpenAI DALL-E model is a generative image model designed to create original images from textual descriptions (prompts). According to the Microsoft Learn documentation and the AI-900 study guide, DALL-E’s primary function is text-to-image generation—it converts creative or descriptive text input into visually relevant imagery.

“Generate captions for uploaded images” → NoDALL-E cannot create image captions. Captioning an image (describing what’s in an uploaded image) is a vision analysis task, not an image generation task. That functionality belongs to Azure AI Vision, which can analyze and describe images, detect objects, and generate captions automatically.

“Reliably generate technically accurate diagrams” → NoWhile DALL-E can create visually appealing artwork or conceptual sketches, it is not designed for producing precise or technically correct diagrams, such as engineering schematics or architectural blueprints. The model’s generative process emphasizes creativity and visual diversity rather than factual or geometric accuracy. Thus, it cannot be relied upon for professional technical outputs.

“Generate decorative images to enhance learning materials” → YesThis is one of DALL-E’s strongest use cases. It can generate decorative, conceptual, or illustrative images to enhance presentations, educational materials, and marketing content. It enables educators and designers to quickly produce unique visuals aligned with specific themes or topics, enhancing engagement and creativity.

“When evaluating the performance of a model, the confusion matrix displays the predicted and actual positives and negatives by using a grid of 0 and 1 values.”

According to the Microsoft Azure AI Fundamentals (AI-900) official study materials and Microsoft Learn module “Identify features of common machine learning types”, a confusion matrix is a tool used to evaluate the performance of classification models. It visually summarizes how many predictions were correctly or incorrectly classified by comparing the predicted labels to the actual (true) labels.

A confusion matrix is a table, typically 2×2 for binary classification, with the following components:

True Positives (TP): The model correctly predicted the positive class.

True Negatives (TN): The model correctly predicted the negative class.

False Positives (FP): The model incorrectly predicted the positive class.

False Negatives (FN): The model incorrectly predicted the negative class.

The confusion matrix allows data scientists and analysts to derive important performance metrics such as accuracy, precision, recall, and F1-score, which together provide a more complete understanding of how well a model performs beyond a single number.

In Microsoft Learn’s AI-900 curriculum, the confusion matrix is highlighted as a key visualization tool that “compares actual values to predicted values to evaluate classification performance.” The grid format (using 0s and 1s for predicted classes) helps identify where misclassifications occur.

By contrast:

AUC metric (Area Under Curve) and ROC curve evaluate model discrimination ability.

Threshold defines decision cutoffs but doesn’t display classifications.

Therefore, based on the official Microsoft AI-900 study guide and Microsoft Learn resources, the correct answer is Confusion Matrix, as it provides a grid view comparing actual versus predicted values in classification models.

Generative AI focuses on creating new content rather than just analyzing existing data. As per Microsoft’s AI-900 curriculum and Azure OpenAI documentation, typical use cases include generating text, images, code, or other creative outputs based on input prompts.

A. Generating draft responses for customer service agents — ✅ Correct. GPT-based models can automatically generate draft replies to customer queries, enabling agents to refine responses and increase efficiency.

B. Creating original artwork from textual descriptions — ✅ Correct. DALL-E, available through Azure OpenAI, can produce unique images based on natural language prompts.

Options C and D are incorrect because they involve predictive or classification models, not generative ones:

C. Predicting sales revenue → Regression (machine learning).

D. Classifying email messages → Classification (machine learning).

Correct answers: A and B.

The Azure AI Language Service includes several natural language processing features, such as question answering, language understanding, entity recognition, sentiment analysis, and more. Each feature serves a distinct purpose, and understanding their differences is key to selecting the correct AI workload.

“You can use Azure AI Language Service ' s question answering to query an Azure SQL database.” — NOThe question answering feature is designed to retrieve answers from text-based knowledge sources (for example, FAQs, documents, or website content). It cannot directly query a database such as Azure SQL. Querying databases requires Azure Cognitive Search, Azure OpenAI, or custom integration using application logic, not the question answering model.

“You should use Azure AI Language Service ' s question answering when you want a knowledge base to provide the same answer to different users who submit similar questions.” — YESThis is the primary use case of question answering. It allows developers to build a knowledge base (KB) of predefined question-answer pairs or extract answers from documents. When users submit semantically similar questions (e.g., “What are your office hours?” or “When are you open?”), the service returns the same consistent answer.

