Sets out the public-interest considerations relevant to generative AI in education, including legal context, accountable implementation and the treatment of material risk.
Consideration of generative AI in education should retain the date and status of International guidance released on 7 September 2023. Later developments should not be read into the position available at publication. Analysis of the issue should state the unit of analysis, reference period, coverage, exclusions and treatment of missing information. Material differences in population, setting or method should remain explicit in any comparison. Review of the implementation question should give particular attention to adverse cases, unequal effects and errors that learners may be unable to identify or remedy after the event.
At the publication date, International guidance released on 7 September 2023 provides the relevant international context for the policy matter. Any consequential application still requires evidence from the affected jurisdiction or institution. Data used for the issue should be interpreted against stable definitions and an identifiable population. A revision or break in series should not be reported as a change in performance. Public information on the issue should state the applicable scope and limitations in terms that affected users can understand, including the basis for any later correction.
Scope of this analysis
At the publication date, International guidance released on 7 September 2023 provides the relevant international context for generative AI in education. Analysis of the implementation question should state the unit of analysis, reference period, coverage, exclusions and treatment of missing information. Review of the policy matter should give particular attention to adverse cases, unequal effects and errors that learners may be unable to identify or remedy after the event.
For the issue, the implementation record should distinguish binding duties, policy expectations and institutional choices, including any transition or jurisdictional limitation. Responsibility for the relevant measure should be identifiable at each consequential decision point. Delegation should identify both the operating role and the body retaining oversight of learner impact. Any indicator used in relation to the issue should distinguish description from causal explanation. Interpretation should retain uncertainty, distributional differences and limits on generalisation.
A focused examination of the implementation question requires a clear analytical discipline. The principal risks associated with the issue should be assessed as connected conditions. A failed safeguard may conceal another weakness or prevent timely correction. Frequency is relevant, but a rare event may still be material where the effect is serious or irreversible. An imprecise scope or measure may produce a credible-looking record that does not answer the relevant decision question.
Implementation of the relevant measure should be organised around a decision that can be tested. The record for the relevant measure should identify the responsible function, decision authority and escalation route. Gaps between public oversight and provider control should not remain implicit. Apparent movement caused by revision should not be attributed to educational performance.
The evidential record for the relevant measure should permit a reviewer to trace the matter from decision to outcome. Any indicator used in relation to the relevant measure should distinguish description from causal explanation. Material variation and uncertainty should be reported together with any restriction on wider application. The sample should be extended when records conflict, a material group is missing or earlier corrective action may not have been sustained.
- Prohibit uses for which evidence or authority is insufficient within a defined period and review the result.
- Review incidents and supplier changes, and retain the basis, responsible function and affected scope.
- Control personal and confidential information and retain evidence sufficient for independent review.
- Retain accountable human decision-makers before it informs a consequential decision.
- Classify uses by effect on learners, including material exceptions and unequal effects.
Operational significance
Analysis of generative AI in education should state the unit of analysis, reference period, coverage, exclusions and treatment of missing information. Reliability depends on preserving the material distinctions between the matters compared. The relationship between the risks is material: one failed safeguard may remove the evidence needed to activate another. Analysis of the policy matter should state the unit of analysis, reference period, coverage, exclusions and treatment of missing information. Comparative findings should not conceal differences capable of changing their meaning.
The review method for the issue should be reproducible. Data used for the policy matter should be interpreted against stable definitions and an identifiable population. Changes in method, definition or series should remain separate from changes in the underlying result. Reassess materiality when new evidence changes the likely scope or consequence. Risk assessment for the issue should consider severity, reach, duration, recurrence and detectability, with escalation where learner impact may be material.
A conclusion should not imply uniform application where the governing law differs between jurisdictions. Public communication should not present an aspiration, recommendation or proposed measure as an existing legal duty.
The principal risks associated with the relevant measure should be assessed as connected conditions. A decision concerning the issue should recognise that a technical capability is not evidence that a use is educationally justified. A formal commitment concerning the implementation question does not establish effective operation. Review should test how the measure is applied, how exceptions are handled and what remedy is available.
Evidence concerning the issue should be current, attributable and representative of the affected scope. Material gaps or contradictions should remain visible in the conclusion. Where reliance has occurred, correction may require review of affected decisions as well as amendment of published information.
Any indicator used in relation to the policy matter should distinguish description from causal explanation. A reported result should state how outcomes are distributed and where transfer beyond the observed setting is not supported. Records concerning the issue should remain traceable from source evidence to decision and follow-up. Superseded conclusions should be retained where they informed a material outcome. Records concerning the policy matter should remain traceable from source evidence to decision and follow-up.
Neither one indicator nor one control can establish the complete position on the issue. Review of the issue should give particular attention to adverse cases, unequal effects and errors that learners may be unable to identify or remedy after the event.