A good grade does not show who did the thinking
A preprint examines 150 student-AI conversations. Similar assessed performance can hide different visible roles in setting direction, bringing ideas together and judging answers. Chat records alone cannot reveal learning or prove misconduct; this is not a K-12 trial.
What the source reports
Jorge Fábrega studies 150 chats between students and AI. The preprint looks beyond the work students turned in. Some work received similar scores, but students did not all play the same visible part. They differed in how they set goals, joined ideas and checked answers. A good final score cannot show every step of that work. But a chat record has limits too. It cannot tell us everything a student learned. It cannot prove cheating. The abstract does not show a cause and effect result for K-12 schools. A teacher could ask for a short account of key choices. An unassisted task might show what a student can do alone. Those are ideas from the digest, not proven effects of this study.
Original source
- Title
- Same Performance, Different Process: Epistemic Ownership in AI-Mediated Education
- Author
- Jorge Fábrega
- Publication
- arXiv
- Date
- Friday, October 2, 2026