Identify AI Concepts and Responsibilities

Domain 01 covers three skill areas: (1) Describe principles of responsible AI — all six Microsoft Responsible AI principles and their considerations; (2) Identify AI model components and configurations — how generative AI models work, how to select the right model by capability, and deployment options; and (3) Identify AI workloads — recognising the right workload type for a scenario across generative AI, agentic AI, text analysis, speech, computer vision, and information extraction.

Responsible AI questions appear in two main formats. The first presents a real-world practice or task and asks which of the six principles it exemplifies. The second format is True/False, where a statement describes a specific behaviour or practice and candidates confirm whether it correctly represents a principle. The same six principles appear repeatedly across both formats, so knowing their precise definitions and being able to distinguish between similar ones, particularly fairness, transparency, and accountability, is critical.

Transparency is about explainability. AI systems should be understandable, and users should be informed when they are interacting with AI and how decisions are made. Accountability is about human responsibility. People and organisations must remain responsible for AI-generated decisions through practices like human-in-the-loop review, ongoing monitoring after deployment, and disclosure of the teams responsible for the system. Both are tested in scenario and True/False questions, and they are the most commonly confused principle pair on the exam.

Through True/False questions that test precise conceptual boundaries. Key facts to know: generative AI models are not retrained before each user request, as inference uses the already-trained model; responses are generated by predicting the next token based on patterns from training data, not by copying stored documents; and the inference stage is when the model generates a response to a prompt. Multiple questions test the same underlying concepts from slightly different angles.

Workload identification questions present a scenario and ask candidates to name the correct AI workload type. The key skill is matching the type of output required to the correct category: tasks producing new text content map to generative AI; tasks producing new images from text descriptions map to image generation; tasks identifying named entities in text map to text analysis (NER); and tasks converting spoken audio to text map to speech recognition. Candidates must focus on what the solution produces, not just what its input is.

Max Completion Tokens is the most directly tested parameter. Questions present a requirement to control response length and minimise cost, and candidates must identify it as the correct setting rather than Temperature or Top P. The exam also tests True/False statements about when parameters can be configured and how deployment names are used to route inference requests. Foundry portal behaviour around model deployment and testing, including what is not required before deploying a model, is tested through True/False questions.