In the rapidly evolving landscape of higher education in the United Kingdom, a pressing concern has emerged among academics regarding the increasing reliance on artificial intelligence (AI) for teaching assessment and student feedback. Recent research highlights the dangers of outsourcing these critical functions to AI systems, warning that educators risk losing the nuanced 'craft of feedback'—a skill honed over years of professional practice that involves far more than automated comments. This craft encompasses relational dialogues, contextual understanding, and empathetic guidance tailored to individual student needs, elements that generative AI struggles to replicate authentically.
The debate intensified with a new paper published in Assessment & Evaluation in Higher Education, which argues that while AI promises efficiency amid growing class sizes and resource constraints, the productivity gains represent a 'false economy.' By delegating feedback generation to tools like ChatGPT or specialized marking platforms, universities may inadvertently erode educators' expertise, student-teacher relationships, and the pedagogical processes that foster deep learning. As UK higher education faces financial deficits and workload pressures—with many institutions implementing job cuts—the allure of AI as a cost-saving measure is strong, yet experts urge caution to preserve the human essence of education.
This issue is particularly relevant in the UK, where surveys indicate near-universal AI adoption among students, amplifying the need for balanced policies. The Higher Education Policy Institute's (HEPI) 2026 Student Generative AI Survey reveals that 95% of undergraduates use AI in some form, with 94% applying it to assessed work, underscoring the urgency of addressing these risks without stifling innovation.
Understanding the 'Craft of Feedback' in Higher Education
The 'craft of feedback,' as conceptualized by researchers like Professor Naomi Winstone from the University of Surrey, refers to the intricate, professional artistry educators employ in assessment practices. Unlike a mere product—such as a written comment on an assignment—this craft is a dynamic process involving multimodal interactions: classroom discussions, impromptu corridor chats, iterative revisions, and personalized encouragement that considers a student's 'back story' and future aspirations. For instance, an experienced lecturer might soften critique for a first-year student overwhelmed by transition challenges or highlight growth potential in a mature learner balancing family commitments.
This process draws on two decades of scholarship emphasizing feedback as relational and meaning-making, rather than transmissive. In UK universities, where modules often feature diverse cohorts from Russell Group elites to post-1992 institutions, the craft adapts to cultural, socioeconomic, and disciplinary contexts—vital in fields like humanities where subjective interpretation reigns, or STEM where precision matters. Outsourcing to AI risks reverting to outdated models, producing generic outputs that ignore these nuances and fail to build trust essential for students to act on advice.
Winstone and colleagues stress that feedback encounters are 'matters of care,' demanding ethical sensitivity and relational labor. Without this, students receive 'dangling data'—information without uptake—potentially widening attainment gaps, especially for underrepresented groups less adept at navigating multiple feedback modes.
Risks to Educators: Skill Atrophy and Professional Identity
One of the most overlooked dangers is the impact on academics themselves. If AI handles routine marking, lecturers may stagnate in their feedback proficiency, losing the tacit judgment that defines expert practice. Winstone notes, 'What’s lost for teachers if they don’t engage in assessment and feedback? It’s our primary source of insight into student struggles and our own teaching effectiveness.' In cash-strapped UK universities, where deficits persist into 2026, this could exacerbate burnout, as remaining human elements become overburdened with complex cases.
- Skill stagnation: Reduced practice leads to 'unthinking' habits, diminishing nuanced ethical decisions like contextualizing feedback for personal hardships.
- Professional devaluation: Marking as 'core academic work' risks peripheral status, mirroring broader casualization trends in UK higher education.
- Feedback loop disruption: Teachers gain less diagnostic data on cohort needs, hindering curriculum refinement.
A University of Bath study echoes this, warning AI erodes 'human capital thinking' across workplaces, including academia, urging 'learning vaults' to safeguard creativity.
Student Learning Under Threat: From Connection to Isolation
For students, AI feedback lacks the psychological connection that motivates engagement. Research shows human comments are trusted more due to perceived empathy, while AI risks 'hallucinations,' biases from training data, and over-generalization—issues amplified in diverse UK cohorts including international students facing language barriers. The HEPI survey found 49% report improved experiences with AI for time-saving and clarity, but 12% now include AI text in submissions (up from 3% in 2024), raising integrity concerns.
Over-reliance could erode critical thinking, as students outsource sense-making. In a polarized landscape, 65% note assessment changes, with anxiety over misconduct accusations prevalent. For full details on student AI use, see the HEPI 2026 report.
Photo by Toa Heftiba on Unsplash
Real-World Case Studies: Jisc Pilots in UK Universities
The Jisc AI Marking and Feedback Pilot, involving 15 UK institutions like London South Bank University (LSBU), tests tools such as Graide and Keath. At LSBU, AI generated draft feedback on essays, with academics editing for personalization—saving time on low-level tasks while preserving oversight. Early reflections show suitability for formative assessments, but summative use lags due to stakes. Staff valued consistency gains, students appreciated timeliness, yet concerns persist over devaluing human judgment.
| Institution | Tool | Outcomes |
|---|---|---|
| LSBU | Graide | Time savings on drafts; improved feedback volume |
| 10 others | Graide/TeacherMatic | Workload reduction; need for human review |
Organizers assure no human replacement intent, emphasizing hybrid models amid union worries on staffing. For pilot insights, explore Jisc case studies.
Stakeholder Perspectives: Academics, Students, and Administrators
Academics like Winstone advocate 'care-full' integration, prioritizing trust and equity. Students are divided: HEPI data shows 68% deem AI skills vital, yet only 36% feel institutionally encouraged. Administrators, facing £307m deficits, eye AI for efficiency but risk backlash—UCU strikes highlight workload tensions.
- Academics: Protect craft via training; use AI for analytics.
- Students: Demand literacies; fear bias/misconduct flags.
- Admins: Balance costs with quality; policy gaps in 49% unis.
Broader Implications for UK Higher Education
Amid 2026 crises—visa rejections, enrollment drops—AI outsourcing could widen inequalities. Biased algorithms might disadvantage non-native speakers, echoing A-level algorithm scandals. Ethical frameworks, like University of Exeter's, stress integrity policies.
Equity risks loom: Free AI tools disadvantage premium users; 77% rely on basic versions per surveys.
Solutions: A Manifesto for Care-Full Feedback
The paper outlines 10 principles: Multimodal entanglement, relational focus, trust-building via literacies. Recommendations include:
- Hybrid models: AI for drafts, humans for synthesis.
- Staff development: Feedback craft workshops.
- Student induction: AI evaluation skills.
- Equity audits: Access, bias checks.
Read the full open-access paper for the manifesto.
Photo by Annie Spratt on Unsplash
Future Outlook: Thoughtful Integration Ahead
By 2030, AI could transform UK HE if guided by scholarship. Pilots suggest hybrids enhance outcomes; with OfS oversight, policies may standardize. Ultimately, preserving the human spark ensures feedback remains a catalyst for growth, not a commodified output.