AI and higher education: teachers' perceptions of students' ethical tensions in online pedagogical methods

Synthesis: Baoyi and Khan interviewed 25 university language teachers in China, online and in person, about the ethical tensions they perceive in their students' use of generative AI across five online language-teaching methods: the direct method, content and language integrated learning, task-based language teaching, communicative language teaching, and community language learning. The semi-structured interviews averaged 51.52 min (shortest 45 min, longest 59 min) and were analyzed inductively with thematic analysis, yielding 20 named tensions, four per method, grouped into a three-domain framework: authenticity of learning (11 tensions), authority challenge (4), and collaborative dynamics (5). The paper reports no effect sizes: its contribution is a qualitative framework of named, method-specific tensions, and its evidence is entirely teachers' self-reported perceptions rather than independently verified student behavior.

Key Findings

• Twenty tensions, four per method. Thematic analysis of the interviews identified 20 unique ethical tensions across the five methods, and the authors state that each method confronted four of them.
• Authenticity of learning is the dominant domain, with 11 of the 20 tensions. It asks whether demonstrated competence reflects real understanding or AI-based performance, and it gathers Academic Integrity concerns such as conversational ventriloquism, conceptual hollowing, Scaffolding collapse, and the authenticity crisis.
• Authority challenge holds 4 tensions, covering shifts in power from the teacher toward AI tools: student autonomy, teacher authority, the translation dependency trap, and the confidence mirage.
• Collaborative dynamics holds 5 tensions around trust, equity, and participation among peers: invisible inequity, the trust erosion cycle, the AI alpha, collaborative hollowing, and the counselor's dilemma.
• The domains interact rather than sit side by side. The authors state that a lack of authenticity can increase difficulties with authority, and that weak teamwork can also weaken meaningful learning; they describe the tensions as method-salient rather than method-exclusive, since the five methods are frequently combined in one online course.
• The tensions are method-specific, not generic. The paper's central claim is that policy built on generic, method-agnostic guidance is unlikely to address the distinct pressures each method places on authenticity, authority, and collaboration.

How the study was done

The design is a qualitative thematic analysis of interviews, not an experiment, and the paper reports no effect sizes or inferential statistics. Data came from 25 teachers engaged in online teaching at various universities in China, interviewed online and in person through a semi-structured protocol. Teachers who had never taught online were dropped as irrelevant, and the sample was focused on university language instructors who actively used AI-based learning environments and employed the five methods. Participation was voluntary with no time pressure, which the authors say helped reduce social desirability bias; the average interview ran 51.52 min, with the shortest at 45 min and the longest at 59 min. Interviews were recorded with consent and transcribed with "Turboscribe," then cross-checked by the authors.

Coding moved through transcription, initial coding in NVivo with a parallel manual Excel list, first-order coding, and main coding. Tension names were generated inductively rather than drawn from prior literature, and each label was refined through discussion between the two authors until both agreed it represented the coded pattern. The study was approved by the Research Office of Kashgar University (approval code KSU—2026(042), dated February 10, 2026).

The named tensions by method

In the direct method, teachers reported conversational ventriloquism (answers generated in real time, so the exchange misrepresents whose competence is displayed), student autonomy, teacher authority, and memory erosion — students performing well with tools but poorly on AI-restricted tasks.

In content and language integrated learning, the tensions are conceptual hollowing, the bilingual bluff, the translation dependency trap, and the synthesis illusion. One transcript quoted in the paper: "The students are using AI tools to do the work for them, and this means they are not really understanding the subject matter because when I ask them to explain the subject matter, they cannot do it."

Task-based language teaching produced scaffolding collapse, the shortcut paradox, invisible inequity, and task realism corrosion — the last describing teachers pushed toward redesigning activities to be AI-resistant rather than authentic.

Communicative language teaching yielded the authenticity crisis, risk-free fluency, the confidence mirage, and pragmatic blindness, where AI output is grammatically correct but wrong in tone, formality, or culture.

Community language learning produced the trust erosion cycle, the AI alpha, collaborative hollowing, and the counselor's dilemma, all turning on concealed or unequal AI use inside group work.

What this means for practice

• Design assessment around process, not only product. The authors call for process-based assessment standards precisely because authenticity-related tensions dominated their data; a polished final artifact is the least diagnostic evidence you can collect.
• Make AI use visible in group work. The collaboration-related tensions (trust erosion cycle, collaborative hollowing) follow from concealment, so transparency requirements for group tasks address the mechanism rather than the symptom.
• Set teacher-AI role boundaries explicitly. Authority-related tensions recur across methods, and the authors advise teachers to act as guides and "consolers" who steer students toward using AI as an assistive tool rather than an exclusive one.
• Write method-specific guidance, not one campus-wide AI policy. The study's main recommendation to policymakers is to move beyond general AI strategies and tailor ethical policy to each pedagogical method.
• Do not read the framework as evidence of student behavior. It records what teachers perceive; pair it with student self-reports, observation, or AI-use logs before acting on a specific tension.

Limitations

• The sample is 25 teachers in Chinese universities, which the authors say limits geographic scope and may not transfer to other regions.
• The study covers online pedagogy only; the authors call for future work in physical classrooms.
• The findings rest entirely on teachers' self-reported perceptions and should be read as evidence of how teachers interpret students' AI-related behavior, not as verified evidence of what students did.
• Coding and cross-checking were done by only two authors, without formal inter-rater reliability statistics, and because the sample was language instructors using five language-acquisition methods, the authors note the findings may not generalize to other content areas.

Citation

Baoyi, W., & Khan, F. U. (2026). AI and higher education: teachers' perceptions of students' ethical tensions in online pedagogical methods. Frontiers in Psychology, 17, 1881032.