AI-900: AI Workloads

Topic snapshot

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: Describe Artificial Intelligence Workloads and Considerations

A team is comparing a generative AI solution in Azure OpenAI Service with a classification-only model for customer messages. Which capability is the main value proposition of the generative AI workload?

Options:

• A. Group similar messages without labeled examples

• B. Predict a numeric satisfaction score for each message

• C. Assign each message to a predefined category

• D. Generate new draft replies and summaries from prompts

Best answer: D

Explanation: Generative AI is valuable because it can produce new content, such as summaries or draft replies, from prompts and context. A classification-only workload is limited to assigning existing labels, so it cannot create original output for the user.

The core difference is the type of output produced. A classification workload maps an input to one of several predefined labels, such as billing, support, or sales. A generative AI workload can create new content based on a prompt and optional context, such as a summary, answer, or draft reply. That content-generation ability is the main value proposition when a business needs synthesized text rather than simple labeling.

Predicting a numeric value is a regression task, and grouping similar items without labels is a clustering task. The key takeaway is that generative AI creates new outputs, while classification chooses from existing categories.

• The option about predefined categories describes classification, which labels data but does not generate new content.
• The option about a numeric satisfaction score describes regression, which predicts continuous values.
• The option about grouping similar messages describes clustering, an unsupervised learning task rather than a generative one.

Question 2

Topic: Describe Artificial Intelligence Workloads and Considerations

A warehouse wants an AI solution that can examine a photo, identify each forklift and pallet, and show where each one appears in the image. Which AI workload type best fits this requirement?

Options:

• A. Optical character recognition (OCR)

• B. Object detection

• C. Sentiment analysis

• D. Image tagging

Best answer: B

Explanation: The key requirement is to find specific items and indicate where they are in the photo. That makes this an object detection task, which is a computer vision workload focused on identifying objects and their positions within an image.

Object detection is the best fit when an image can contain multiple items and the solution must both recognize them and locate them. In this case, the warehouse needs to identify forklifts and pallets and show where each appears in the photo, which is commonly represented with bounding boxes. That is more specific than simply labeling the overall image.

Image tagging adds descriptive labels to an image, but it usually does not identify the exact position of each object. OCR is used to extract text from images, and sentiment analysis evaluates opinion in text. The deciding clue is the need for both object identity and object location.

• Image tagging can label a photo with terms like “forklift” or “warehouse,” but it typically does not return each object’s position.
• OCR is for reading printed or handwritten text from images, not finding pallets or forklifts.
• Sentiment analysis is an NLP task used to assess opinion in text rather than analyze object placement in photos.

Question 3

Topic: Describe Artificial Intelligence Workloads and Considerations

A logistics company wants a mobile app that reads tracking numbers and delivery addresses from photos of package labels and stores the text in a database. Which computer vision workload should the company use?

Options:

• A. Image classification

• B. Optical character recognition (OCR)

• C. Facial analysis

• D. Object detection

Best answer: B

Explanation: The requirement is to extract text from an image, not to label the entire image or find general objects. OCR is the computer vision workload designed to read characters such as tracking numbers and addresses from photos or scanned documents.

The key clue is that the app must read text from package-label photos and convert it into usable data. That is an OCR workload. OCR detects characters in an image and returns machine-readable text, which is commonly used for labels, receipts, forms, and scanned documents.

Image classification assigns a label to the whole image, such as “package label” or “box.” Object detection identifies and locates items within an image, such as a package or barcode region, but it does not by itself extract the written text. Facial analysis is used for face-related scenarios, not document text. When the main goal is reading words or numbers from an image, OCR is the best match.

• Whole-image labeling Image classification predicts a category for the overall image rather than extracting the text inside it.
• Finding items Object detection can locate regions or objects, but the scenario requires reading characters from the label.
• Face-focused analysis Facial analysis applies to human faces and is unrelated to package-label text extraction.

Question 4

Topic: Describe Artificial Intelligence Workloads and Considerations

A company stores scanned invoices as PDF files and wants to extract fields such as vendor name, invoice number, and total amount from documents that follow similar layouts. Which AI workload is the best fit for this requirement?

Options:

• A. Document processing

• B. Face detection

• C. Object detection

• D. Image classification

Best answer: A

Explanation: Document processing is intended for documents such as invoices, receipts, and forms. It extracts text and understands layout elements like fields and tables, making it the best match for structured or semi-structured files in business workflows.

The key distinction is that document processing treats the input as a document, not just as a picture. For invoices, forms, and receipts, the goal is usually to extract text together with its structure, such as key-value pairs, tables, and line items. That is different from general computer vision tasks, which focus on understanding image content such as objects, scenes, or categories. A general vision model might detect regions or classify the whole page, but it is not primarily designed to return business fields like invoice number or total. When the requirement is to read structured or semi-structured files and capture named fields, document processing is the appropriate workload.

