AI-900 Azure AI Fundamentals - Set 2 - Part 1

Correct Answer: Supervised classification

Privacy and security is the correct principle because healthcare applications handle personally identifiable information that must be protected from leaks, unauthorized disclosure, and improper exposure during model training.

Option a is incorrect because fairness focuses on eliminating bias and ensuring equal treatment across different groups, not specifically on protecting sensitive data from exposure or unauthorized access.

Option b is incorrect because transparency relates to making AI systems understandable and explainable to users, not primarily concerned with safeguarding personally identifiable information from security breaches.

Option d is incorrect because accountability focuses on establishing responsibility frameworks and ethical standards for AI development, rather than directly addressing data protection and privacy preservation measures.

Correct Answer: Edge computing deployment

Edge computing allows AI models to run locally on user devices, keeping personally identifiable information on the device itself, thereby avoiding vulnerabilities associated with transmitting sensitive data to external servers.

Option a is incorrect because centralized cloud processing requires data to be transmitted to and stored on external servers, which contradicts the requirement of keeping personal transaction information on local devices.

Option b is incorrect because federated learning is a training methodology where models learn from distributed data, but the question specifically asks about running inference on device without cloud communication.

Option d is incorrect because blockchain storage is primarily for creating immutable distributed ledgers, not specifically designed for running AI models locally on devices to protect personal information.

Correct Answer: Design for minority user needs

Microsoft's inclusiveness principle states that designing AI solutions for minority users including those with different physical abilities ensures the solution works effectively for the majority of users as well.

Option a is incorrect because focusing only on majority preferences can exclude users with disabilities or different needs, contradicting the principle that inclusive design for minorities benefits everyone.

Option c is incorrect because creating separate specialized versions goes against the inclusive design philosophy, which advocates for single solutions that accommodate diverse needs rather than segregated experiences.

Option d is incorrect because prioritizing cost-effectiveness over accessibility needs contradicts the inclusiveness principle, which emphasizes empowering everyone regardless of physical ability, gender, ethnicity, or other factors.

Correct Answer: Transparency through system explanations

Transparency requires AI systems to be understandable to end users, allowing them to comprehend why the system behaves as it does, which directly addresses the user confusion issue.

Option a is incorrect because while fairness is important for unbiased recommendations, the core issue described is user confusion about system behavior, which is specifically addressed by transparency principles.

Option c is incorrect because security and encryption protect data from unauthorized access but do not help users understand why specific content recommendations are being presented to them.

Option d is incorrect because accountability frameworks establish responsibility structures for AI development, but they don't directly resolve the immediate user need to understand system behavior and recommendation logic.

Correct Answer: Accountability through responsibility structures

Accountability principle ensures people are responsible for AI systems, requiring structures that consistently enact principles within government and organizational frameworks, directly addressing who is responsible for decisions and errors.

Option a is incorrect because privacy focuses on protecting personal data from leaks and unauthorized disclosure, not on establishing clear lines of responsibility for decision-making and error handling.

Option b is incorrect because inclusiveness focuses on ensuring AI empowers everyone regardless of physical ability or background, rather than establishing responsibility frameworks for system decisions and outcomes.

Option c is incorrect because transparency helps users understand system behavior through open frameworks, but accountability specifically addresses who is responsible and answerable when the system makes errors.

Correct Answer: Make clear what system can do

This guideline requires helping users understand what the AI system is capable of doing, which PowerPoint Designer demonstrates by displaying suggested topics that clarify its feature capabilities.

Option a is incorrect because making clear how well the system performs relates to communicating accuracy and limitations, not simply explaining what features and capabilities the system offers.

Option c is incorrect because supporting efficient dismissal focuses on making it easy to ignore or turn off AI features, not on initially communicating what capabilities exist.

Option d is incorrect because encouraging feedback relates to enabling users to indicate preferences during interactions, rather than upfront communication about what the system is designed to do.

Correct Answer: Uncertainty in predictions

Using tentative language like "we think" communicates uncertainty to users, helping them understand that AI recommendations are predictions that may not always be accurate or perfectly aligned.

Option a is incorrect because system capabilities refer to what features the AI can perform, while the tentative language specifically addresses the confidence level in predictions rather than capability boundaries.

