AI in education should increase human agency
The important question is not whether a classroom “uses AI.” It is whether the use helps learners think, teachers teach, and institutions act more responsibly.
Begin with a learning purpose
An AI feature should serve a specific educational goal: practicing retrieval, receiving formative feedback, exploring alternative explanations, improving accessibility, or reducing an administrative burden. “Innovation” is not a learning outcome. If the goal is unclear, success cannot be evaluated.
Keep the learner intellectually active
A system that produces the final answer may save time and remove the practice a learner needs. Design interactions that ask for a prediction, explanation, draft, or reflection before assistance appears. The tool can then challenge reasoning, generate a contrasting case, or identify a gap.
Preserve the teacher’s role
Teachers understand local context, relationships, curriculum, and the consequences of a decision. AI can support preparation and feedback, but it should not quietly determine a grade, diagnosis, disciplinary response, or learning pathway without accountable human judgment.
Protect privacy and age-appropriate use
Education data may reveal identity, ability, behavior, disability, or family circumstances. Institutions should know what is collected, why it is needed, where it goes, and how it is deleted. UNESCO’s guidance calls for a human-centered and age-appropriate approach, including data protection and ethical validation.
Test equity in the real setting
A tool may work differently across languages, accents, devices, connectivity levels, and learning needs. Pilot with the actual community, not only ideal examples. Provide an alternative path for people who cannot or should not use the system.
Require evidence proportional to impact
A low-stakes brainstorming aid can be tested informally. A tool influencing assessment, placement, or student well-being needs stronger evidence, documentation, and routes for challenge. Measure whether learning improves—not only whether users generate more output.
- The learning goal is stated.
- Students know when AI is involved.
- Data collection is minimized.
- Teachers can inspect and override results.
- Non-AI alternatives remain available.
- Errors and complaints have a clear route.
Teach judgment, not only prompting
Students need to recognize uncertainty, check citations, compare sources, disclose assistance, and decide when a tool should not be used. Prompt technique changes quickly; judgment lasts longer.
AI in education is most defensible when it expands a person’s capacity to understand and act. It is least defensible when convenience hides a transfer of agency that learners and teachers cannot see or contest.
Sources and further reading
Editorial note: This article was prepared with AI assistance and reviewed against the cited UNESCO guidance by LifeTechGlow.