Microsoft Azure AI Fundamentals Practice Test (AI-900)

Utilize the text generation capabilities of Azure OpenAI Service - This is the correct answer. The text generation capabilities of Azure OpenAI Service, such as models like GPT, are ideal for building a chatbot that provides human-like answers to customer queries. These models can generate coherent, contextually relevant responses based on the input provided.

Employ the code generation features of Azure OpenAI Service - This feature is focused on generating code from natural language prompts.

Leverage the image creation functions of Azure OpenAI Service - Image creation functions like DALL·E are used for generating images from text descriptions

Use Azure's QnA Maker to fetch answers from a knowledge base - While QnA Maker is useful for creating a FAQ-style system based on a knowledge base, it does not offer the same level of dynamic, human-like conversation generation as the text generation models in Azure OpenAI Service.

Accountability - This is the correct answer. Accountability ensures that AI systems are developed with mechanisms for oversight, and that organizations take responsibility for the outcomes produced by these systems. It involves ensuring that AI systems are used ethically and that their creators or operators are held responsible for their impact.

Transparency focuses on making AI systems understandable and providing visibility into how they work, but it doesn't directly address the mechanisms for oversight and responsibility for outcomes.

Inclusiveness is about ensuring that AI systems are designed to be fair and accessible, considering the diverse needs of users, but it is not specifically about oversight or accountability for outcomes.

Reliability and Safety focus on ensuring that AI systems perform as expected and do not cause harm, but accountability is the key consideration for ensuring oversight and responsibility for the system's outcomes.

They create new data based on learned patterns from training data - This is the correct answer. AI models that generate new content, such as text, images, or music, are typically generative models. They learn patterns from training data and use this knowledge to produce original content that resembles the data they were trained on.

They classify input data into predefined categories - This describes classification models, which focus on categorizing data into specific labels or categories, rather than generating new content.

They predict numerical values from historical data - This is characteristic of predictive models, which are used for forecasting or regression tasks, rather than generating new content.

They retrieve exact copies of existing content - This describes retrieval-based models, which focus on finding and returning pre-existing content from a database or dataset, rather than generating new, original content.

Including fail-safe mechanisms in the system design helps prevent the application from causing harm if it encounters unexpected situations. Minimizing transparency, reducing data diversity, and ignoring edge cases can compromise safety and lead to undesirable outcomes.

The Azure AI Speech service provides text-to-speech capabilities, allowing developers to convert written text into spoken words with natural-sounding voices. This service supports multiple languages and customizable voice options, making it suitable for applications like audiobooks, accessibility tools, and interactive voice responses.

The Azure AI Language service focuses on natural language processing tasks such as sentiment analysis and key phrase extraction but does not offer text-to-speech functionality. Azure AI Vision service is designed for image and video analysis. Azure Cognitive Search provides indexing and querying over content but does not handle text-to-speech conversion.

Key Phrase Extraction is the appropriate Azure AI capability for this scenario. It analyzes text to identify the most relevant phrases that represent the main topics or concepts within a document. These key phrases can be used as metadata tags to improve searchability and organization.

Sentiment Analysis determines the emotional tone of the text, which is not useful for identifying key concepts.

Entity Recognition identifies and categorizes specific entities like names of people, organizations, or locations but may not capture all significant terms.

Speech Recognition converts spoken language into text and is not applicable to text documents.

They should use Computer Vision. This workload enables machines to interpret and understand visual information from images or videos. In this scenario, the retail company aims to analyze images from cameras to identify and categorize products, which is a typical application of Computer Vision.

Natural Language Processing deals with understanding and processing human language, which is not applicable here.

Knowledge Mining involves extracting information from large volumes of text but doesn't directly address image analysis.

Generative AI focuses on creating new content, such as text or images, and is not suited for product identification tasks.

Removing or masking personally identifiable information (PII) before feature engineering-through data anonymization or strong redaction-ensures that customer identities cannot be recovered, so the training data is no longer classified as personal data under GDPR. Encryption alone protects data at rest and in transit but exposes PII once the data are decrypted inside the training pipeline. Reversible hashing (pseudonymization) still permits re-identification if the lookup table is compromised. Aggregating the data into population-level statistics removes all personalization signals, making it unsuitable for a recommender system.

Evaluate and adjust the training data to remove discriminatory patterns - This is the correct answer. Ensuring that the training data is free from biases and discriminatory patterns is essential for responsible AI. This helps ensure that the job descriptions generated by the model do not inadvertently favor one group over another, promoting fairness and equity.

Increase the model's vocabulary to include industry-specific terms - While increasing vocabulary can improve the model's relevance to specific industries, it does not directly address equity concerns in the generated content.

Optimize the model's performance to generate descriptions faster - Performance optimization for speed may improve efficiency but does not directly relate to ensuring equitable treatment of candidates in the generated job descriptions.

Reduce the computational resources required for deployment - Reducing computational resources can be important for cost or environmental reasons but does not directly address fairness or equity in the AI-generated job descriptions.

DALL·E is an image generation model available in Azure OpenAI Service that creates images from text descriptions, allowing users to generate visual content based on textual inputs.

GPT-3 and ChatGPT are language models designed for generating and understanding human-like text, not images.

