Three Scenarios for AI in Education: From Responsible Assistance to Co-Creation

Abstract

This article proposes a pragmatic and proportionate pathway for integrating generative artificial intelligence into higher education through three scenarios graduated by autonomy, agency, and risk (responsible support, guided collaboration, and co-creation with a strengthened declaration). These scenarios convert broad principles into verifiable and traceable teaching decisions through-out the teaching cycle (planning, material creation, support, and assessment). The common thread is the use of artificial intelligence as a supplement to academic judgment, never as a substitute, with transparency (including disclosure and marking of synthetic content), external verification of facts and citations, and equity and inclusion by design. This is consistent with UNESCO’s guid-ance (a human-centred vision, immediate actions, and capacity building), the AI Act (Article 50 on transparency and marking obligations), the Safe AI in Education Manifesto (human supervision, privacy, accuracy, explainability, transparency), and the SAFE framework (Safety, Accountability, Fairness, and Efficacy) as an operational bridge between policy and the classroom. Scenario 1 prioritises low risk and high transparency; Scenario 2 focuses on traceable iteration with sig-nificant human post-editing; and Scenario 3 demands robust evidence and auditing (prompts, versions, verification, bias/language checks, human/peer review), with strengthened controls due to its higher impact. This gradient aligns with sector guidance, which promotes authenticity, agency, and ownership of the process and advises against relying on detectors, thereby reinforcing designs that verify agency and traceability. Two instruments facilitate adoption and consistent evaluation: firstly, a cross-cutting rubric (veracity and currency, traceability, correction of hallu-cinations, equity and language, and quality of interaction); and secondly, checklists for each type of task. The result is an operational map for marking, verifying, and documenting in proportion to risk, which enables artificial intelligence to be leveraged as a pedagogical opportunity without compromising on rigour, fairness, and responsibility.