Font size
WorksheetsGenerative AI/Fairness and Ethics
Total questions: 64
Worksheet time: 32mins
Some mitigation techniques you can use at the metaprompt and grounding layer include:
Clarifying user intent and providing context
Ignoring user input
Generating random responses
Avoiding any user feedback
A Large Language Model (LLM) is:
a type of artificial intelligence that processes and generates human-like text
a device used for storing large amounts of data
a programming language for building web applications
a tool for creating digital images
The first step in training a transformer model is:
Tokenizing the input data
Evaluating the model
Deploying the model
Fine-tuning the model
LLMs are based on which type of architecture?
Convolutional Neural Network
Recurrent Neural Network
Transformer
Decision Tree
An encoder block is:
a component that processes input data and extracts features in a neural network
a device that converts analog signals to digital signals
a hardware component used for data storage
a type of memory used in computers
A decoder block is:
a component that converts encoded data back to its original form
a device that encodes information for transmission
a block that stores data temporarily
a circuit that amplifies signals
What best describes Generative AI?
AI that analyzes and summarizes existing data
AI that creates new content such as text, images, or code
AI that only classifies data into categories
AI that detects anomalies in data
Which Azure service provides access to pre-trained generative language and image models?
Azure Machine Learning
Azure AI Studio
Azure Cognitive Search
Azure Data Factory
What is the primary advantage of using foundation models?
They require manual training from scratch for each task
They can be fine-tuned for many tasks with less data
They only work for image generation
They eliminate all ethical risks
What is a prompt in generative AI?
A dataset used to train a model
A command that tells a model what to create or output
A data visualization tool
A type of supervised learning
Which of the following is NOT a common generative AI use case?
Text summarization
Image creation
Predicting customer churn
Code completion
What Azure service lets you customize generative AI models safely without writing code?
Azure AI Content Safety
Azure OpenAI Service in Azure AI Studio
Azure Synapse Analytics
Power BI
What is prompt engineering?
Building neural networks manually
Improving a model’s accuracy by tuning weights
Crafting and refining instructions to guide generative model output
Writing Python scripts to deploy models
Large Language Models (LLMs) are primarily trained to:
Identify objects in images
Predict the next word in a sequence
Detect network intrusions
Cluster unlabeled data
What is an example of a multimodal model?
A model that only processes tabular data
A model that handles both text and images as inputs
A model that runs on multiple GPUs
A model with multiple hidden layers
Which Microsoft tool provides guided access to generative AI capabilities for low-code users?
Azure Logic Apps
Power Virtual Agents with Copilot
Microsoft Defender
SQL Server
A marketing team wants to automatically generate product descriptions and social-media posts from a spreadsheet of items. Which Azure service should they use?
Azure Cognitive Search
Azure OpenAI Service
Azure Machine Learning Designer
Azure DevOps
A teacher wants a chatbot that answers students’ questions using only her school’s website content. Which Azure component ensures the model stays within that data?
Azure Cognitive Search indexing + grounding
Azure Content Safety
Azure Cosmos DB
Azure Container Registry
A developer wants to moderate user-generated content for harmful or unsafe language before displaying it. Which service should they add?
Azure AI Content Safety
Azure Monitor
Azure Sentinel
Azure Speech Services
When using Azure OpenAI, which responsibility belongs to the user (not Microsoft)?
Building data centers
Prompt design and responsible deployment
Model pre-training
GPU optimization
Which feature helps prevent data leakage when using enterprise data with generative AI?
Role-based access control and data grounding
Public API endpoints
Anonymous authentication
Hard-coding credentials
An HR department uses generative AI to summarize resumes. Which risk must they consider most?
Overfitting
Data bias and fairness
Under-sampling
Low latency
An e-commerce site uses Azure AI Studio to create product-image variations. Which type of model are they using?
Large Language Model (LLM)
Diffusion or image-generation model
Clustering model
Regression model
Which Azure AI feature supports monitoring for toxicity and bias in generated outputs?
Azure Advisor
Responsible AI dashboard
Azure Resource Monitor
Azure Databricks
Your company wants to integrate generative AI features into a Power App. Which connector or capability allows that?
Power Automate flows
AI Builder with Azure OpenAI integration
Microsoft Fabric
Azure Kubernetes Service
Which principle of Responsible AI relates to giving users clear explanations of how AI decisions are made?
Reliability and safety
Inclusiveness
Transparency
Accountability
Which of the following are examples of generative AI models?
GPT-4
DALL-E 3
ResNet-50
K-Means
Which Azure services can be used to build or customize generative AI solutions?
Azure OpenAI Service
Azure AI Studio
Azure Key Vault
Azure Monitor
Which practices support responsible use of generative AI?
Providing user disclaimers about AI-generated content
Training models only on private data
Evaluating outputs for bias and harm
Allowing unrestricted public prompts
What are common input types for generative AI systems?
Text prompts
Numerical regressions
Images
Hardware addresses
Which statements about Azure OpenAI Service are true?
It provides access to models like GPT-4 and DALL-E through an API
It requires users to build models from scratch
It integrates with Azure AI Content Safety
It can only run on-premises hardware
What is the main goal of responsible AI?
To make AI models run faster
To ensure AI systems are built and used in ways that are ethical, fair, and safe
To increase company profits through automation
To replace human decision-making entirely
Which of the following best describes fairness in AI systems?
