LeaderTech

Out tomorrow! Link here: https://lnkd.in/ecnUf5FM

Students who received the best AI-generated notes scored the lowest on comprehension tests. Students who had to do more work — evaluating real-time summaries — scored the highest. They preferred the automated notes anyway. Xinyue Chen and colleagues (University of Michigan, 2025) tested three levels of AI assistance in note-taking. The difference in post-test scores between automated and intermediate conditions: statistically significant (p = 0.002). The mechanism: note-taking has two functions — encoding (active processing in real time) and storage (the record for later review). Automated AI handles the encoding for you. The notes are complete. The encoding is not yours. The tool that writes your notes for you may also be taking the understanding with them. My full analysis — what it supports, what it can't tell us, and what it means for teachers, researchers, and institutional leaders — in Wedensday's AI Research Insights. Subscribe here (for free): https://lnkd.in/e-pHpFzh #AIineducation #notetaking #highereducation

Since August 2nd, texts produced by Claude carry a mandatory AI watermark. Yet this signal does not say whether it was the student or the model who wrote the text. It only indicates that an AI probably participated in the process, at some point. Teachers hoping to finally have a reliable answer to the question of academic dishonesty will find themselves with a new tool — and the same unresolved questions. Institutions will have to decide whether to use this signal as evidence in their assessment procedures. A statistical signal only becomes proof if we collectively decide to treat it as such. That is the choice the 2026 academic year is going to force. Tomorrow in our LeaderTech newsletter: what the AI Act actually requires, how watermarking works at Anthropic, the documented limits of classical detectors — a complete FAQ with 6 references to prepare your conversations with colleagues and students. Link: https://lnkd.in/eN7B9Me9 (Original text in French - with automatic translation in English) #AI #HigherEducation #Governance

In a controlled study, automated AI note-taking produced lower comprehension than intermediate AI assistance. The difference was statistically significant. Participants rated the automated condition higher quality and said they'd use it again. The tool that feels more helpful may be the one doing the cognitive work that was supposed to be yours. Full study breakdown in this week's AIRI: https://lnkd.in/dvBMrcNj

It is practical to have an AI take notes during a conference or meeting. Unfortunately, this does not improve either the understanding or retention of the information presented. https://lnkd.in/eaMRG462

85 studies measured AI in experiential learning. In 48% of them, AI improved outcomes. In the other 52%, it didn't and AI was present in both halves. Bedoya, Chiappe, and Durand (2026) ran a PRISMA systematic review across those studies. The split tracked one variable: whether AI was designed to prompt students to justify a decision, or to speed up task completion. Same technology. Opposite results. The governance weight lands after that finding. When AI can generate a plausible final artefact — a finished case analysis, a defensible-looking placement report — grading that artefact alone no longer confirms the student did the reasoning behind it. Most institutions in the review hadn't built an assessment model that survives this. Publishing an AI use policy doesn't close the gap. A safeguard that lives in a document and never touches a rubric doesn't address the reasoning problem. Moving from giving students AI to designing curriculum around it is a governance decision about what counts as evidence that learning happened. My Substack post discusses what this means in practice for 1. programme leaders deciding what "pass" means on a placement report, 2. for curriculum committees drafting AI use policies, and 3. for assessment designers working before the artefact problem compounds. Read it here and subscribe (for free): https://lnkd.in/eki9HKeK Full study (open access): Bedoya, N. I., Chiappe, A., & Durand, J. (2026). European Journal of Education → https://lnkd.in/e9vRmCvh #AIGovernance #HigherEducation #AssessmentDesign #ExperientialLearning #EdPolicy

Teachers need training in AI. That’s what emerges from the AI policies of most U.S. states. Still, we need to know which kind of training… for what purpose… and, more generally: “why”? https://lnkd.in/ehKpXVe3

Quand les étudiants les plus satisfaits de l'IA sont ceux qui réfléchissent le moins ChatGPT a généré des lettres de motivation académiques pour 121 étudiants universitaires chinois à partir de leur CV (British Journal of Educational Technology, méthode mixte). Trois comportements ont émergé : 11 % ont tout accepté sans modification, 23 % ont corrigé la forme, 66 % ont repris le contenu en profondeur. Le résultat contre-intuitif : les étudiants les plus satisfaits du texte généré (médiane 7/7) sont ceux qui n'ont rien changé. Ceux qui ont exercé le plus d'esprit critique ont donné la note la plus basse (médiane 5/7, p < 0.001). ChatGPT avait systématiquement exagéré leurs expériences — « deux conférences » devenues «de nombreuses conférences » dans le texte généré. Question pour une direction académique: quand l'outil qui écrit au nom de l'étudiant surestime systématiquement son parcours, qui en porte la responsabilité institutionnelle? Jiang, Y., Wu, Q., Yang, Y., Jian, C. & Zhao, J. (2026). BJET, 57, 965–983. lien DOI https://lnkd.in/e3BqBgnh Ce type de lecture croisée entre recherche, enseignement et gouvernance nourrit chaque édition de mon Substack — lien: https://lnkd.in/eNm2F9Ec #GouvernanceIA #LeadershipÉducatif #IntégritéAcadémique #IAenÉducation #EnseignementSupérieur

Every major AI guidance document published by a state or intergovernmental body in the past three years includes some version of the same commitment: equitable access to artificial intelligence across education systems. None of them price it. Not the infrastructure. Not the digital literacy training. Not the ongoing cost of maintaining the systems through which access is supposed to flow. Institutions expected to operationalize this commitment receive a mandate, not a budget. What looks like access policy on paper becomes an unfunded directive in practice. The new issue of AI Research Insights examines the gap between what state AI guidance promises and what it actually funds — and what institutional leaders need to account for before adopting a framework built on resources the policy does not provide. Link here to subscribe (for free) and receive it : https://lnkd.in/dXVzzaB5