Codex is tailored for code generation and completion tasks and does not generate images.

Implementing algorithms that promote fairness in the model's decision-making process helps to mitigate discriminatory patterns and ensure equitable outcomes across different demographic groups. Merely increasing data from the affected group or removing demographic features might not fully resolve the underlying biases. Switching to a different algorithm without addressing fairness considerations may not eliminate the disparity.

Regression analysis is used when the goal is to predict a continuous numerical value based on input variables. Forecasting the future price of a stock is a typical regression problem because it involves predicting a numeric value.

Classifying emails as spam or not spam is a classification task because it involves assigning discrete categories to data points.

Grouping customers based on purchasing behavior is a clustering task, as it involves identifying natural groupings within the data without predefined labels.

Detecting fraudulent transactions in financial data is also a classification problem, aiming to categorize transactions as fraudulent or legitimate.

Computer Vision's text reading capability - This is the correct answer. Computer Vision's text reading capability (including OCR features) in Azure AI services is the most suitable solution for extracting both printed and handwritten text from scanned documents and images. It can process unstructured data by detecting and reading text in various image formats, making it ideal for the organization's needs.

Form Recognizer to analyze and extract structured data - Form Recognizer is useful for extracting structured data (e.g., from forms or invoices), but it is not primarily designed to handle the extraction of general unstructured text from images, especially when dealing with a mix of printed and handwritten text.

Text Analytics to analyze and interpret text sentiment - Text Analytics is primarily used for analyzing and interpreting the sentiment, entities, or key phrases in text data, but it does not extract text from images or scanned documents.

Face API to detect and analyze faces in images - The Face API is focused on detecting and analyzing human faces in images, not for extracting text from documents or images. This is unrelated to the task of text extraction.

Estimating emotional states of people in images - This is the correct answer. Solutions that analyze human faces in images often include the ability to estimate emotional states, such as detecting whether a person is happy, sad, angry, or surprised based on their facial expressions.

Converting handwritten text into digital format - This feature is associated with Optical Character Recognition (OCR), which focuses on recognizing and converting text from images, not facial analysis.

Detecting objects like vehicles and furniture - This is the functionality of object detection solutions, which focus on identifying and locating specific objects in an image, not analyzing human faces.

Categorizing images into predefined classes - This refers to image classification, where the goal is to classify an image into one of several predefined categories, but it is not specifically focused on analyzing human faces.

Computer Vision - This is the correct answer. An AI system that processes images to recognize and categorize objects is an example of computer vision. This type of workload focuses on enabling machines to interpret and understand visual information, such as identifying objects within images.

Natural Language Processing (NLP) is focused on understanding and generating human language, such as text or speech. It is not used for processing images.

Knowledge Mining involves extracting insights and patterns from unstructured data, such as documents and media, but it is not specifically focused on recognizing or categorizing objects within images. It may include visual data analysis but is broader in scope.

Object Detection is the appropriate capability because it can identify and locate multiple instances of objects within an image, providing both the class of each object and its position through bounding boxes. This enables the company to count and track products in images effectively.

Image Classification assigns a single label to an entire image without providing location data.

Optical Character Recognition (OCR) extracts text from images but doesn't detect objects.

Facial Recognition is specialized for detecting and identifying human faces, not general products.

The correct answer is DALL-E. DALL-E models are specifically designed to generate original images from textual prompts within the Azure OpenAI Service. GPT-4 is a large language model focused on understanding and generating text. Azure Computer Vision is an Azure AI service used for analyzing existing images, such as recognizing objects and extracting text, not for generating new ones. Codex is a model family that is optimized for generating code.

They can generate new data instances similar to the training data - This is the correct answer. Generative AI models are designed to create new data instances that resemble the training data. This makes them ideal for generating synthetic data to augment testing datasets, ensuring variety and realism in the data used for testing.

They can classify data into specific categories with high precision - This feature is characteristic of classification models, which categorize data into predefined labels but do not generate new data instances.

They can reduce data dimensionality while retaining key features - This describes dimensionality reduction techniques (such as PCA), which focus on simplifying the data without losing important features. It is not related to generating new data instances.

They can identify anomalies by learning normal data patterns - This is a feature of anomaly detection models, which are used to identify outliers or unusual patterns in data, not for generating synthetic data.

Transparency involves making AI systems understandable and explainable to users and stakeholders. By enhancing transparency, the company allows users to comprehend how data is processed and how decisions are made, building trust in the system.

Privacy focuses on protecting personal and sensitive data, which is important but does not specifically address understanding AI decision-making processes.

Fairness aims to prevent biases and ensure equitable outcomes.

Inclusiveness ensures AI systems are accessible and beneficial to diverse users.

Azure Speech service is designed for converting spoken language into text (speech-to-text) and vice versa (text-to-speech). It enables applications to transcribe and understand spoken words, which is essential for a virtual assistant that processes voice commands.

Azure Bot Service helps in building conversational agents but does not handle speech recognition.

Azure Text Analytics processes textual data for tasks like sentiment analysis and key phrase extraction but does not convert speech to text.

Azure Cognitive Services Translation translates text from one language to another but does not perform speech recognition.