Identify AI concepts and responsibilities
Free Microsoft Certified: Azure AI Fundamentals practice — 6 questions on Identify AI concepts and responsibilities, with explanations. No sign-up. Full 12-question mixed test →
A healthcare provider is deploying an AI system to analyze patient X-ray images for potential abnormalities. The system occasionally flags healthy tissue as abnormal, requiring radiologists to review false positives. Which responsible AI principle is MOST directly challenged when the system disproportionately generates false positives for patients with darker skin tones due to training data imbalance?
Fairness directly addresses the requirement that AI systems produce equitable outcomes across different groups. When a model shows systematically different error rates for specific demographics due to biased training data, this is a core fairness violation requiring dataset rebalancing and validation across subgroups.
A financial services company needs to build an AI solution that analyzes quarterly earnings call transcripts to extract key financial metrics, sentiment trends, and regulatory compliance risks mentioned by executives. The solution must process both the spoken audio and presentation slides simultaneously. Which combination of AI workloads BEST addresses this scenario?
This scenario requires three distinct specialized workloads: speech (converting audio to text), document intelligence (structured extraction from presentation slides including tables and forms), and NLP (sentiment analysis, entity extraction, and understanding regulatory language). Each workload is purpose-built for its task, and combining them provides the most accurate and reliable solution.
An AI development team is evaluating whether to use a large language model (LLM) or a small language model (SLM) for a customer service chatbot that will run on mobile devices with limited connectivity. The chatbot needs to handle common product questions with 80% accuracy but must respond within 2 seconds and work offline. Which factor MOST strongly favors selecting an SLM over an LLM for this scenario?
The critical requirement is offline functionality with limited connectivity. SLMs typically have millions to low billions of parameters (vs. hundreds of billions for LLMs), making them small enough to deploy directly on mobile devices. This enables local inference without cloud API calls, meeting the offline and 2-second latency requirements. Edge deployment is the key architectural advantage of SLMs for this scenario.
A legal firm is implementing a generative AI system to draft contract summaries from 50-page lease agreements. During testing, the system occasionally includes clauses that don't exist in the source document. The firm implements a solution where the AI must cite specific page numbers for every statement in the summary and retrieves relevant contract sections before generating each summary paragraph. Which term BEST describes this mitigation approach?
Grounding with RAG (Retrieval-Augmented Generation) specifically addresses hallucination by retrieving relevant source content before generation and constraining the model to reference only retrieved information. Requiring page citations enforces grounding by making the model base responses on actual document content rather than generating plausible-sounding but fabricated clauses. This is the standard approach for reducing hallucination in document-based generative AI applications.
A manufacturing company wants to deploy an AI agent that monitors production line cameras, detects equipment anomalies, retrieves maintenance manuals from a database, and automatically generates work orders with repair instructions. The solution must coordinate multiple steps: analyzing visual input, making decisions based on historical data, and producing structured output. Which statement BEST describes why an AI agent architecture is appropriate for this scenario compared to a single-purpose AI model?
AI agents are characterized by their ability to autonomously plan and execute multi-step workflows, use tools (like database queries and work order systems), and make decisions about next actions based on intermediate results. This scenario requires orchestrating computer vision (anomaly detection), information retrieval (manual lookup), decision-making (severity assessment), and structured generation (work orders)—exactly what agent architectures are designed to handle through tool use and sequential reasoning.
A university is developing an AI-powered admissions assistant that answers prospective student questions about programs, requirements, and campus life. During compliance review, legal counsel requires that every AI response include a mechanism for students to understand what information influenced the answer and a way to challenge potentially incorrect information. Which responsible AI principle is the legal requirement MOST directly enforcing?
Transparency specifically requires that AI systems provide understandable information about how they operate and how decisions are made. Requiring users to understand what information influenced responses and providing challenge mechanisms directly implements transparency—it makes the system's reasoning visible and verifiable. This allows users to assess trustworthiness and detect potential errors.
Ready for the real thing?
The full course has two full-length practice tests, video lessons for every exam domain, hands-on labs and detailed answer explanations.
Start my full course on Udemy →