Identify AI Concepts and Capabilities – AI-901 Study Guide
This domain accounts for 43% of the exam (Microsoft's published range is 40–45%) and is the "describe" half of AI-901: responsible AI principles, how generative AI models work and how to choose between them, and the shape of the common AI workloads — computer vision, natural language, speech, and information extraction. Nothing here is hands-on; that's the other domain, Implement AI Solutions by Using Microsoft Foundry (57%), which covers actually building with the Foundry portal and SDK. Here you only need to recognize and name capabilities from a scenario description.
Responsible AI: The Six Principles
Microsoft frames responsible AI around six principles. Each exam question describes a situation and asks which principle applies — so anchor each to a hook and a scenario:
| Principle | Hook | Canonical scenario |
|---|---|---|
| Fairness | Treat all people equitably; avoid bias | A hiring model favors one gender for identical qualifications |
| Reliability & safety | Operate consistently and safely; fail gracefully | An autonomous system tested across conditions, handling the unexpected |
| Privacy & security | Protect data from unauthorized access | A healthcare system safeguarding patient records |
| Inclusiveness | Empower everyone, all abilities | Designing for users who rely on assistive technology |
| Transparency | Understandable; disclose capabilities and limits | Publishing how the system works and where it's unreliable |
| Accountability | Humans stay responsible; governance and oversight | A board owning outcomes, able to review and override decisions |
The confusion pairs the exam exploits:
- Fairness vs. inclusiveness — fairness is about equitable outcomes (no biased results); inclusiveness is about accessibility (everyone can use it). "Biased against a group's outcomes" → fairness. "Usable by people with disabilities" → inclusiveness.
- Transparency vs. accountability — transparency is about understandability and disclosure; accountability is about human responsibility, governance, and the ability to intervene. "Explain how it decides / state its limits" → transparency. "Humans own the outcome and can override" → accountability.
- Reliability & safety vs. privacy & security — reliability & safety is about behaving correctly and safely; privacy & security is about protecting data. Don't let "security" pull you to the safety principle when the scenario is really about data protection.
Common AI Workloads
Given a scenario, name the workload:
| Workload | What it does | Scenario tell |
|---|---|---|
| Computer vision | Interpret images and video | "identify/locate objects in a photo", "read a sign" |
| Natural language processing (NLP) | Understand text and speech | "analyze reviews", "transcribe calls", "detect language" |
| Document / information extraction | Extract structured information from documents, images, audio and video | "pull fields from invoices/receipts/forms" |
| Generative and agentic AI | Create new original content, or act autonomously toward a goal | "write drafts", "generate images from a prompt", "an agent that completes a multi-step task" |
The discriminator is the input and the goal: images → computer vision; text/speech → NLP; forms/media → information extraction; creating or acting → generative/agentic AI.
Sign in free to read the full guide
5 more sections (about 1,498 words) continue below. The first domain's study guide is free for every AI-901 learner with a free account.