Toward Partnership Ethics for Generative Artificial Intelligence in Higher Education

Synthesis: A mixed methods study at one Swedish university asked what ethical concerns actually shaped how students and teachers used, or declined to use, generative AI. A census survey of roughly 10,000 students and 800 teachers in a social science faculty returned 541 student and 128 teacher responses, followed by five focus groups with 13 students and two teachers. The sharpest number is a gap: 80% of students had used genAI in their studies, but only 20% of teachers had actively integrated it into teaching, so the practice had outrun the institution's response. Ethics came up unprompted for 31% of participants (209 of 669), and those concerns ran well past academic integrity: participants worried about critical thinking, Creativity and autonomy, about moral character, and about a tendency to treat model output as objective. The authors read these findings through environmental ethicist Carolyn Merchant's partnership ethics (mutual interdependence, respect for diversity, inclusiveness, accountability and a shared commitment to long-term Sustainability) and argue it offers a better frame for institutional decisions than a compliance-and-detection stance.

Key Findings

• Ethics surfaced without being asked. 209 of 669 participants (31%) raised ethics in free-text survey comments or interviews, though no question asked about it directly. Of those 209, 158 were students and 51 were teachers. Only 13, all students, took an entirely positive view of genAI; the remaining 94% (n = 196) were ambivalent or negative. The authors read that as evidence the concern is live rather than imposed by the design.
• Developmental harm was the sharpest concern. 51% (n = 106) worried that genAI use could hamper individual development by affecting a person's knowledge, skills, capacity for "free thinking" and moral character. Teachers feared students would "learn less", "stop thinking" and "take shortcuts"; students feared losing what one called "actual acquired knowledge". One teacher put it bluntly: "If AI interrupts that, it's brain poison."
• One asymmetry went unremarked by participants. Many worried about what genAI affected in what students knew. No one voiced concern about its impact on what teachers knew.
• Moral character entered the conversation. 22% (n = 47) worried that genAI made users "lazy", a result the authors connect to the "moral judgements" reported in an Australian focus group study. A small number instead saw genAI as a way for the individual to be "effective".
• Dependency replaced the resistance other studies found. Where the Australian students in Bearman et al. (2025) "generally wanted agency" and "resisted dependency", several students here described becoming dependent, and said so as a loss. One wrote that genAI "gives a dopamine kick because everything takes less time" and "rewards bad behaviours - or rather ones that make the learning process more difficult in the long run."
• Students were more positive than teachers, with one shared line. Students took a more favorable view overall, but both groups agreed that submitting an entirely AI-generated text as one's own work is wrong.
• Most did not see themselves as agents in the decision. 86% of the 209 who raised ethics (n = 179) did not describe themselves as having an active role in determining ethical genAI use. Teachers were likelier than students to see themselves as ethical agents; students wanted the university to supply "lectures", "guidelines" and policy.
• Environmental cost appeared, but rarely. Three teachers raised the "huge environmental cost" and "theft of data" behind model training, as did three students, while the university's policy documents stayed silent, a gap the authors report as mirrored internationally.

How participants used and valued genAI

Students used genAI far more than they valued it for writing, and the pattern matters for anyone reading the ethical concerns as technophobia.. Among students who did use it, the valued uses were cognitive and preparatory: finding inspiration (73.6% agreeing at least somewhat), saving time (64.7%), studying and reviewing materials (56%), and correcting grammar or spelling (50.6%). Only 11.6% agreed that genAI offered "no benefits" at all.

Writing was the exception. More than eight in ten students did not see genAI as advantageous for writing essays or reports, the very use that dominates institutional anxiety about integrity. Students described the tradeoff themselves. One explained the pressure: "it takes me three minutes to understand the first sentence. And then I see that it is going to take me weeks to read this entire text, which is one of 10 texts that I should read." Two participants mentioned using genAI to cope with neurodivergence; another named the cost, saying genAI "prohibited me from actually reading and finding out myself what is important in the text".

The harm participants feared was not rule-breaking but Trust erosion. Students said they trusted classmates less, and several described classmates using genAI in group work without telling the group, which they experienced as "detrimental to the trust and ethics of academia". A few suspected teachers used genAI but withheld that from students, noting that faculty had room to experiment while "students are not given that kind of room to test it out."

