Read with the model in view
Use the visual as a map for the concepts, evidence, and decisions described in this note.
Artificial Intelligence projects in class often move quickly from a promising demo to a confident presentation. A risk register slows the process just enough for students to name what could go wrong, who will watch for it, and what evidence should trigger a review.
The goal is not to make student projects feel bureaucratic. The goal is to teach learners that responsible AI work includes visible decisions about uncertainty.
What to capture
A lightweight AI risk register can use five columns:
- Risk statement: What harmful, misleading, unfair, insecure, or unreliable outcome could occur?
- Evidence to monitor: Which test result, evaluation log, user comment, or operational signal would reveal the risk?
- Owner: Who is responsible for reviewing the signal and deciding whether action is needed?
- Mitigation: What guardrail, fallback, human review step, or communication plan reduces the risk?
- Review trigger: What condition requires the team to pause, revise, or escalate the project?
Classroom prompt
Ask each team to identify one technical risk, one user-experience risk, and one governance risk. For example, an AI advising assistant might produce inaccurate recommendations, confuse students about official policy, or lack an accountable review pathway when advice is challenged.
Teaching note
Encourage students to write risks as testable statements instead of vague concerns. “The model may be biased” is less useful than “The model may recommend different support resources for similar student cases without a documented reason.”
Learning outcome
Students learn to connect AI evaluation with project governance. They practise making uncertainty reviewable through evidence, ownership, and practical decision triggers.
Turn the note into a classroom decision
- Choose one concept from the note.
- Name an artifact that would make it observable.
- Write the next action and how you would verify it.