A new academic paper argues that catching AI-written essays is the wrong fight. The real fix, it says, is redesigning how students get graded.
The paper proposes what it calls a "Dynamic Evidence Collection Ecosystem," a framework that replaces single-submission grading with continuous, multi-source evidence of a student's work. That means iterative drafts, design logs, records of separate activity rounds, self-reflection notes, and peer collaboration, all gathered over the course of an assignment rather than judged from one final file. An AI-enabled layer sits on top to handle learning analytics and formative feedback, with an emphasis on transparency rather than flagging suspected cheating. The authors also lay out an implementation scenario meant to help colleges actually adopt the framework, not just theorize about it.
Most schools have responded to generative AI with detection software, which is unreliable and puts the burden on catching students rather than making assignments harder to fake convincingly. This paper's bet is different: if only the finished essay gets graded, tools built to produce finished essays will keep winning. Grading the paper trail instead, drafts, logs, reflections, changes what a student actually has to produce to pass.
It is also a lot more for instructors to review than a single PDF, and this is a design framework, not a tested classroom pilot. Whether it survives contact with a real grading queue, and a real overloaded professor, is still an open question.