• Image classification labels an entire image or document page but does not extract specific fields.
• Object detection can locate regions on a page, but it is not the primary workload for reading key-value pairs and tables.
• Face detection is used for finding human faces in images, which is unrelated to invoice field extraction.

Question 5

Topic: Describe Artificial Intelligence Workloads and Considerations

Which responsible AI principle emphasizes human oversight and clear responsibility for the outcomes of an AI system?

Options:

• A. Accountability

• B. Transparency

• C. Reliability and safety

• D. Fairness

Best answer: A

Explanation: Accountability is the responsible AI principle that requires people and organizations to remain answerable for what an AI system does. It includes human oversight, governance, and clear ownership of decisions and results.

Accountability means AI systems should not operate without clear human and organizational responsibility. Even when an AI model makes recommendations or automates part of a process, people must be able to oversee its use, review important outcomes, and take responsibility when something goes wrong. In practice, this principle supports assigning owners, defining approval paths, and keeping humans involved for impactful decisions.

This differs from other responsible AI principles that focus on understanding the system, avoiding biased treatment, or making sure it works safely. The key idea is that responsibility for AI outcomes always remains with humans and the organization using the system.

• Transparency is about making AI behavior and limitations understandable, not about who is responsible for outcomes.
• Fairness is about avoiding unjust bias and treating people equitably across groups.
• Reliability and safety focuses on consistent, dependable, and safe system behavior under expected conditions.

Question 6

Topic: Describe Artificial Intelligence Workloads and Considerations

A car rental company is automating accident claims. Customers submit a damage photo, a photo of the rental agreement, and an optional voice note. The company wants AI to read the agreement image and extract the contract number, renter name, and return date. Which AI workload is the primary workload for this requirement?

Options:

• A. Image classification

• B. Document processing

• C. Speech recognition

• D. Sentiment analysis

Best answer: B

Explanation: The overall claims process could involve several AI workloads, but the stated requirement is to pull named values from a rental agreement image. That makes document processing the primary workload because the system must read and interpret document content.

A single business process can include more than one AI workload, so the key is to identify the exact task being asked about. Here, the task is to read a document image and extract structured fields such as contract number, renter name, and return date. That is document processing because the AI must recognize text and understand where important values appear in the document.

The damage photo and voice note suggest other possible workloads in the same workflow, but they are secondary to this specific requirement. A general image workload would analyze the photo itself, while document processing is aimed at forms, receipts, agreements, and similar files where text and fields must be captured.

• Image classification would fit analyzing the damage photo, but it does not extract named values from a document.
• Speech recognition would apply to the optional voice note, not the agreement image.
• Sentiment analysis evaluates tone or opinion in text, which is unrelated to capturing contract details.

Question 7

Topic: Describe Artificial Intelligence Workloads and Considerations

A retailer has a multilingual customer-support center. It needs a prebuilt Azure AI solution that can transcribe live phone calls and translate the spoken conversation between English and Spanish for agents. Which Azure AI service family is the best fit?

Options:

• A. Azure AI Language

• B. Azure AI Vision

• C. Azure AI Speech

• D. Azure OpenAI Service

Best answer: C

Explanation: Azure AI Speech is the best fit because the input is live spoken audio, and the company needs both transcription and translation. Those are core speech AI capabilities available as prebuilt services.

The key concept is matching the AI workload to the input type and task. Here, the company is working with live phone audio and wants to turn speech into text and translate the spoken conversation between languages.

• Speech-to-text is a speech workload.
• Translating spoken audio is also a speech workload.
• A prebuilt service means the company does not need to train a custom model.

Azure AI Language is mainly for understanding and analyzing text that already exists as text. Azure AI Vision focuses on images and video. Azure OpenAI Service is for generative tasks such as creating, summarizing, or transforming content, not for core live call transcription. When the requirement starts with spoken audio, Azure AI Speech is usually the right choice.

• Text analysis fits tasks like sentiment analysis or entity extraction on text, but it does not directly handle live call transcription.
• Image processing is used for photos, scanned documents, and video, not spoken conversations.
• Generative output can help create or summarize content, but it is not the primary service for speech transcription and speech translation.

Question 8

Topic: Describe Artificial Intelligence Workloads and Considerations

Contoso is reviewing four planned Azure AI solutions before release. Which scenario most directly raises a transparency concern?

Options:

• A. A kiosk that uses Azure AI Face detection service has lower accuracy for people with darker skin tones.

