01

Start with academic authority

An institution should decide which course material is approved, who may update it and what an AI tutor is allowed to do with it. The generated explanation should be visibly distinct from the source. If it points a learner in the wrong direction, the academic team needs a practical correction path.

This is consistent with a human-centred approach to educational AI. UNESCO’s guidance emphasizes data privacy, age-appropriate use and ethical validation of generative AI in education. The FLOOR9 pilot should turn those principles into observable checks rather than a slogan.

02

Test learning, not just engagement

A longer conversation is not automatically a better lesson. An evaluation should look at whether students can explain or apply the concept afterward, whether misconceptions are corrected and whether the experience supports learners who need a different explanation.

A useful pilot combines task-level evaluation, student feedback and lecturer review. It should compare the tutor’s output with approved sources and record the types of error that matter most in the subject.

03

Separate teaching from assessment

A tutor may guide practice with hints, examples and questions. During independent assessment, the permitted assistance can change. If the system does not make that boundary clear, neither the learning experience nor the result can be interpreted confidently.

Access controls, retention, usage cost and device support also matter. An institution needs to know who could reach a course, what information was processed, whether the service remained available and what each useful learning hour cost.

04

Decide expansion from evidence

The first pilot should produce a documented account of improvement, limitations and operational problems. If the tutor is accurate but difficult for students to use, that matters. If it is engaging but does not improve understanding, that matters too.

The purpose of evaluation is to decide what to change before scaling. An ambitious technology system earns academic trust through that discipline.