AI gives everyone identical outcomes
AI provides equal and appropriate treatment across groups without bias
AI models perform faster for all users
AI avoids human involvement
What is an example of bias in an AI system?
A model predicting the same result every time
A model performing worse for certain demographic groups
A model improving with more data
A model learning to predict new patterns
Which of the following is an ethical risk of AI?
Overfitting a model
Collecting personal data without consent
Optimizing hyperparameters
Using cloud compute resources
Which Microsoft Responsible AI principle focuses on giving users understandable explanations of how AI works?
Reliability and Safety
Transparency
Privacy and Security
Fairness
What does accountability mean in responsible AI?
Only the AI is responsible for its own outcomes
People and organizations remain responsible for AI decisions and impacts
AI systems self-report errors
Accountability is optional in AI
When training a model, which action helps reduce bias?
Collecting diverse and representative datasets
Using fewer training samples
Ignoring outliers
Using a more complex algorithm
What is inclusive design in AI?
Designing AI that only works for specific users
Designing AI systems that consider a variety of abilities, languages, and backgrounds
Removing accessibility features for simplicity
Focusing only on technical accuracy
Which Microsoft service helps identify and mitigate unfairness in models?
Azure AI Content Safety
Responsible AI Dashboard in Azure Machine Learning
Azure Cognitive Search
Azure Synapse Analytics
Why is reliability and safety important in AI?
It reduces server costs
It ensures models behave as intended and handle unexpected inputs safely
It improves the visual design of an app
It helps AI generate more creative outputs
A bank uses an AI model for loan approval. The model denies loans to women more often than men, even with the same income and credit score. Which principle is being violated?
Fairness
Transparency
Inclusiveness
Reliability
A medical AI tool misclassifies darker-skinned patients due to underrepresentation in the dataset. What is the main cause?
Biased training data
Too much model regularization
Model drift
Hyperparameter tuning
A school uses an AI system to predict which students might need academic support. What step promotes ethical use of this AI?
Making predictions public
Explaining the AI’s reasoning to parents and teachers
Using AI without human review
Collecting sensitive data without consent
A company deploys a chatbot that generates offensive responses to users. Which Azure service can help prevent this issue?
Azure AI Content Safety
Azure Kubernetes Service
Azure Monitor
Power Automate
An organization wants to ensure its AI products follow Microsoft’s Responsible AI standards. Which tool provides guided assessments and best practices?
Responsible AI Standard
Azure Key Vault
AI Builder
Azure Sentinel
What is a best practice for ensuring transparency in an AI project?
Documenting data sources and model design choices
Keeping datasets secret
An AI team discovers bias in model predictions. What should they do first?
Delete all data
Identify and analyze the source of the bias in data or labeling
Reduce the model’s accuracy threshold
Ignore it if accuracy is high
When might privacy and security concerns arise in AI systems?
When sensitive personal data is used or shared without proper protection
When using open-source libraries
When models are optimized
When using synthetic data
A company uses an AI hiring tool. To maintain accountability, what should they do?
Allow the AI to make final hiring decisions
Keep humans in the loop to review and approve decisions
Hide AI use from applicants
Automate all HR processes
Which of these is a key feature of Microsoft’s Responsible AI approach?
Encouraging experimentation without limits
Publishing tools and frameworks like the Responsible AI dashboard
Allowing models to evolve independently
Avoiding human oversight
Which are Microsoft’s six principles of Responsible AI?
Fairness
Reliability and Safety
Scalability
Accountability
Efficiency
Which actions improve fairness in an AI solution?
Auditing models for demographic performance differences
Ignoring sensitive attributes
Using diverse data samples
Reducing training data to simplify the model
Which practices promote transparency?
Providing clear explanations of model predictions
Sharing data sources and methods with stakeholders
Encrypting all outputs
Hiding how decisions are made
Which Azure features help support responsible AI?
Responsible AI Dashboard in Azure ML
Content filtering in Azure AI Content Safety
Azure DevOps Pipelines
Azure Monitor
Which actions demonstrate accountability in AI?
A. Assigning clear human ownership for AI decisions
B. Testing and auditing models regularly
C. Removing all human review
D. Automating every decision
What is a potential consequence of using non-representative data in AI training?
Reduced need for model evaluation
Faster model deployment
Improved model accuracy for all users
Biased predictions against certain groups
Which Microsoft Azure service can help detect and mitigate harmful outputs from generative AI models?
Azure Virtual Machines
Azure Logic Apps
Azure AI Content Safety
Azure Blob Storage
What does LLM stand for in the context of artificial intelligence?
Large Language Model
Linguistic Learning Machine
Longitudinal Learning Model
Logical Language Matrix
What does 'GPT' stand for in the context of AI?
Generative Pre-trained Transformer
Generative Predictive Text
General Processing Technology
Graphical Processing Tool
What is the first step in training a transformer model?
Attention
Tokenization
Decoding
Embeddings
How does generative AI create new content, such as text or images?
By using random combinations of elements from its training data.
By applying fixed rules programmed by developers.
By learning patterns in the training data and generating outputs based on these patterns.
By mimicking exact examples from its training data without any variation.
What is the potential impact of foundation models in various industries?
Cause confusion and misinformation in industries
Only be used for academic research
Have no impact on any industry
Revolutionize industries such as healthcare, finance, and customer service
What is the potential problem of hallucinations in generative AI?
They improve the model's accuracy
They have no impact on the model's output
They are necessary for the model to function properly
They can make the output text difficult to understand and generate incorrect information
