Browse Certification Practice Tests by Exam Family

AI-901: Identify AI Concepts and Capabilities

Try 10 focused AI-901 questions on Identify AI Concepts and Capabilities, with explanations, then continue with IT Mastery.

Open the matching IT Mastery practice page for timed mocks, topic drills, progress tracking, explanations, and full practice.

Try AI-901 on Web View full AI-901 practice page

Topic snapshot

FieldDetail
Exam routeAI-901
Topic areaIdentify AI Concepts and Capabilities
Blueprint weight43%
Page purposeFocused sample questions before returning to mixed practice

How to use this topic drill

Use this page to isolate Identify AI Concepts and Capabilities for AI-901. Work through the 10 questions first, then review the explanations and return to mixed practice in IT Mastery.

PassWhat to doWhat to record
First attemptAnswer without checking the explanation first.The fact, rule, calculation, or judgment point that controlled your answer.
ReviewRead the explanation even when you were correct.Why the best answer is stronger than the closest distractor.
RepairRepeat only missed or uncertain items after a short break.The pattern behind misses, not the answer letter.
TransferReturn to mixed practice once the topic feels stable.Whether the same skill holds up when the topic is no longer obvious.

Blueprint context: 43% of the practice outline. A focused topic score can overstate readiness if you recognize the pattern too quickly, so use it as repair work before timed mixed sets.

Sample questions

These questions are original IT Mastery practice items aligned to this topic area. They are designed for self-assessment and are not official exam questions.

Question 1

Topic: Identify AI Concepts and Capabilities

A team is building a lightweight help-desk chat feature. It will use a generative AI model deployed from Microsoft Foundry, and a web app must collect employee questions, send prompts, and show the replies. Which description best distinguishes the model from the application?

Options:

  • A. The web app generates replies; the deployed model stores the chat history.

  • B. The deployed model generates replies; the web app sends prompts and displays responses.

  • C. The Foundry portal answers users directly; the model formats the user interface.

  • D. The system prompt displays replies; the web app trains the model.

Best answer: B

Explanation: In a generative AI solution, the deployed model is the component that produces new content from prompts. The application is the client that gathers user input, adds any needed instructions such as a system prompt, sends the request to the model deployment, and displays the model’s response. The app may also manage conversation flow or UI behavior, but it is not the content-generating model. The key distinction is generation versus orchestration and presentation.

  • Reversed roles fails because the web app calls the model; it does not generate the AI response by itself.
  • Prompt as component fails because a system prompt is an instruction, not the runtime component that displays replies.
  • Portal as chatbot fails because Foundry helps deploy and test models, but the user-facing app sends prompts and shows responses.

Question 2

Topic: Identify AI Concepts and Capabilities

A developer is building a lightweight Python chat client that will send user prompts to a generative AI model in Microsoft Foundry. Which need maps to the principle that a model deployment must exist first?

Options:

  • A. Reading responsible AI documentation

  • B. Comparing model descriptions in the catalog

  • C. Calling the model from application code

  • D. Writing a user prompt in the chat UI

Best answer: C

Explanation: In Microsoft Foundry, a deployment makes a selected model available for use by applications. A lightweight client does not call an undeployed catalog model directly; it sends requests to a specific deployed model endpoint or deployment name. Exploring models, drafting prompts, or reviewing guidance can happen before deployment, but application code needs a deployed target to invoke.

  • Prompt drafting can be done conceptually or in a portal experience, but it is not the same as application code invoking a model.
  • Catalog comparison helps choose a model, but browsing available models does not create a callable endpoint.
  • Documentation review supports responsible use, but it does not provide the runtime target a client needs.

Question 3

Topic: Identify AI Concepts and Capabilities

A generative AI assistant summarizes insurance policy clauses. Some source documents may be outdated, and the model may infer missing context. The product team wants each summary to state uncertainty, identify source limitations, and flag cases that require human review. Which responsible AI concept is this primarily addressing?

Options:

  • A. Privacy and security

  • B. Transparency and user disclosure

  • C. Inclusiveness

  • D. Fairness

Best answer: B

Explanation: Transparency and user disclosure focus on helping people understand AI system behavior and limitations. In this scenario, the key need is not to make the summary sound more confident, but to tell users when the output may be incomplete, uncertain, or dependent on limited sources. That kind of disclosure reduces over-reliance by signaling when human judgment or review is needed. Accountability may involve assigning review responsibility, but the communication of uncertainty and limitations is primarily a transparency practice.

