An AI tutor may explain an answer confidently or assign a quality label to written work. That output should not become an unchallengeable judgment about a learner.
Define the pedagogical role
UNESCO’s guidance on generative AI in education and research advocates a human-centred approach. It is a reference for design and governance, not proof that a particular tutoring product improves learning.
An illustrative language-learning tool could provide formative suggestions while teachers retain responsibility for consequential assessment under the institution’s process.
Distinguish help with practice from grading, admissions or disciplinary decisions. The latter require different authority and safeguards.
Make feedback inspectable
Connect the response to the task, relevant source or marking criteria. Avoid unsupported claims about a student’s ability, motivation or personal characteristics.
Give students a comprehensible way to ask for clarification or human review. The process should not assume that an articulate challenge is the only signal of an error.
Use appropriately authorised data and follow institutional rules for learner information, particularly where children are involved.
Evaluate the learning interaction
Review factual errors, confusing explanations and the effort teachers spend correcting output. Engagement time alone does not establish educational benefit.
Where an outcome study is proposed, use a suitable design and report its limits. A successful demonstration is not evidence of sustained learning gains.
Ask teachers to review examples of both useful and misleading feedback before approving the classroom workflow. The comparison can reveal whether the interface gives learners enough context to distinguish a suggestion from a verified explanation. It should inform pedagogical design rather than become a ranking of students based on model output.
GAT’s EdTech coverage benefits from this distinction. The value of educational AI depends on how it supports a defensible learning process, including the ability to recognise and correct its own influence when the feedback is unsuitable.
Published by the Global Apex Tech Editorial Desk. Partner involvement, when applicable, is disclosed above the headline. For editorial questions or source material, contact editor@globalapextech.org.