AI Misuse in Education Is a Measurement Problem: Toward a Learning Visibility Framework

A new research paper by Eduardo Davalos and Yike Zhang, titled 'AI Misuse in Education Is a Measurement Problem: Toward a Learning Visibility Framework,' challenges the current approach to AI in education. The authors argue that institutional responses focusing on AI detection tools and restrictive policies have proven unreliable and ethically problematic. Instead, they reframe the issue as a measurement problem: when AI enters the assessment loop, educators lose visibility into how learning outputs are produced, retaining only the final results.

Drawing from cognitive offloading, learning analytics, and multimodal timeline reconstruction research, the authors propose a three-principle framework. First, educators should clearly specify and model acceptable AI use. Second, they should recognize learning processes as assessable evidence alongside final outcomes. Third, they should establish transparent timelines of student activity. The framework emphasizes transparency and shared evidence rather than surveillance, aiming to preserve trust between students and educators while aligning AI use with educational values.

The paper, submitted and accepted to AIR-RES2026, represents a significant shift in thinking about educational integrity in the age of conversational AI systems like ChatGPT and Claude. By moving from an adversarial detection mindset to one focused on process visibility, the framework offers a principled pathway for integrating AI tools ethically. This approach acknowledges that AI is here to stay in education and focuses on making learning processes more transparent rather than trying to police AI use through unreliable detection methods.