EditLens: Quantifying the Extent of AI Editing in Text
Summary
EditLens analyzes the extent of AI editing in text, introducing lightweight similarity metrics to quantify edits made by AI relative to original human text. It trains a regression model to predict the degree of AI editing and reports strong performance on classification tasks, with a case study examining Grammarly. The work has implications for authorship, education, and policy, and teams plan to publicly release their models and data.