AI Workloads & Scenarios

A supermarket wants to use live camera feeds to monitor stock levels on its shelves and to identify items at self-checkout. Which AI workload is this?

• AText analysis
• BComputer vision✓
• CInformation extraction
• DSpeech recognition

Answer: Computer vision is the area of AI that deals with analysing visual input such as photographs, videos and live camera feeds, and monitoring stock levels and identifying items for checkout in retail are among its listed uses. Speech recognition transcribes spoken audio, text analysis makes sense of written language, and information extraction pulls fields and insights out of documents, forms and recordings.

A consumer brand wants to analyse thousands of social media posts about its products to determine whether each one expresses a positive, negative or neutral opinion. Which text analysis technique is this?

• ASentiment analysis✓
• BLanguage detection
• CKey-term extraction
• DSummarisation

Answer: Sentiment analysis is a form of text classification that determines whether a body of text is positive, negative or neutral, and analysing social media posts, product reviews or articles for sentiment and opinion is a common use. Key-term extraction finds important words and phrases, language detection identifies which language a document is written in, and summarisation reduces the volume of text while keeping its main points.

A law firm must remove names, addresses and telephone numbers from client documents before sharing them with an outside analytics partner. Which text analysis capability is needed?

• ASummarisation that reduces each document to its main points
• BText classification that assigns each document to a sensitivity category
• CEntity detection that detects and redacts personally identifiable information✓
• DLanguage detection that flags documents written in more than one language

Answer: Detecting and redacting personally identifiable information such as names, addresses and telephone numbers is a particularly specialised form of entity detection, and redacting PII before sharing or analysing data is one of the standard text analysis scenarios. Classifying documents by category, detecting their language or summarising them would not remove the private details from the text.

A hotel chain wants every recorded customer-service call transcribed into text automatically so that it can be searched later. Which AI capability is needed?

• ASpeech synthesis
• BSemantic segmentation
• CKey-term extraction
• DSpeech recognition✓

Answer: Speech recognition is the ability of AI to hear and interpret speech, usually as speech-to-text in which the audio signal is transcribed into text, and automated transcription of calls or meetings is a common speech scenario. Speech synthesis goes the other way by vocalising text, semantic segmentation is a computer vision technique, and key-term extraction analyses text that already exists.

A finance app must read a photographed receipt and match the values on it to the fields of an expense claim. Which combination of technologies forms the basis of this kind of document analysis?

• AImage classification to identify the receipt type, combined with a language model that writes a natural-language description of it
• BOptical character recognition to locate text in the image, combined with an analytical model that interprets the individual values✓
• CSpeech recognition to read the receipt aloud, combined with summarisation to condense the transcript into the required fields
• DObject detection to draw a box around each item on the receipt, combined with semantic segmentation to isolate the pixels of every printed line

Answer: Optical character recognition (OCR) is a computer vision technology that can identify the location of text in an image, and it is the basis for most document analysis solutions. OCR is often combined with an analytical model that interprets individual values so that specific fields can be extracted, with matching receipt text to expense-claim fields being the classic example. Classifying or describing the receipt, reading it aloud, or boxing and segmenting its pixels would not extract the field values.

A company wants a digital assistant that can not only answer questions in natural language but also send emails and update calendars on the user's behalf by using tools. Which kind of solution is this?

• AA text analysis solution
• BA generative AI chat bot
• CA speech synthesis solution
• DAn agentic AI solution✓

Answer: Agents are software applications built on generative AI that can reason over and generate natural language, automate tasks by using tools and respond to contextual conditions, and sending emails or updating calendars are typical action tools. A generative AI chat bot answers questions or holds a conversation but does not act through tools, and text analysis and speech synthesis are single capabilities rather than task-automating assistants.

An AI agent is given a tool that lets it query a product database to look up stock information before answering a customer. Which type of agent tool is this?

• AA knowledge tool✓
• BAn action tool
• CA system prompt
• DA large language model

Answer: Knowledge tools provide an agent with access to information, such as search engines or databases, so a database lookup is a knowledge tool. Action tools instead enable the agent to perform tasks such as sending emails, updating calendars or controlling devices. The system prompt is the agent's instructions defining its role, and the large language model is the agent's brain that reasons over language.

Generative AI models can handle many natural language tasks today, yet specialist NLP tools are still used for some text analysis use cases. Why?

• ATo generate original documents as a starting point for editing
• BTo produce predictable results or to apply custom rules✓
• CTo handle images, audio and video inside chat completions
• DTo reason over tools and take actions in response to conditions

Answer: Specialist NLP tools remain in use for common text analysis cases because they produce predictable results or apply custom rules, in contrast to the open-ended generative approach. A typical example is a chatbot that answers frequently asked questions or orchestrates predictable dialogues without the complexity of generative AI. Creating original content is a generative AI use, acting through tools describes agents, and multimodal chat describes models that accept images, audio and video.

A support desk receives written tickets from customers around the world and must route each one to a team that reads the right language. Which text analysis technique should run first in the workflow?

• ALanguage detection✓
• BSummarisation
• CEntity detection
• DText classification

Answer: Language detection determines which language or languages a document is written in, and it is often the first step in a multi-stage text processing workflow. Only once the language is known can the ticket be routed to the right team. Text classification, summarisation and entity detection are later analysis steps that operate on text whose language is already established.

A computing history website wants visitors to type questions about key figures and events and receive original, conversational answers generated for each query. Which AI workload is this?

• ASpeech synthesis
• BComputer vision
• CGenerative AI✓
• DText classification

Answer: Generative AI is the branch of AI that enables applications to generate new content, often natural language dialogues, and a computing history site with a chat interface that generates original responses to visitors' questions is Microsoft's own example. Text classification assigns existing text to categories, computer vision analyses visual input, and speech synthesis converts text into spoken audio.