Option c is incorrect because privacy protection involves safeguarding personal data, whereas the phrasing about thinking you'll like something relates to conveying prediction confidence, not data security.

Option d is incorrect because while personalization is part of recommendations, the specific use of uncertain language addresses prediction confidence rather than explaining how the system learns from behavior.

Correct Answer: Time services based on context

This guideline requires timing actions or interruptions based on the user's current task and environment, which Outlook demonstrates by considering location, destination, and traffic conditions.

Option a is incorrect because efficient invocation focuses on making it easy for users to manually request system services when needed, not on the system proactively timing notifications.

Option c is incorrect because matching social norms ensures the experience is delivered appropriately for social and cultural contexts, not specifically about timing actions based on environmental factors.

Option d is incorrect because showing contextually relevant information focuses on displaying appropriate data for the situation, while timing services emphasizes when to act or interrupt the user.

Correct Answer: Show contextually relevant information

This guideline focuses on displaying information that is pertinent to the user's current task and environment, which Walmart demonstrates by recommending accessories that complement the gaming console being viewed.

Option a is incorrect because timing services relates to when to interrupt or act based on user context, whereas this scenario is about what information to show during browsing.

Option b is incorrect because remembering recent interactions involves maintaining short-term memory for efficient references, not about showing complementary product suggestions during current browsing sessions.

Option d is incorrect because learning from behavior involves personalizing experiences over time through past actions, while this scenario demonstrates immediate contextual relevance based on current viewing activity.

Correct Answer: Match relevant social norms

This guideline ensures the AI experience is delivered in a way users expect given social and cultural contexts, which Editor demonstrates by presenting suggestions politely rather than dictatorially.

Option a is incorrect because mitigating bias focuses on ensuring language and behaviors don't reinforce unfair stereotypes, while the polite suggestion approach specifically addresses culturally appropriate interaction styles.

Option b is incorrect because making clear capabilities involves helping users understand what the system can do, not how it communicates suggestions in a socially appropriate manner.

Option d is incorrect because global controls allow users to customize overall system behavior and monitoring, whereas polite phrasing relates to culturally appropriate communication styles during interactions.

Correct Answer: Social bias mitigation

This guideline requires ensuring AI system language and behaviors don't reinforce undesirable unfair stereotypes and biases, which MyAnalytics addresses by using gender-neutral icons instead of gendered representations.

Option a is incorrect because transparency focuses on making AI decisions understandable and explainable to users, not on preventing reinforcement of gender stereotypes through visual design choices.

Option c is incorrect because efficient dismissal relates to making it easy to ignore or turn off AI features, not about preventing stereotypical representations in visual elements.

Option d is incorrect because contextual timing addresses when to act or interrupt based on user environment, whereas gender-neutral icons address preventing bias reinforcement in visual representations.

Correct Answer: Support efficient dismissal

This guideline specifically focuses on making it easy to dismiss or ignore undesired AI system services, which directly addresses the need to easily turn off automatic design suggestions.

Option a is incorrect because efficient invocation is about making it easy to request services when needed, not about turning off or ignoring unwanted automatic suggestions that interrupt workflow.

Option c is incorrect because scoping services when uncertain relates to disambiguation or graceful degradation when the system is unsure about user goals, not dismissing unwanted suggestions.

Option d is incorrect because making clear capabilities helps users understand what the system can do, whereas the issue is about easily disabling features users don't want.

Correct Answer: Support efficient dismissal options

This guideline emphasizes making it easy to dismiss or ignore undesired AI services, which Siri demonstrates by allowing simple voice commands to cancel actions without complex navigation.

Option a is incorrect because efficient correction focuses on making it easy to edit or fix AI system errors, whereas saying "never mind" dismisses the interaction entirely rather than correcting it.

Option c is incorrect because scoping services addresses disambiguation when the system is uncertain about goals, not providing an easy method to cancel ongoing interactions.

Option d is incorrect because timing services relates to when the system should act or interrupt based on context, not about providing quick dismissal methods for unwanted interactions.

Correct Answer: Support efficient correction

This guideline focuses on making it easy to edit, refine, or recover when the AI system is wrong, which the Alt Text feature demonstrates by providing accessible correction options.

