Researchers from the University of Maryland, Google DeepMind, and Simon Fraser University have developed a new technical tool called "IdeaLens," aimed at detecting AI-generated ideas within texts, even when humans rewrite them in their own style. The tool works by analyzing core ideas and arguments rather than focusing solely on words and linguistic structures, with the goal of distinguishing between texts based on human ideas and those relying on AI-generated ideas, particularly in scientific research and educational documents.
Researchers relied on training their models using approximately one million pre-classified documents, converting texts into summaries focusing on key ideas and arguments while reducing the impact of writing style. This approach helps the system analyze the text's content rather than relying on linguistic patterns that may change upon rewriting, distinguishing it from traditional detection tools that look for linguistic patterns and repeated phrases.
Tests showed that the tool correctly identified 91 percent of texts fully generated by AI, while only 0.6 percent of human texts were misclassified. This mechanism differs from traditional detection tools that look for linguistic patterns and repeated phrases that may indicate AI use, as "IdeaLens" seeks to identify the source of the ideas themselves, not the entity that formulated the final words.
Comparisons showed that the tool's performance varies depending on the method of human-AI collaboration; when a human outlines the ideas and AI writes the text, the traditional "Pangram" tool detected AI involvement in 92 percent of cases, compared to 7 percent for "IdeaLens." Conversely, when a human writes a text based on an outline prepared by AI, "IdeaLens" detected AI influence in 68 percent of cases, compared to 8 percent for "Pangram."
Despite these results, the tool still requires further testing, as it faced difficulties dealing with some of the latest AI models, and its results do not definitively prove the source of every idea, necessitating caution when using it in educational institutions and scientific journals.
Source: Syrian Arab News Agency (SANA)
Comments (0)