• B. A loan screening app uses Azure Machine Learning, but applicants are not told AI is used and staff cannot explain the score.

• C. An Azure AI Language sentiment analysis solution stores identifiable customer comments without adequate safeguards.

• D. An Azure OpenAI Service chatbot can issue refunds, but no team is assigned to review harmful decisions.

Best answer: B

Explanation: Transparency is about making AI use visible and understandable to affected users. A screening system that hides the use of AI and cannot provide a reason for its score most directly violates that principle.

Transparency in responsible AI means people should know when AI is being used and should be able to understand the basis for an output, especially when the output affects them. In the loan screening scenario, both parts are missing: applicants are not told that AI is involved, and employees cannot explain how the score was produced. That makes the system harder to trust, question, or appropriately use.

The other scenarios point to different responsible AI principles. Uneven performance across demographic groups is mainly a fairness issue. Weak protection of identifiable customer data is a privacy and security issue. Lack of assigned human ownership for reviewing harmful outcomes is an accountability issue. The key takeaway is that hidden AI use and unexplained outputs are classic transparency concerns.

• Bias across groups lower accuracy for people with darker skin tones is primarily a fairness concern.
• Weak data protection storing identifiable comments without safeguards is a privacy and security concern.
• No clear owner failing to assign human review responsibility is an accountability concern.

Question 9

Topic: Describe Artificial Intelligence Workloads and Considerations

A company uses an AI model in Azure Machine Learning to rank job applicants. Reviewers find that candidates with similar qualifications receive different scores, and applicants from one demographic group are consistently ranked lower. Which responsible AI principle is most directly implicated?

Options:

• A. Privacy and security

• B. Fairness

• C. Transparency

• D. Inclusiveness

Best answer: B

Explanation: The scenario shows unequal outcomes for similarly qualified applicants, especially across a demographic group, which is a fairness concern. In responsible AI, fairness means an AI system should avoid biased treatment and unjustified disadvantage.

Fairness is the responsible AI principle most directly related to equitable treatment and avoiding harmful bias. In a hiring scenario, if people with similar qualifications receive different outcomes and one demographic group is consistently ranked lower, the clearest concern is that the model is producing unfair results. That is exactly what fairness is meant to address.

Transparency is about understanding how a model reaches its decisions. Inclusiveness is about designing AI systems that can be used effectively by people with diverse needs and abilities. Privacy and security are about protecting personal data and controlling access to it. Those principles can still matter, but they are not the main issue described here. The key takeaway is that unequal model outcomes across groups point first to fairness.

• Transparency would fit a scenario about explaining how scores were produced, but this scenario centers on biased outcomes.
• Inclusiveness is about designing systems for broad and accessible use, not about whether one group is scored unfairly.
• Privacy and security concerns protecting applicant data, not whether the ranking results disadvantage a demographic group.

Question 10

Topic: Describe Artificial Intelligence Workloads and Considerations

A company uses Azure AI Language to classify incoming customer emails as high, medium, or low priority. Which action best demonstrates the responsible AI principle of transparency?

Options:

• A. Encrypt email records and restrict access to administrators

• B. Explain the model’s purpose, confidence scores, and escalation rules to agents

• C. Compare error rates across customer groups before deployment

• D. Assign one team to approve updates and handle incidents

Best answer: B

Explanation: Transparency is about helping stakeholders understand what an AI system does, how to use its outputs, and where its limits are. Explaining the classifier’s purpose, confidence scores, and escalation rules directly supports informed use by the people relying on it.

Transparency in responsible AI means making an AI system understandable to the people who use it or are affected by it. In this scenario, support agents need to know what the email classifier is designed to do, how confident it is in a result, and when they should override or escalate a decision. That helps them use the system appropriately instead of treating it like an unexplained black box.

Privacy and security focus on protecting data. Fairness focuses on whether the system performs equitably across groups. Accountability focuses on who is responsible for oversight and decisions. The key takeaway is that transparency is about clear communication of AI behavior, limitations, and proper usage.

• Encrypting records and limiting access addresses privacy and security, not making the model understandable.
• Comparing error rates across groups addresses fairness by checking for unequal impact.
• Assigning a team to approve updates and handle incidents addresses accountability through human oversight.

Continue with full practice

Use the AI-900 Practice Test page for the full IT Mastery route, mixed-topic practice, timed mock exams, explanations, and web/mobile app access.

Try AI-900 on Web View AI-900 Practice Test

Related focused pages

• Free AI-900 Full-Length Practice Exam
• Machine Learning
• Computer Vision
• NLP Workloads
• Generative AI

Free review resource

Read the AI-900 Cheat Sheet on Tech Exam Lexicon, then return to IT Mastery for timed practice.