How artificial intelligence transforms the feedback ecosystem in higher education

Synthesis: Bearman and colleagues argue that AI does not slot into Feedback as another source of comments; it reworks the web of relations among students, educators, materials and peers, and they borrow Barnett's ecological framing to call that web a feedback ecosystem. Their evidence comes from three two-hour online workshops with 12 students and 18 educators, run so that both groups could hear each other's firsthand accounts. The findings are asymmetric in an instructive way. Students treated AI as an additional but flawed information source, weighed it against rubrics, lecture notes, discussion boards and educator comments, and mostly privileged educators they trusted, while still using AI as a holding pattern when staff feedback was slow. Educators were far less engaged, mostly piloting AI in single activities rather than integrating it. The paper's constructive proposal is a shift from human-in-the-loop oversight of AI output to humans-as-the-loop: people helping people build stronger feedback relationships over time.

Key Findings

1. AI changes relations, not just the supply of comments. Students generated and interpreted AI feedback in association with rubrics, lecture notes, discussion boards and educator comments, so each source altered the meaning of the others rather than being interchangeable.
2. Students preferred educator judgment but often could not reach it. They privileged comments from educators they had relationships with, and one student described booking a meeting that took over a month and never happened, ending in a few email bullet points.
3. AI functioned as a holding pattern under time pressure. Rather than wait for human responses, students continued substantial work with AI output and asked on the discussion board in parallel, editing later if the AI had misled them.
4. AI was used to translate thin human feedback. One student pasted a generic set of comments into a chatbot to work out what to do differently next time, and described reaching a better understanding of what the educator had meant.
5. Its presence made peer feedback look suspect. One student said she no longer trusted peers' discussion board comments because they had probably had AI write the answers, illustrating a loss of faith in peer feedback as a channel.
6. Educators were cautious and mostly at pilot stage. Their caution was about ceding authority to general-purpose AI, and one institutional trial where students prompted the tool and received finished answers rather than commentary on their own work.
7. AI both increased educator workload and produced distance. Commenting on longer AI-assisted submissions took more time without efficiency gains, and the authors sensed growing distance between staff and students.

Why time is the hinge

Temporality runs through the analysis. AI output is privileged because it arrives immediately, and the authors note that speed can descend into diminished quality when efficiency is over-prioritized. That trade is unevenly available: students with lives that make it hard to reach the preferred teacher feedback are the ones most likely to accept the efficient option, which is why the paper treats timing as an equity question and not only a matter of convenience. Time also changed on the educator side, where AI-assisted submissions lengthened grading without easing it, and where the significant investment required to rework feedback practices with AI competes with everything else. Set against that, the students who gave two hours to a workshop about improving their own feedback practice suggest feedback work is not only instrumental: for some it is time invested in a meaningful relation.

Human-as-the-loop, not human-in-the-loop

The automated feedback literature proposes a human-in-the-loop who oversees and amends machine-generated comments, and that framing now runs through AI assessment work. The authors argue it is insufficient because current systems address outputs rather than learning trajectories: an AI provides instant information about a task, while an educator relates, challenges and evaluates a learner against a trajectory over time. Their alternative puts people in the business of helping other people build stronger feedback relationships, whether educator, a longitudinal cohort of peers, or students managing their own network of sources as a form of Feedback Literacy. The paradox they draw out is that concentrating on the whole ecosystem brings the teacher back into relief rather than displacing them: educators introduce students to feedback relations, use assessment design to create immediately useful ones, and can point to relationships beyond the course. Peer or family relations can serve the same function, which matters because professionals rely on a feedback community long after university.

What this means for practice

• Instructors. Ask what a feedback source is doing in a student's network rather than whether it can substitute for your comments, and design assessments so that the relations you want students to have are immediately useful.
• Assessment designers and professionals. Treat rubric clarity and the availability of human response as part of the same system: thin or late human comments push students toward AI, and the same students have the least room to wait.
• Researchers. The ecosystem frame is offered as a sensitizing concept for qualitative work, so studies that record only one actor's experience will miss the shifts in relations reported here.

Limitations

• The study is exploratory with a limited participant pool from a single institutional context: 12 students and 18 educators from one university attending workshops on AI and feedback.
• Participants self-selected through interest in AI and feedback, so students and educators who avoid AI may hold different views.
• The authors state explicitly that the ecosystem conceptualization did not emerge inductively from the data and is not wholly original, even though the analysis itself was inductively derived.

Connected Concepts

• Feedback
• Feedback Literacy
• Formative Assessment
• Human AI Collaboration
• Teaching
• AI Literacy
• Equity
• Self-Regulated Learning
• Generative AI
• Large Language Models (LLMs)
• Higher Education

Connected Articles

• The care-full craft of feedback in an age of generative AI — feedback encounters as matters of care when generative AI mediates them
• TeachingCoach: A Fine-Tuned Scaffolding Chatbot for Instructional Guidance to Instructors — a chatbot built to give instructors feedback on their own teaching
• Access is Not Enough: Human Support Improves Engagement with AI Tutoring — human support as the condition for engagement with AI tutoring
• Feedback futures: beyond the limits of human and GenAI capacities — what human and GenAI feedback capacities can and cannot reach
• Comparing Generative AI and teacher feedback: student perceptions of usefulness and trustworthiness — student perceptions of usefulness and trust in AI versus teacher feedback

Citation

Bearman, M., Corbin, T., Walton, J., Tai, J., Nieminen, J. H., Dawson, P., Crawford, N., & Boud, D. (2026). How artificial intelligence transforms the feedback ecosystem in higher education. Assessment & Evaluation in Higher Education. Advance online publication.