Did AI Write This? The Question Education Must Go Beyond
I asked Claude whether it could detect AI use in texts.
“Not reliably, no,” it replied.
It explained that stylistic clues are weak evidence, that editing can confuse detectors, and that human writing may be wrongly classified as machine-generated.
That admission matters. But it opens a larger educational question:
What do we actually need to establish when we assess a student’s work?
AI detection and human authorship are different questions.
A detector attempts to infer how a text was produced. Education must also establish whether the student understands its claims, has exercised judgment, and can take responsibility for what it says.
A student may use AI to translate, reorganise an argument, challenge an interpretation, or improve a draft—and still make the decisive intellectual choices.
Another student may write every sentence without AI while merely repeating ideas they have not understood.
The presence or absence of AI does not settle the question of authorship.
The limits of detection
We should be precise. Some detectors perform well under particular test conditions. That does not make a detector score proof about an individual student’s work.
Research has documented false positives affecting non-native English writers and shown that rewriting can substantially alter detection results. These findings concern particular systems and conditions; they are reasons for caution, not a claim that detection can never improve. Research: Liang and colleagues.
Comparing a student’s work with earlier writing can provide context. But a change in style is still something to explore, not a verdict. Students learn, receive help, experiment, and revise.
From the finished answer to the authored trace
In Memecraft, the authored trace is the record of how meaning formed:
What did the student notice?
What did they accept too quickly?
Which suggestion did they reject?
What changed after checking the evidence?
Where did their own judgment enter?
The Field Report collects these moments over time: observations, sources, disagreements, revisions, and decisions.
But the trace alone is not enough. A polished logbook can also be generated. A prompt history does not establish understanding.
This is why we need the Defend moment.
The teacher selects one actual decision from the student’s work and opens a conversation:
- What led you to this interpretation?
- Which alternative did you consider?
- What did you change, and why?
- What evidence would make you reconsider?
- What are you prepared to stand behind?
The aim is to make judgment available for examination. A student should have room to hesitate, acknowledge uncertainty, and revise an answer. Fluency alone is not understanding.
The teacher as witness
Here, the teacher acts as witness: attending to how the student connects an observation to a claim, responds to a challenge, and accepts responsibility for a choice.
This does not reconstruct every keystroke or prove that no AI was used. It provides evidence of understanding and authorship through dialogue.
AI-use rules still matter and should be explicit. But compliance with those rules and the educational quality of a student’s judgment require separate attention.
MoMo starts where detectors stop
MoMo examines claims, distinctions, evidence, and reasoning. It does not certify who wrote the words.
Its educational purpose is to help students distinguish fact from interpretation, metaphor from mechanism, confidence from evidence, and repetition from understanding.
That is the movement from media literacy toward symbolic literacy: learning to examine how an answer becomes meaningful—and whether we have accepted it before understanding the question.
The Pause creates space for that examination.
The Field Report records it.
The Defend moment brings it into dialogue.
“Own your story” means being able to explain the choices that made it yours—and to revise them when there is reason to do so.
Poul Goldschadt
Memecraft – Digital Phenomenology
