AI, school and honesty
Rules about AI differ by school and by assignment. The safe approach is to know the rule, say when you used it, and use it to learn.
Words on this page (4)
- policy
- : A rule for choosing actions. In a language model, the model itself is the policy.
- tool
- : A function the model can ask your program to run, described by a name, description and argument schema.
- prompt
- : The input text you give a model, including instructions, examples and data.
- skill
- : A folder of instructions and scripts an agent loads only when a task needs it.
What it is
Some teachers allow AI for brainstorming and checking, some ban it for graded work, and some require you to show how you used it. Read your school's and ask when it is unclear. A simple test: after using it, could you explain every part of the work in your own words, without the chat open? If not, you have borrowed the result, not learned it.
Why detectors are not the answer
that claim to detect AI writing are unreliable. OpenAI launched its own text classifier in early 2023 and withdrew it a few months later because of its low accuracy. A 2023 Stanford study found that detectors wrongly flagged writing by non-native English speakers much more often. A flag is not proof, and false accusations can hurt real students. If you wrote it yourself, keep your notes and drafts, because they show your work.
A worked example
Two students write on the causes of the French Revolution. Student A pastes the and submits the reply. Student B writes an outline, asks the model "what is missing from my argument?", rewrites two paragraphs herself, checks the dates, and adds a note: "I used an AI assistant to critique my outline." Student B can answer any follow-up question. Student A cannot answer the first one.
Where the line sits
Think of three levels. Allowed help is like asking a classmate to explain a topic or check your spelling. Gray help is like having someone restructure your essay. Not allowed is submitting work you cannot explain as yours. Your teacher decides where each task falls, so the same use can be fine in one class and cheating in another.
A second example: a coding assignment
You must write a function that sorts a list. A student asks a model for the code, pastes it in and passes the tests. In the next lesson the class builds on that function and the student is lost. Another student writes her own attempt, gets stuck, and asks the model to explain the error message. She fixes it herself. Both finished the task. Only one can now write the next one. The test is the same as before: could you redo it alone?
Trade-offs
Using AI heavily saves time now and costs later. Avoiding it entirely keeps the skill but means you do not learn a tool that employers expect. A middle path works for most courses: use it for explanations, practice questions and feedback on your own draft, and do the graded thinking yourself.
What goes wrong
The common failure is a gap between your marks and your ability. You score well on homework and badly on a closed-book exam. That pattern, which teachers and students both notice, suggests the homework was done by the tool, not by you. A second failure is invented sources in an essay, which a teacher can check in a minute.
How to tell you are using it well
After a session, close the chat and write three sentences about what you learned. If you cannot, you borrowed rather than learned. Keep dated drafts and a short note on how you used the tool, so you can show your process if asked.
Common mistakes
Assuming that because no one can see it, it is allowed. Using AI to skip the exact skill the course is trying to build. Citing nothing: where the rules allow AI, say what you used it for.
A student's essay is flagged by a detector and she is called in. She opens her notes app and shows the outline, the three drafts and the sources she read, all with dates. The teacher reads the draft history and drops the matter. Keeping your own trail is the best defense, and a good learning habit anyway.
Check yourself
What is a good test of whether you used AI to learn or to skip learning?
Not quite. If you can explain and redo the work yourself, you learned it. If not, you only borrowed it.
Why is it risky to accuse someone based on an AI detector?
Not quite. Evidence shows high error rates, so a flag alone is not proof.
Your school has no clear rule for an assignment. What should you do?
Not quite. The person who set the assignment decides what is fair. Asking first protects you.
You got 0 of 3 right.
Go deeper
- Article
Liang et al. · The study showing that detectors misjudge non-native writing.
- Article
OpenAI · OpenAI's announcement, with its note that it later retired the tool for low accuracy.
- ArticleOne Useful ThingArticle
Ethan Mollick · Practical writing on how teachers and students are adapting to AI.
Spotted a mistake or something unclear? Report it on GitHub.