'AI Should Help Them Learn, Not Learn for Them': University Staff Perspectives on the Role of Generative AI in Education

Synthesis: Cross-sectional survey of 76 academic staff at an Australian university (conducted August 2023–June 2024) exploring educators' perspectives, attitudes, and experiences with generative AI in learning and teaching. While staff saw clear productivity gains — streamlined administration, curriculum and Assessment design, and student-centered activities — they voiced substantial concerns about academic integrity, ethical implications, the erosion of core skills such as critical thinking and Creativity, and widening equity gaps in student access and digital proficiency. A recurring refrain — that AI "should help them learn, not learn for them" — points to an urgent need for governance frameworks, structured professional training, interdisciplinary collaboration, and the active inclusion of student perspectives. Published in the Journal of University Teaching and Learning Practice, DOI https://doi.org/10.53761/d5n2yh02. (CC BY-ND license — see log.)

Key Findings

• Opportunity without integration. Most staff (61.8%) had used GenAI at work and rated the experience positively, yet few used it frequently or had woven it into teaching and curriculum design (only 40.4% had integrated it on an occasional or regular basis). The dominant perceived benefit was enhanced productivity and efficiency, especially for brainstorming, background research, and content summarization.
• Academic integrity is the central fear. By far the most common concern was students using AI to plagiarise or cheat on assessments, with some staff seeing any attempt to stay on top of it as futile given the speed of technological change.
• AI as an impediment to genuine learning. Roughly a third saw GenAI as a threat to learning quality, worried students would use it as a "shortcut" and miss the cognitive work needed to build writing, referencing, critical analysis, and creative skills. Staff flagged inaccuracy and bias (hallucinations) and students' poor AI Literacy as enabling uncritical overreliance — captured in the study's title sentiment that AI should "help them learn, not learn for them."
• Workload and labor concerns. Some staff reported that GenAI increased their workload (e.g., marking AI-generated essays) and feared that university adoption would be used to replace human functions such as marking, feedback, and student services rather than genuinely assist them.
• A clear skills and support gap. While most staff wanted training and supported embedding AI Literacy into the curriculum, over three-quarters said adequate resources were not provided, and the majority reported no clear institutional guidelines on ethical use. Only one respondent rated their university as above average for AI utilization compared to peers.
• "A tool, not a solution." Across open-ended responses, staff broadly agreed that GenAI should support rather than replace human educators, preserving teacher–student relationships, disciplinary expertise, and human oversight, while managing routine administrative "shovel work."

Study Design & Method

The study used an exploratory cross-sectional design with an online survey (hosted on RedCap) built on Kelly et al.'s (2023) Generative Artificial Intelligence Survey, trialled with colleagues before launch. Academic staff from ten schools at an Australian university were recruited through Associate Deans of Learning and Teaching via email, Teams, and campus posters. The instrument contained up to 55 items across six sections — demographics, digital and AI literacy, training and university adoption, prompt engineering, Assessment development, and academic integrity, ethics, and equity — with branching logic, closed (multiple-choice and Likert) and open-ended items. After listwise deletion, 76 of 94 initial responses were analyzed. Descriptive statistics were computed in SPSS 28; because of the small sample and low cell counts, no inferential statistics were run. Open-ended responses were analyzed using an inductive thematic approach (Braun & Clarke), with codes grouped into broad themes and saturation judged when repeated review yielded no new codes.

What this means for practice

• Instructors. Set your own course-level expectations instead of waiting for institutional guidance: more than three-quarters of the 76 staff surveyed said adequate resources were not provided, and the majority reported no clear institutional guidelines on ethical use.
• Instructors. Keep the cognitive work visible in task design. Staff reported students using GenAI as a shortcut around the writing, referencing, critical analysis, and creative effort that builds skill, and flagged poor student AI Literacy as what enables uncritical overreliance on inaccurate or biased output.
• Instructors. Ask for training that pairs technical skill with pedagogical strategy, and embed AI literacy in the curriculum: most respondents wanted training and supported embedding it, yet few felt confident teaching students how to prompt.
• Instructors. Redesign assessment around higher-order thinking and evaluative judgment rather than leaning on detection, and protect the human, dialogic role of Feedback: staff saw marking AI-generated essays as added workload and judged GenAI feedback to lack nuance and personalization.
• Instructors. Do not assume uniform access or proficiency. With a socioeconomically diverse cohort, staff linked GenAI to widening digital and educational disparities, so build tasks and alternatives that do not presume paid tools or confident use.

Limitations

• Cross-sectional online survey at a single Australian university: 94 responses were received and 18 were removed by listwise deletion for not proceeding past the demographic items, leaving 76 academic staff analyzed — a sample the authors themselves call modest, affecting external validity.
• Recruitment was voluntary and self-selected — staff with greater interest in GenAI were more likely to respond — and respondents do not proportionally reflect the institution's broader staff profile, so the sample is illustrative rather than representative of the sector.
• Because the sample was small with low cell counts, only descriptive statistics were computed in SPSS version 28 and no inferential tests were run.
• Data were collected between August 2023 and June 2024; the authors note that AI technologies and policies have evolved rapidly since, so some findings may no longer reflect current practice, and the study captures self-reported attitudes and experience rather than observed teaching or student outcomes.

Citation

Enright, H., Horvath, D., Petrovic, K., & Šarkić, B. (2026). "AI should help them learn, not learn for them": University staff perspectives on the role of Generative AI in education. Journal of University Teaching and Learning Practice, 23(7).