A lecturer reads 60 end-of-term essays and notices that more than half share the same five-paragraph shape, the same transitions (“Furthermore, it is important to note that…”) and the same balanced-to-the-point-of-bland conclusion. She cannot prove any of them used AI, and she does not need to. The real problem lies elsewhere: the assignment no longer measures what she wants to measure.
For educators, the useful question is not “how do I catch students using AI?” but “how should my assessment change, and how can AI help me teach better?”
Which assessments AI can easily do for students
| Format | Vulnerability | Redesign direction |
|---|---|---|
| Generic take-home essay | Very high | Tie it to local data, specific cases or in-class experiences |
| Unproctored online quiz | High | Use for self-practice with low weighting |
| Group project report | Moderate | Assess the process: logs, drafts, a presentation |
| Timed in-class writing | Low | Keep for foundational skills |
| Oral examination, viva | Low | Add a five-minute conversation to take-home work |
| Critiquing AI output | Low | Ask students to evaluate and improve an AI-generated answer |
Five redesigns that work
- Assess process, not just product. Ask for an outline, a commented draft and a short reflection on what changed between versions.
- Use data or context that is yours. Analyse data the class collected, respond to a specific seminar discussion, study a particular local organisation.
- Add a spoken element. Five minutes of questions about a submission reveals understanding more reliably than any detection tool.
- Make AI the object of analysis. Give students an AI-generated answer and ask them to find errors, check sources and identify shallow reasoning.
- Break tasks down and do some in class. Skills that must be practised unaided, such as building an argument paragraph, happen during contact time.
Writing a course AI policy
A clear policy means honest students do not have to guess. A good one covers:
- Permitted use per assignment: none; specific tasks (brainstorming, grammar checks); open use with disclosure.
- The reason: “Assignment 1 develops your own reasoning, so AI may not be used to generate content.” Students respect rules that come with reasons.
- How to disclose: a short template naming the tool and what it was used for.
- Consequences for breaches, linked to institutional integrity rules.
Talk the policy through in the first session rather than just posting it online. Many breaches stem from misunderstanding rather than intent.
Using AI to prepare teaching
- Practice questions: ask AI for 20 questions at different cognitive levels for a chapter, then select and edit. Check every answer; AI often produces wrong keys or distractors that are also correct.
- Examples and scenarios: examples closer to students' lives, and several variants for different groups.
- Draft rubrics: AI proposes an initial framework, which you align with the course learning outcomes.
- Multiple explanations of a difficult concept.
Whatever you use in class is still your responsibility; a question with a wrong answer key costs more trust than the time it saved.
AI in feedback and grading: lines to hold
Using AI to draft feedback or grade is tempting with hundreds of scripts. The risks are real:
- Student data: submissions are students' personal data and intellectual work. Do not upload them to external tools unless your institution has an approved service and students have been informed.
- Inconsistency: the same script graded twice may receive different results.
- Bias: AI may favour certain styles, disadvantaging students writing in a second language or arguing in unconventional but valid ways.
- Accountability: students are entitled to know who assessed their work and why it received its mark.
A safer pattern: you grade and jot brief notes, and AI helps turn those notes into clear, structured, constructive feedback. The grade is always a human decision. If AI plays any part, tell students.
Teaching AI literacy
Students will work in AI-rich environments. One session on checking AI outputs, spotting fabricated references and disclosing properly is worth more in the long run than a blanket ban. Model it live: put a disciplinary question to an AI assistant and work through what is right, wrong and missing with the class.
A plan for next term
- Review each assessment against the table above and identify which need redesigning.
- Write an AI policy for each assignment, with reasons and a disclosure template.
- Add at least one process or oral element to the most heavily weighted task.
- Try drafting a practice question set with AI and check every answer.
- Ask your institution whether an approved AI tool exists for handling student work.
Something to do this week: take the most heavily weighted assessment in your course, put the brief to an AI assistant and read the response as if a student had submitted it. If it would earn a decent mark, you know where your redesign should start.
Câu hỏi thường gặp
Should teachers ban students from using AI?
It depends on each task's purpose. Restricting AI for foundational skills is reasonable; elsewhere, permitting it with disclosure and teaching responsible use often works better than a blanket ban.
How can I make assignments harder for AI to complete?
Assess the process, use class-specific data or context, add a short oral component, and ask students to analyse and critique AI-generated output.
Can I use AI to grade student work?
Only to help phrase feedback, using an institutionally approved tool with students informed. Grading decisions should stay with the teacher, since AI can be inconsistent and biased.
Can AI write good multiple-choice questions?
It can produce useful drafts. Check every answer key and distractor, because AI often generates wrong keys or more than one correct option.
What should a course AI policy include?
Permitted use for each assignment, the reasons for the rules, how students should disclose AI use, and consequences for breaches under institutional policy.
