Researchers build AI detector designed to spot automated ideation

Researchers from universities and tech institutions have developed a tool called IdeaLens to identify whether the underlying concepts in a piece of writing originated from artificial intelligence. Traditional detection software focuses on statistical quirks in word choice, which means it often fails when a human rewrites machine text or uses an AI outline to write an article. IdeaLens addresses this by converting a document into a simplified outline, removing the specific phrasing to analyze only the progression of ideas.

According to a research paper published on arXiv, this strategy detects machine assistance that conventional software misses. When human authors wrote stories based on AI generated outlines, older detection tools flagged only eight percent of the texts, while IdeaLens flagged sixty eight percent. Similarly, when an AI model wrote prose based on an outline designed by a human, IdeaLens recognized the human conceptual origin and flagged the draft as AI generated only seven percent of the time.

This distinction is important for academic and professional settings where rules often focus on who conceived an idea rather than who drafted the words. However, real world creative workflows are rarely strictly human or strictly machine. It remains uncertain how effectively such detectors will perform on collaborative writing, such as documents where a human writer iteratively refines an AI outline with their own thoughts.