Syntes: Bauer, Greiff, Graesser, Scheiter & Sailer (2025) — AI can invert learning when learners over-rely on it

Källa: Bauer, E., Greiff, S., Graesser, A. C., Scheiter, K. & Sailer, M. (2025). Looking Beyond the Hype: Understanding the Effects of AI on Learning. Educational Psychology Review 37, 45. https://doi.org/10.1007/s10648-025-10020-8

Verified content

• [full text] Reflection on the field (accepted 5 April 2025, published 24 April 2025). Focus is cognitive learning processes and outcomes. The authors introduce the ISAR model: inversion, substitution, augmentation and redefinition, comparing AI with no-AI conditions (building on Puentedura's SAMR model).
• [full text] "The ISAR model introduces inversion effects, referring to reduced cognitive learning when learners over-rely on AI as demonstrated by initial research on generative AI." Inversion "occur[s] when AI, intended to support deep learning, instead leads to reduced cognitive processing and learning outcomes". They cite studies of students using ChatGPT for constructive tasks (information search and writing) as linked to shallower processing and diminished outcomes (Fan et al., 2024; Stadler et al., 2024).
• [full text] Substitution: AI matches a non-AI alternative without changing processing depth (they point to intelligent-tutoring metas finding no significant difference from human tutoring). Augmentation: extra cognitive support versus a weaker or unsupported alternative. Redefinition: the task itself is changed so that learning becomes more constructive or interactive.
• [full text] They also note a workplace parallel: replacing human actions with AI can erode skills maintained through practice, and over-reliance may weaken judgement when AI is later unavailable.

Limits

• [assessment] A framework paper, not a new trial or meta-analysis of generative AI in school practice. The inversion examples cite other studies rather than original data. Linked to C2 because that claim is about answer-giving AI during practice harming later unaided performance — the same direction as inversion — not because this paper tests C2.

Why it is in the library

A clear, sourced framework for how AI can help or hurt cognitive learning. The inversion warning (over-reliance, shallower processing, worse outcomes) is the adjacent idea behind C2, and the workplace parallel touches C18 without being a new deskilling trial.

Added 2026-11-08 by the project (weekly source sweep, kind: adjacent (human-AI interaction); read: full text). [abstract] = from the published abstract; [full text] = read in the full text; [assessment] = the project's own assessment.