  • Fairness is about reducing unfair treatment or biased outcomes, not mainly disclosing uncertainty in a summary.
  • Privacy and security protects data and access, but the stem focuses on interpreting output safely.
  • Inclusiveness aims to support diverse users and accessibility, not primarily source-limit disclosure.

Question 4

Topic: Identify AI Concepts and Capabilities

A retail company wants an AI feature that reads customer review comments and identifies the overall sentiment, key phrases, and product names mentioned in each review. Which primary AI workload does this scenario describe?

Options:

  • A. Computer vision

  • B. Image generation

  • C. Speech

  • D. Text analysis

Best answer: D

Explanation: Text analysis is the primary workload when an AI solution processes written language to find meaning, patterns, or labels. In this scenario, the input is customer review text, and the desired outputs are sentiment, key phrases, and product names. Those are common text analysis tasks because they interpret existing text rather than create images, understand visual content, or process audio. The key clue is that the source data and the results are both centered on language understanding.

  • Speech is not the best fit because no spoken audio, transcription, or voice output is involved.
  • Computer vision fails because the scenario does not analyze images or video frames.
  • Image generation fails because the solution is not creating new visual content from a prompt.

Question 5

Topic: Identify AI Concepts and Capabilities

A team is working in Microsoft Foundry. They have already chosen a generative AI model that supports their chat scenario. Which action best represents deploying the model rather than selecting it?

Options:

  • A. Choosing a model based on task fit

  • B. Creating an endpoint for test and app calls

  • C. Reviewing model descriptions and capabilities

  • D. Comparing models by supported modalities

Best answer: B

Explanation: Selecting a model means deciding which available model best fits the workload, such as choosing a chat-capable or multimodal model based on the task. Deploying a model is the next implementation step: making that selected model available so it can be tested in Foundry or called by a lightweight application. In fundamentals terms, deployment is about creating a usable model instance or endpoint, not comparing model families or reading capability descriptions. The key distinction is decision versus consumption readiness.

  • Capability comparison is part of model selection because it evaluates what different models can do.
  • Task-fit choice describes selecting the appropriate model before it is made callable.
  • Model review supports selection, but it does not make the model available to an application.

Question 6

Topic: Identify AI Concepts and Capabilities

A team is testing a Microsoft Foundry chat app before release. During a test, the app returns a table that appears to include confidential employee compensation data. The team needs to diagnose the issue, minimize further exposure, and avoid deleting useful incident evidence. What is the best handling approach?

Options:

  • A. Lower the model temperature and continue testing

  • B. Delete the transcript and remove the test prompt

  • C. Add a disclaimer to warn users about sensitive output

  • D. Restrict the output and escalate for access review

Best answer: D

Explanation: When an AI system reveals information that should not be broadly visible, the safer handling approach is to reduce exposure and involve the appropriate owner rather than tuning around the symptom. The team should restrict access to the output or transcript, preserve only the evidence needed for investigation, and escalate to the data owner, security team, or responsible AI process. The likely root cause may be overly broad permissions, inappropriate grounding data, or missing controls around what the app can retrieve and display. Changing generation settings or adding a warning does not address unauthorized disclosure. The key takeaway is to treat sensitive AI output as a privacy and security incident, not just a prompt-quality issue.

  • Temperature tuning affects response variability, not whether confidential data is authorized to appear.
  • User warnings may improve transparency but do not prevent disclosure of protected information.
  • Deleting evidence can hinder investigation and does not fix the underlying data-access problem.

Question 7

Topic: Identify AI Concepts and Capabilities

A bank uses one automated loan-screening model for all applicants. The team requires the same scoring rule for every person, uses past approval data as input, and wants to understand why complaints of unfair outcomes may still be valid. Which explanation best fits this situation?

Options:

  • A. The model needs a larger compute deployment.

  • B. The issue is only transparency, not fairness.

  • C. The model must use separate rules for each group.

  • D. The training data may reflect biased past decisions.