“Azure AI Language Service ' s question answering can determine the intent of a user utterance.” — NODetermining user intent is part of the Language Understanding (LUIS) capability, not question answering. LUIS models map natural language inputs to intents and entities, typically used in bots or applications that execute tasks (like booking a meeting or checking weather).

Hence, correct answers are: No, Yes, No — aligning with the AI-900 official study guide and Microsoft Learn module “Identify Azure AI Language capabilities.”

According to the Microsoft Azure AI Fundamentals (AI-900) official study materials and the Microsoft Learn module “Describe features of Natural Language Processing (NLP) workloads on Azure”, this scenario combines two major capabilities of AI: speech recognition and natural language understanding.

Speech-to-Text (E) – This is the first step in processing verbal commands. The Azure Speech service converts the spoken words of a user into textual data that can be understood and processed by downstream components. This workload is commonly referred to as speech recognition, and it falls under the speech capabilities of Azure Cognitive Services. Without this transcription process, the system could not interpret the user’s voice input.

Language Modeling (C) – After the speech input is converted into text, the next step is to interpret the meaning of the text so the system can take appropriate action. Language modeling, also known as language understanding, is responsible for identifying the user’s intent (for example, “turn on the lights” or “set the thermostat to 72 degrees”) and extracting entities (such as device name or temperature value). In Azure, this function is handled by Language Understanding (LUIS) or Conversational Language Understanding (CLU). These models allow smart systems to process commands and map them to defined actions.

Other options are not correct:

A. Text-to-speech converts text output into spoken language, which is not mentioned as a requirement.

B. Translation converts text from one language to another, irrelevant to this scenario.

D. Key phrase extraction identifies important terms in text but doesn’t interpret or execute commands.

Therefore, the solution uses speech-to-text to transcribe verbal commands and language modeling to understand and act upon them — the two key NLP workloads enabling voice-controlled smart devices.

According to the Microsoft Azure AI Fundamentals (AI-900) official study guide and the Microsoft Learn module “Identify features of the Computer Vision and Custom Vision services”, the Custom Vision service is used to train, deploy, and improve custom image classification and object detection models using your own labeled data.

Multilabel or Multiclass Selection – NOThe statement is false because the Multilabel or Multiclass choice applies only to image classification models, not object detection models. In image classification, “Multiclass” means one label per image, while “Multilabel” means multiple labels per image. In contrast, object detection models identify and locate multiple objects in an image using bounding boxes; thus, this classification-type selection does not apply.

Object Detection Locates Content in an Image – YESThis statement is true. The object detection functionality in Custom Vision is designed to both identify what objects appear in an image and determine their location through bounding box coordinates. For example, a model could detect and locate multiple products on a store shelf. Microsoft documentation describes object detection as “identifying the presence and location of objects in an image.”

Predefined Domains – YESThis statement is true as well. When you create a new Custom Vision project, you must select a domain, which is a predefined optimization setting tailored to specific use cases such as retail, food, landmarks, or general images. These domains are designed to improve model accuracy by applying specialized transfer learning features based on the type of images you will analyze.

In summary:

Classification type (Multilabel/Multiclass): No (only for classification models)

Detect object location: Yes

Choose predefined domain: Yes

The correct matches are based on the Microsoft Azure AI Fundamentals (AI-900) Official Study Guide and the Microsoft Learn module “Explore computer vision in Microsoft Azure.” These materials explain that facial recognition tasks can be categorized into four major operations: verification, identification, similarity, and grouping. Each task serves a distinct purpose in facial recognition scenarios.

Verification – “Do two images of a face belong to the same person?”The verification task determines whether two facial images represent the same individual. Azure Face API compares the facial features and returns a confidence score indicating the likelihood that the two faces belong to the same person.

Similarity – “Does this person look like other people?”The similarity task compares a face against a collection of faces to find visually similar individuals. It does not confirm identity but measures how closely two or more faces resemble each other.

Grouping – “Do all the faces belong together?”Grouping organizes a set of unknown faces into clusters based on similar facial features. This is used when identities are not known beforehand, helping discover potential duplicates or visually similar clusters within an image dataset.

Identification – “Who is this person in this group of people?”The identification task is used when the system tries to determine who a specific person is by comparing their face against a known collection (face database or gallery). It returns the identity that best matches the input face.

According to Microsoft’s AI-900 training, these tasks form the basis of Azure Face API’s capabilities. Each helps solve a different type of facial recognition problem—from matching pairs to discovering unknown identities—making them essential components of responsible AI-based vision systems.