AI and I—Distributed Authorships in the Swedish Upper Secondary Writing Classroom

Synthesis: Silverdal, Hermansson and Lindgren examine what authorship becomes when upper secondary students write with GenAI in an ordinary Swedish language classroom. The empirical material is three students' recorded writing sessions on a Kafka-inspired short-story assignment, plus complementary interviews, analyzed with rhizomatic mapping and Deleuze and Guattari's notions of assemblage, affect and becoming. Three assemblages emerge, and the contrast between the first two is the paper's finding. In one, an eight-minute event, the student delegates the story to ChatGPT, declares the output better than he could write and copies it across, repositioning himself as the less capable writer — yet the analysis refuses to read this as an absence of ability, tracing instead how he interprets the task, negotiates the word count with a peer, and reorients the genre through successive prompts. In another, co-writing appears as intense critical labor. Because the same activity produced both, the authors argue that distributed authorship is a relational accomplishment rather than a per-student trait, and that teacher guidance is what decides whether writing with GenAI is equitable.

Setting and Method

The study sits inside the regular curriculum. A teacher introduced Kafka's Metamorphosis, discussed its atmosphere and themes, and asked students to write an 800–1000 word short story in past tense with a narrative turning point, imagining waking as an animal aligned with their personality. Writing sessions were screen-recorded, and the analysis treats the recordings and the interviews as "continuations of the writing events themselves" rather than neutral representations of them. Consent was handled as ongoing practice, with attention to verbal and non-verbal cues and the right to withdraw.

Three writing events were selected purposively — not for representativeness but because they made different configurations of authorship visible — and mapped through three layers of analysis: content logs documenting temporal, embodied and material progression (prompting, reading, revising, peer interaction, bodily position toward the screen), Multimodal AI transcription tracing how prompts, responses, speech, silence, gesture and tool shifts related to one another, and finally narrative reassembly. Pauses mid-sentence, prompt reformulation, hesitation before accepting a suggestion and screen-switching are treated as significant reconfigurations rather than interruptions, and affect is read as shifts in entities' capacity to act.

Three Assemblages

"AI is my writer." In the shortest event, roughly eight minutes, the student opens a browser, prompts ChatGPT with a long hybrid instruction combining narrative imagination, genre labels, stylistic cues and an embedded word count, asks for a fairytale register, scans the output, says "Damn, I wish I could write like this!" and copies it into the document. The analysis is careful about what this is. The student is positioned as less able by his own account, and his textual engagement is limited — but the event does not read as simple absence of writing ability: he interprets the assignment, negotiates length with a peer who supplies the requirement, and directs the genre. The authors connect this to evidence that students who already write well use GenAI productively while weaker writers may not recognize what it offers, and to findings of reduced cognitive engagement when GenAI solves the task rather than supporting it.

"AI and I as co-writers." A second student's writing appears as "intense labor of critical engagement," with an iterative, curious prompting pattern that the authors align with critical AI literacy: the GenAI is interrogated rather than obeyed, and the student's own knowledge of how to engage shapes the process. Co-writing here is reciprocal and distributed rather than delegated.

"AI and I—no thanks!" The third assemblage foregrounds resistance: authorship is articulated by reconfiguring writing practice away from the tool, with AI experienced as a threat to writing, Creativity and — in this student's framing — the climate.

None of the three is offered as a stable category. They are analytically constructed assemblages that show what authorship and writing ability can become, and the point of presenting them together is the coexistence: the same assignment, teacher, room and tool produced all three.

Why the Contrast Is the Finding

Across the assemblages the authors read three themes: equity, authorship and Ethics. The equity argument is the sharpest. Because GenAI rewards exactly the competence it threatens — knowing how to interrogate a text, judge an output, notice what is missing — students who are already strong writers extract more from it, and the student who found the output better than his own work was the one least positioned to use it well. That is a mechanism for widening rather than narrowing gaps, and the authors conclude that guidance is what decides which side of it a class lands on.

Authorship, meanwhile, emerges as distributed across students, model, interface, teacher, curricular expectations and peer talk rather than residing in the person at the keyboard. Peer joking and side conversations are shown to be constitutive of the writing event, not interruptions to it. The policy implication the authors draw is that questions of authorship and of what counts as writing ability have to be settled in the classroom and in assessment design, not only in integrity rules.

What this means for practice

• Instructors. Do not infer a student's writing ability from whether they used AI. The weakest-looking event here still contained task interpretation, negotiation and genre direction, and the same lesson produced a critically engaged co-writer.
• Instructors. Teach the interrogation of output explicitly, because the students who most need it are the least likely to do it unprompted. Leaving guidance implicit rewards prior writing strength.
• Instructors. Make the assignment's constraints do pedagogical work. Here the word count and genre became relations the student manipulated — visible handles for a teacher to attach Reflection to.
• Researchers. Treat pauses, reformulations and hesitation as data rather than noise. Most of what distinguished the assemblages was in the moments between prompts, which a chat log alone would not record.

Limitations

• Three students in one Swedish upper secondary classroom completing one Kafka-inspired assignment, selected purposively for analytical productivity rather than representativeness; nothing here estimates prevalence.
• Screen recordings capture what happens at the machine, so peer talk is registered only as far as microphones and frame edges allow, and the analysis of affect is interpretive.
• The three assemblages are analytical constructions, and the authors say so; readers should not treat them as a typology of students, and no student was followed across tasks to test whether a pattern is stable.
• Interviews were conducted alongside the recordings but the paper's claims about reasoning rest substantially on the recorded events, and the interviewer's presence is itself treated as an affective force in one assemblage — which is honest but leaves reactivity unquantified.
• No outcome measure exists. Nothing in the study establishes whether any of the three configurations produced a better or worse story, or more learning.

Citation

Silverdal, S., Hermansson, C., & Lindgren, E. (2026). AI and I—Distributed authorships in the Swedish upper secondary writing classroom. Postdigital Science and Education.