Best answer: D

Explanation: Fairness in AI is not guaranteed by applying the same automated rule to everyone. If the model learns from biased historical approvals, incomplete inputs, or output labels that reflect unfair decisions, it can treat applicants identically in process while still producing unequal or unjust results. The problem is in what the system learned and what outcomes it optimizes, not necessarily in whether the same scoring code runs for each applicant.

The key takeaway is that fairness requires checking inputs, outputs, and impacts, not only confirming that the automation is uniform.

  • Separate rules is not the best explanation because different rules by group are not required to identify biased outcomes.
  • Transparency only fails because explainability may help investigate the model, but the complaint is about unfair impact.
  • Compute deployment is unrelated because more capacity does not correct biased data or biased labels.

Question 8

Topic: Identify AI Concepts and Capabilities

A city agency uses Azure Content Understanding in Foundry Tools to extract fields from uploaded inspection videos. The result must include only defects that are visible or audible in the recording, avoid assumptions about causes, and flag uncertain findings for review. Which approach is the best fit?

Options:

  • A. Generate a complete repair plan for every detected defect

  • B. Infer likely defect causes from the building age and neighborhood

  • C. Use image generation to recreate missing inspection views

  • D. Extract observed defects with evidence and confidence, then review low-confidence items

Best answer: D

Explanation: Extraction results should be grounded in the source content. For video, that means details that are visible in frames, audible in the audio, or otherwise provided by the uploaded source. If a defect is unclear, the safer design is to include evidence such as timestamps or extracted snippets, use confidence indicators when available, and route uncertain items to human review. The system should not add likely causes, missing views, or downstream plans unless those details are explicitly present in the content or supplied by another trusted source. The key takeaway is that extraction identifies information from content; it should not invent or infer facts beyond that content.

  • Cause inference fails because building age and neighborhood can add assumptions not present in the inspection video.
  • Repair planning goes beyond extraction by creating recommendations that may not be stated or shown in the source.
  • Recreated views are generated content, not validated evidence from the uploaded recording.

Question 9

Topic: Identify AI Concepts and Capabilities

A city services team wants to analyze resident feedback messages. The solution must find references to named parks, departments, dates, phone numbers, and monetary amounts so staff can route and summarize issues. Which text analysis capability is the best fit?

Options:

  • A. Sentiment analysis

  • B. Language detection

  • C. Entity detection

  • D. Key phrase extraction

Best answer: C

Explanation: Entity detection is the text analysis capability used to identify and label specific references in unstructured text, such as locations, organizations, dates, phone numbers, and quantities. In this scenario, the team needs to locate structured references inside feedback messages so the information can support routing and summarization. That requirement is different from determining emotional tone, extracting general topics, or identifying the language of the text. The key signal is the need to find named or typed references rather than classify the overall message.

  • Sentiment analysis fails because it assesses positive, negative, or neutral tone rather than locating named references.
  • Key phrase extraction fails because it returns important phrases or topics, not typed entities such as dates or money amounts.
  • Language detection fails because it identifies the text language, not parks, departments, dates, or quantities.

Question 10

Topic: Identify AI Concepts and Capabilities

A developer is planning a vision feature for a home-assistance app. Users will upload a photo of a room and ask questions such as, “Is there a clear path from the door to the desk?” Which vision capability best matches this need?

Options:

  • A. Image generation

  • B. Visual question answering

  • C. Image classification

  • D. Object detection only

Best answer: B

Explanation: Visual question answering maps a user’s question to information found in an image. In this scenario, the app must interpret the room layout and answer whether a path is clear, not just label the whole image or locate one object. A multimodal vision-capable model can combine the image input with the user’s text question to produce an image-based answer.

Object detection may help identify items, but by itself it does not answer the user’s specific question about the scene.

  • Image generation fails because the user needs interpretation of an existing photo, not creation of a new image.
  • Object detection only is too narrow because locating objects does not directly answer the path question.
  • Image classification is too broad because assigning a category to the whole image does not provide question-specific assistance.

Continue with full practice

Use the AI-901 Practice Test page for the full IT Mastery practice bank, mixed-topic practice, timed mock exams, explanations, and web/mobile app access.

Try AI-901 on Web View AI-901 Practice Test

Free review resource

Read the AI-901 Cheat Sheet for compact concept review before returning to timed practice.

Revised on Monday, May 25, 2026