AI study says 'scaffold, don't substitute' to break education's harm cycle
80% of student essays say AI reliance reduces thinking — researchers propose a fix.
A new research paper from Lucile Favero, Juan Antonio Pérez-Ortiz, Tanja Käser, and Nuria Oliver, accepted at the ACM AI Leadership Summit and posted on arXiv (2608.17451), argues that the biggest risk of AI in education is misalignment: systems that eliminate human effort erode the very capabilities education aims to build. The authors organize this into an integrative framework spanning four interrelated dimensions — cognition, agency, emotional well-being, and ethics — linked by a self-reinforcing cycle where cognitive offloading reduces effort, weakens agency, and compounds emotional and ethical harm.
Grounding this in student voices, they analyzed 49 International Baccalaureate argumentative essays about AI's impact. 80% of essays reported that relying on AI reduces thinking. Yet students consistently described wanting AI that supports rather than replaces learning: withholding immediate answers, prompting recall, and encouraging reflection through questions. The authors distill this into one principle: "scaffold, do not substitute." They argue this extends beyond education to any AI system that mediates human thinking, and outline a research agenda for building AI that strengthens enduring human capacity — an imperative for democratic societies.
- 80% of 49 student essays say AI reliance reduces thinking, based on an analysis of International Baccalaureate argumentative essays.
- The paper defines a self-reinforcing harm cycle across four dimensions: cognition, agency, emotional well-being, and ethics.
- Proposed design principle: "scaffold, do not substitute" — AI should prompt recall and reflection instead of handing out solutions.
Why It Matters
As AI enters classrooms at scale, this framework pushes developers to build systems that strengthen human capability, not erode it.