M11 — Use AI without surrendering judgement
Use AI without surrendering judgement
Module outcome: inspect, correct or reject generated teaching resources. Model limits understood, hints requested instead of answers, generated material audited, and model choices made from recorded evidence rather than enthusiasm.
Prerequisite: any taught unit from this course, so there is material to audit. Planned module budget: about 16 of the 200 activity hours. Built today: nothing complete. All four units are specifications.
M11-U01 · Understand model limits
Specification. The unit page does not exist yet.
Planned content: distinguish model, prompt, skill, tool and workflow; generated language is not a source citation. Evidence to keep: a tool-choice explanation and a five-term example card.
M11-U02 · Ask for hints
Specification. The unit page does not exist yet.
Planned content: provide the source and the attempt; request a hint without a completed answer. Evidence to keep: a hint-and-retry record built on a source-bound tutoring prompt.
M11-U03 · Audit generated material
Specification. The unit page does not exist yet.
Planned content: verify meaning, levels, answers, facts and rights; reject convincing but unsupported content. Evidence to keep: an annotated AI output and an adversarial review case.
M11-U04 · Choose models empirically
Specification. The unit page does not exist yet.
Planned content: run the same cases under a cost ceiling; record correctness, retries, latency and unsupported claims. Evidence to keep: a comparison record without invented scores.
Game · Human or machine?
Any group, two worksheet versions, 15 minutes. The trainee prepares two mini-explanations of the same language point for their own learners: one written by themselves, one generated by an AI tool, then lightly tidied. The audience gets both unlabelled and votes on each question in turn: which explanation would you trust, which example would you use, which answer key would you check before class. Every vote needs a reason, and the trainee records the reasons without defending either version.
Trainee note: the record trains the audit skill directly. Vague reasons, “it sounds strange”, get refined into checkable ones, “the example uses a word the reading never taught”. When the room confidently picks the machine version and is wrong about an error you planted, that moment does more for judgement than any warning lecture.
Original design for this course. No book in the standard resource shelf covers AI audit; this game was written for this module.
Play it now: explanation court
Two rounds. Each round puts one language point on trial with two explanations, A and B. One is human, one is machine — the labels are hidden. You have three votes per round, and each vote needs a reason. A reason like it sounds clear earns nothing; a reason like it shows a sentence I could copy earns a point. The verdict matters less than the quality of your reasons. No timer.
The spoken debate version is above.
Interactive version built for this course; original design, as stated on this page.
How to work through this module
No built units yet. Use the specifications as your AI checklist for every tool you let near your teaching: know what layer you are using, ask for hints against a source, audit every generated page before a learner sees it, and record comparisons instead of adopting tools on impression.
Sources and review status
Source anchors C3, R6, R7, R8 and T3 in the evidence register. No source certifies this module. Educator review pending. Corrections go to the contact page.
Previous: M10 Teach online and with low bandwidth · Next: M12 Teach, reflect and demonstrate · Course modules