The reassurance of ‘the human in the loop’ can be found in many places: in vendor marketing, institutional policies, and in wider discourse around AI. It’s become the standard response to concerns about AI making consequential decisions.
But what does it mean in practice? If a marker glances at AI-generated feedback and clicks “accept”, was a human meaningfully in the loop? If an admissions adviser reviews an AI recommendation under time pressure, with no easy way to see how it was produced, can they realistically contest it?
The encouraging news is that good practice is already emerging. Across the sector, many institutions and individuals are thinking carefully about where human judgement needs to sit and what form it needs to take. What could be developed further is a shared set of practical tools — rules, examples, checklists, review processes, etc. — that help institutions tell the difference between meaningful oversight and a tick-box review.
We think our next pilot could help build exactly that — and we want to consult with members before finalising the design.
A different kind of pilot
Our pilots have evolved alongside the sector and the market. When access to AI tools was scarce, it made sense for pilots to introduce new products, usually with free licences attached. Some of the conditions that made that model work well have changed: most colleges and universities now hold licences for relevant tools, the sector has enough adoption experience to know which products are effective, and the question members are asking has shifted from “is this tool any good?” to “how do we use AI well?”
The free licence itself has also become less of a gift than it initially seems. A licence may be free to the participant, but integrating a new product is never free to the institution. It brings training needs, due diligence and risk assessments, and the work of building a new vendor relationship — costs that land before there’s a weight of evidence the tool suits the use case.
This has been one of the most significant challenges for the pilots, with months of approvals standing between sign-up and substantive work. And when an institution does that work alone and then finds the tool isn’t quite right for them, it has little to show for the effort.
Our current proposal is designed to avoid these issues in three ways. First, participants would work with AI-enabled tools they already have access to, rather than being asked to adopt something new. The focus would be on the features and use cases where institutions want more confidence: where AI functionality is available but not yet used, used cautiously, or not used as effectively as it could be. Second, discovery — working out what meaningful human oversight looks like, what policies and processes it requires, and how to evaluate it — would be treated as valuable work in its own right, not as a hurdle to clear before the “real” pilot begins. Third, the thinking would be done together: institutions developing their governance approaches in parallel, with peer support and shared templates, rather than each rebuilding from scratch behind its own walls. The artefacts the cohort co-produces could then become resources for the whole sector.
At this stage, we are proposing a two-phase model, with a deliberate dividing line between discovery and practice.
Phase A — Discovery
In the proposed model, Phase A would be a structured programme of workshops for member colleges and universities. Together we would explore what “human in the loop” means across different contexts, and what institutions need in order to make human oversight meaningful rather than tokenistic.
We see this as co-production rather than simple consultation. One aim would be to develop a set of concrete materials that could include:
- Formal rules and policy positions
- Guidance and support materials
- Ethical checklists
- Review processes
Each participating institution could also leave with an evaluation approach shaped around the questions it most wants answered, and a project plan it can choose to act on if it wants to continue into a practice phase.
We would want Phase A to be a genuine exit point. An institution that completes the discovery stage and decides not to continue should still leave with policy, guidance, checklists and review processes that are applicable in, or significantly inform, wider uses of AI.
Phase B — Putting it into practice
The second phase would be for institutions that want to go further and test the discovery-stage outputs in practice. We expect this to be a self-selecting stage for those with a suitable use case, enough internal capacity, and an appetite to examine how the proposed resources hold up in live settings.
Participants in this phase would apply the rules, guidance, checklists and review processes to AI-enabled tools and use cases they already have access to, then reflect honestly on what worked, what created friction, and where the resources need to be refined.
We would expect the cohort to be supported through a mix of live sessions and asynchronous discussion, but the exact format and level of support will be shaped through consultation.
The output of this phase could be a more informed and practical framework for keeping humans meaningfully in the loop — evidence-based, tested in practice, and published for the wider membership.
We are also considering whether participants in the practice phase could receive a short bespoke report, setting out how their work demonstrates transferable AI maturity in policy, governance, staff capability and evaluation.
What could the pilot aim to achieve?
- Develop a practical, shared understanding of what meaningful human oversight of AI requires, and how this varies across contexts
- Co-produce policies, guidance, ethical checklists and review processes that institutions could adapt for their own settings
- Explore how those resources might be tested in live practice and refined based on practical learning
- Publish a framework for meaningful human oversight, with supporting resources, for the wider membership
- Support participating institutions to build AI maturity in policy, governance, staff capability and evaluation
How we want to consult
Before confirming the pilot design, we want to hear from members about whether this is the right focus, whether the proposed two-phase structure is useful, and what would make participation realistic and valuable. We will use this feedback to refine the final model, including the level of support, the expectations on participants, and the outputs that would be most useful to the sector.
This consultation form lets you sign up for a drop-in consultation session, submit your views online, or do both. You do not need to attend a session in order to respond, and you do not need to submit written comments if you would prefer to discuss your views in a drop-in.
What about tools and vendors?
Participants should bring AI-enabled tools they already have access to, including tools or features they are not yet using confidently or consistently. The pilot is tool-agnostic, and we are emphatically not evaluating or endorsing products. The tool is the medium; the practice is the subject.
We’re also considering inviting vendors to support the pilot by offering free trials to participants who don’t currently have access to a relevant tool, and vendors may be invited to explain how the human oversight features of their products are intended to work, and to answer participants’ questions. To be clear about the terms of that involvement: this pilot will scrutinise oversight claims, not showcase them. Findings will be published independently, and a vendor’s participation is not an endorsement.
Who can take part?
The pilot is open to all Jisc member colleges and universities. We’d welcome involvement from anyone whose work involves AI supporting judgements or content that a human is expected to oversee — educators and assessment leads, quality and governance teams, student services, researchers, and professional services teams alike. The breadth is deliberate: a framework for human oversight is only credible if it’s been tested across genuinely different contexts.
When might the pilot run?
We expect the pilot to run across the 2026-27 academic year, split equally between an initial discovery stage followed by an optional practice stage for institutions that want to go further. The detailed timetable will be confirmed after consultation, so that the level of commitment, spacing of sessions and overall timeframe are realistic for participating institutions.
Next steps
If you are interested in this proposal, please use this consultation form to register for a drop-in session, submit your views online, or both. The form is intended to gather views before the pilot model is finalised, rather than to confirm participation in a settled programme.
We will review the feedback we receive and use it to confirm the final scope, format and timetable. If you have questions in the meantime, you can contact the team at ai@jisc.ac.uk.
Find out more by visiting our Artificial Intelligence page to explore publications and resources, learn more about our communities and sign up for our AI Literacy training.
For regular updates from the team sign up to our mailing list.
Get in touch with the team directly at AI@jisc.ac.uk