Option a is incorrect because efficient dismissal is about ignoring or turning off features, whereas this scenario involves fixing incorrect AI-generated descriptions rather than dismissing the feature entirely.

Option c is incorrect because granular feedback enables users to indicate preferences during regular interactions, while the correction button specifically allows fixing errors rather than just providing preference signals.

Option d is incorrect because global controls allow customization of overall system behavior and monitoring, whereas the ribbon button provides targeted correction for specific AI-generated content instances.

Correct Answer: Scope services when in doubt

This guideline specifically requires engaging in disambiguation or gracefully degrading service when uncertain about user goals, which autocorrect demonstrates by presenting multiple options when unsure.

Option a is incorrect because making clear capabilities involves helping users understand what features exist, not how the system handles uncertainty by presenting multiple correction choices.

Option b is incorrect because efficient correction focuses on making it easy to fix errors after they occur, whereas disambiguation prevents errors by offering choices when uncertain.

Option d is incorrect because contextually relevant information relates to showing data pertinent to current tasks, not specifically about how the system handles uncertainty in user intent.

Correct Answer: Reasons for system behavior

This guideline requires enabling users to access explanations of why the AI system behaved as it did, which Office Online provides through descriptive text about recommendation logic.

Option a is incorrect because system capabilities relate to what features the AI can perform, whereas the descriptive text specifically explains why particular documents were recommended.

Option b is incorrect because uncertainty levels communicate prediction confidence, while the explanation text focuses on the reasoning behind which documents were selected for recommendation.

Option d is incorrect because user controls involve customization options for system behavior, whereas the descriptive text provides transparency about decision-making rationale rather than control mechanisms.

Correct Answer: Remember recent interactions

This guideline specifically requires maintaining short-term memory and allowing users to make efficient references to that memory, which Outlook demonstrates by surfacing recent files and contacts.

Option a is incorrect because learning from behavior involves personalizing experiences over time through past actions, whereas remembering recent interactions focuses on short-term memory for immediate efficiency.

Option c is incorrect because contextually relevant information relates to showing data pertinent to current tasks, while this specifically addresses maintaining accessible short-term memory of recent activities.

Option d is incorrect because timing services focuses on when to act or interrupt based on environment, not on maintaining and surfacing short-term memory of recent user interactions.

Correct Answer: Learn from user behavior

This guideline requires personalizing user experiences by learning from actions over time, which Amazon demonstrates by using purchase and browsing history to tailor future product recommendations.

Option a is incorrect because remembering recent interactions focuses on short-term memory for immediate efficiency, whereas Amazon's recommendations use long-term behavioral patterns to personalize experiences.

Option c is incorrect because contextually relevant information relates to showing data pertinent to current tasks, not specifically about adapting recommendations based on accumulated behavioral learning over time.

Option d is incorrect because updating cautiously focuses on limiting disruptive changes when modifying system behaviors, not on the core learning process that personalizes experiences from user actions.

Correct Answer: Update and adapt cautiously

This guideline requires limiting disruptive changes when updating or adapting system behaviors, which Designer demonstrates by integrating new capabilities into existing interfaces without disrupting established workflows.

Option a is incorrect because learning from behavior involves personalizing experiences based on actions over time, not about how new features are integrated without disrupting existing user experiences.

Option b is incorrect because making clear capabilities helps users understand what the system can do, whereas cautious updating addresses how changes are implemented without disruption.

Option d is incorrect because global controls allow users to customize system behavior broadly, while cautious adaptation focuses on implementing updates in non-disruptive ways to existing interfaces.

Correct Answer: Encourage granular feedback

This guideline specifically requires enabling users to provide feedback indicating their preferences during regular interactions with the AI system, which Excel demonstrates with the helpfulness question.

Option a is incorrect because efficient correction focuses on making it easy to fix errors when the system is wrong, whereas the helpfulness question gathers preference feedback rather than corrections.

Option c is incorrect because conveying consequences involves updating or communicating how user actions impact future system behaviors, not soliciting preference feedback on individual suggestions.

Option d is incorrect because learning from behavior involves personalizing experiences through observed actions over time, while granular feedback specifically solicits explicit preference signals during interactions.