What participants proposed

Almost nobody asked for detection. What participants described as useful was dialogue, and a minority had built it. 61% (88 students and 39 teachers) framed their concern around a greater social good, from the classroom to the university to society. 14% (22 teachers and 8 students) named collective experimentation, discussion and reflection as the way to navigate genAI, and one student asked for help removing "the stigma surrounding this subject". Thirteen participants (eight teachers, five students) reported classroom activities that had helped them decide. A teacher had students "compare and critically discuss and assess the output by AI", whether code for R or Stata or a discussion of a substantive topic, alongside an explanation of how chatbots work. Students described seminars comparing genAI's work with their own, which let them see, as one said, "how faulty a lot of the information was".

The institutional tension here is worth naming. Students wanted the university to "do more", while the university's own position is that decisions belong at department, program and course level. Silence filled the gap. Teachers kept quiet partly from time pressure and partly from their own uncertainty, and students described a "culture of secrecy" in which "many students are scared to talk about how much they use it".

Study Design

The study ran during the 2024-2025 academic year at the University of Gothenburg and was supported by internal grant funding awarded after institutional review. It combined an exploratory online survey, administered in English, with focus group interviews conducted in Swedish and English. Work followed the Swedish Ethics Review Act (2003:460) and the Swedish Research Council's guidelines, and all participants were informed of the study's purposes and consented.

The survey went to the entire student body and all teachers in the faculty through departmental email lists, making it a census rather than a sample. Of the 70 students and 13 teachers who volunteered for follow-up, five focus groups were held with 13 students and two teachers; the authors are explicit that this self-selection is a limitation.

Respondents broadly reflected the faculty's composition. 69% of student and 55% of teacher respondents identified as female; 72% of students were at the undergraduate level, which normally takes three years in Sweden; and 80% of teacher respondents reported more than five years of teaching experience. Familiarity with genAI was high, with about 30% of each group calling themselves very familiar and students twice as likely as teachers to claim extreme familiarity (7.2% versus 3.1%). The authors read participants' comments through dominant ethical theories, noting that most reflected egoism or utilitarianism rather than a relational ethic.

What this means for practice

• Treat silence as the problem, not the absence of a problem. The authors' third practitioner note is that silence around teachers' and students' genAI use creates uncertainty and suspicion, reducing opportunities for agency and collaboration. The study documents the mechanism: unclear boundaries, a culture of secrecy, and students afraid to disclose use. A policy that says nothing while 80% of students use the technology leaves both groups guessing, and the 86% who did not see themselves as agents in the decision are asking the institution to close that gap at the level where decisions actually sit, which makes it a question of policy design rather than communication.
• Move beyond the cheating frame. Their second practitioner note states that cheating and plagiarism is too narrow a frame to support ethical (non)use. The study's participants raised development, character and environmental questions that an integrity-only policy cannot address, and the use they themselves rejected was essay writing, not study support.
• Build dialogue into the decision, not just the rule. Partnership ethics asks all members of an academic community to navigate uncertainty together through ongoing dialogue. The 13 participants who described working classroom activities point the same way: have students critique model output against their own work and explain how the system works, rather than only forbidding it.
• Put environmental consequences on the agenda. The fourth practitioner note argues that discussions omitting the environmental costs of developing and deploying these technologies fail to address the concerns teachers and students actually hold. Three teachers and three students raised it unprompted while policy documents were silent.
• Use partnership ethics as a lens on existing policy. Merchant's five principles (mutual interdependence, respect for diversity, inclusiveness, accountability and shared commitment to long-term sustainability) give institutions a way to audit current guidance against the range of concerns the study documents.

Limitations

• A single case, at a single moment. The study is a case study of one social science faculty at one Swedish public university during the 2024-2025 academic year, and the authors are careful that findings reflect that particular time and place.
• One discipline only. The research focused exclusively on a social science faculty, so the views recorded cannot be assumed to generalize across a university's other faculties.
• Low and self-selected participation. Survey response rates were comparatively low and teacher participation in the focus group interviews was low, with only two teachers across five groups. Participants self-selected, and the authors cannot know whether their views were held by non-participants.
• A moving target. The research concerned a rapidly evolving technology, so the specific uses and concerns recorded describe a technology that had already changed by publication.
• Generalisation is unresolved. The authors find significant resonance between their context and international scholarship on genAI, but state that more research is needed to determine how generalizable the findings may be.

Citation

Gillette, M. B., Wackenhut, A. F., & Olsson, E. M. (2026). Toward Partnership Ethics for Generative Artificial Intelligence in Higher Education. Journal of University Teaching and Learning Practice, 23(6). Open access under CC BY-ND 4.0.