Trust over AI control: Redesigning assessment in UK higher education

Generative AI hasn't just entered the classroom - it has shifted the fundamental unit of value in higher education.

At our recent customer roundtables in Birmingham and London, I sat in rooms filled with a sense of urgency. The consensus was clear: the old authorship question of "Did the student write this?" is no longer enough.

Sitting down with leaders from The Open University, LSE, Queen Mary University of London, and the University of Manchester, I saw a realisation take hold. AI isn't the villain here; it is the catalyst for a transformation that was already overdue. An expansion of academic integrity (authorship) to learning integrity (demonstrable knowledge) asks both whether a student produced the work and how they developed their thinking, to get to the very heart of learning.

How are institutions defining clear "rules of the road" for responsible AI use?

We can move from AI policing to responsible use of AI by giving students clear taxonomies that tell them exactly when and how AI belongs in their work.

In London, Professor Alastair Robertson (Queen Mary University of London) demonstrated how they are closing the ‘graduate capability gap’ by embedding AI, when appropriate, directly into the curriculum.

The QMUL AI assessment taxonomy

• 🔴 Red (No AI): The task requires independent human performance.
• 🟠 Amber (AI-assisted): AI is permitted for brainstorming or structure, but the heart of the thinking must remain human.
• 🟢 Green (AI-integrated): Critical engagement with AI is a core learning outcome of the assessment design.

Whether you call it a ‘traffic light system’ or ‘AI assessment taxonomy’, the goal is always the same: ensuring students aren't scared to use technology but taught to use it responsibly.

Why is the question of "who owns the thinking" central to future assessments?

As AI gets better at generating polished prose, the value of a degree shifts toward the uniquely human ability to defend logic and apply context. Dr. Nayat Horozoglu (London School of Economics) challenged us with a vital question: "Who owns the thinking?"

While AI can spit out a summary in seconds, analytical ownership is much harder to fake. To reclaim that ownership, our attending institutions highlighted a shift toward longitudinal insights: evaluating a student's progress over a longer period rather than judging a single moment in time.

When we measure that leap from research to recommendation, we aren't just checking a box - we’re verifying that the student truly owns their logic.

How can authentic, human-centered assessment scale for large student cohorts?

We often hear that authentic assessment doesn't scale, especially when it involves oral exams or 1:1 dialogue. But when we prioritise active, social learning over passive answer production, those scalable models can become reality.

Professor Paul Shore from the University of Manchester shared a powerful example: Team-Based Learning (TBL). He uses it to counter what he calls ‘metacognitive laziness’.

• Scale: Over 3,000 users in 2025/2026 across Medicine, Dentistry, and Nursing.
• Mechanism: TBL uses a structured cycle of Individual and Team Readiness Assurance Tests (iRATs and tRATs).
• Result: Students must justify their thinking in live, peer-to-peer debates. This makes AI-generated outputs a starting point for discussion rather than the final answer.

What are the hidden barriers to the responsible use of AI for students?

I found our Birmingham discussions on the ‘incentive gap’ particularly revealing. If students aren't motivated to learn, they will use AI to bypass what they perceive as ‘busy work’. This creates a ‘race to the bottom’ where students fear they will be disadvantaged if they don't use AI to keep up with the curve.

The tensions we heard during the roundtables were deeply human:

• Tasks easily offloaded to AI (like summarisation) are often the ones critical for developing critical thinking.
• Students remain focused on what ‘counts’ toward a mark, often ignoring the learning process itself.
• Without clear rules, students fear they will be disadvantaged if they don't use AI.

How can the UK higher education sector move from policing to a model of partnership?

Professor Ian Pickup (The Open University) argues for a principles-based approach. He advocates for student-staff partnerships that move away from "detection theatre" and toward a culture of equity and trust.

This shift in philosophy requires a parallel shift in technology.

To move away from policing, tools must evolve from detectors to platforms that offer assurance of learning to educators. We designed Turnitin Clarity to support this specific transition. By providing a window into the writing process and the evolution of a student's work, it allows educators to gain insight into the balance between human and AI contribution. This turns data into a prompt for a supportive, pedagogical conversation—a partnership—rather than a misconduct hearing.

As Professor Gabrielle Finn (University of Manchester) noted within the context of medical education, where assurance of learning is a matter of professional safety:

This philosophy is the ultimate goal. We are moving toward a future where technology doesn’t just check for originality, but actively helps students and teachers demonstrate that meaningful learning has occurred.

One question to take back to your faculty:

Are your current assessments measuring what a student produced, or what they actually learned?

Curious about how Turnitin Clarity bridges the gap between AI innovation and academic integrity?