EditLens: Quantifying the extent of AI editing in text (arxiv.org)

Researchers have introduced EditLens, a sophisticated tool designed to quantify the degree of AI editing present in text, distinguishing it from both human and fully AI-generated content. Unlike previous efforts that primarily focused on detecting entirely AI-produced text, EditLens employs lightweight similarity metrics validated by human annotators to assess how much an AI has altered original human-written material. Leveraging these metrics, the model achieves groundbreaking performance in classification tasks, with F1 scores of 94.7% for binary and 90.4% for ternary categorizations. This development is significant for the AI/ML community as it opens new avenues for understanding the interactions between human authorship and AI editing. The ability to not only detect AI-edited text but also quantify the extent of AI alterations has profound implications for authorship attribution, educational integrity, and policy formulation regarding AI-assisted writing tools. As a practical application, the researchers analyzed edits made by Grammarly, revealing insights into AI's impact on writing. They further demonstrate their commitment to the community by planning to publicly release their models and dataset, encouraging ongoing research in this evolving field.