A detector score is not proof

Module 1 : What AI already does in your job

Plan du cours

Cette leçon n'est pas encore dans votre langue, la voici donc en anglais.

Petit test

Guess first, no penalty. An AI detector says a student's essay is 80% AI-written. What's your next step?

An AI detector is a tool that guesses whether a piece of text was written by AI. Teachers say these tools leave them second-guessing, and with reason. They make mistakes in both directions: they can flag honest writing and miss copied work.

Even OpenAI, the company behind ChatGPT (the best-known AI chat tool), withdrew its own AI text classifier on 20 July 2023, citing its "low rate of accuracy" (Search Engine Land).

So treat a score as a reason to look closer. The evidence is the student's own thinking: can they explain the essay, and can they show how it was made?

Worked example

A 10-minute talk with a student whose essay was flagged:
1. "Tell me your main argument in your own words."
2. "Why did you pick this example over the others?"
3. "Show me your notes, an early draft or the version history."
4. "What does this word in paragraph 2 mean?"

Version history is the saved record of every change to a document.

A student who wrote it can usually answer. A student who didn't usually can't. Either way, you're judging what they know, not a percentage.

Complétez les trous

Fill the gaps in your plan.

When a detector flags work

I treat the score as . I ask the student to . I decide based on .

Try it

Vos étapes, environ 2 minutes

0 sur 1
  1. 1.Add to your card the 3 questions you'd ask a student whose work was flagged, for one assignment you're setting this term.2 min

Exercices

Question 1 sur 2

Why isn't a detector score enough evidence on its own?

Tuteur IA

Posez vos questions sur cette leçon et faites-vous guider dans l'exercice « Essayez ». Le tuteur fait partie de Learn Pro.

Voir Learn Pro