A new tool reveals the source of ideas in texts generated by artificial intelligence.

New Tool Developed to Identify Source of Ideas in Texts

A research team from the University of Maryland, Google DeepMind, and Simon Fraser University has developed a new tool called IdeaLens. This innovative tool aims to determine whether the core ideas in scientific and research texts originate from a human or artificial intelligence. It achieves this by analyzing arguments and thought patterns rather than focusing on linguistic style.

How IdeaLens Works

The tool strips texts of their writing style and converts them into a diagram that highlights the main ideas. This process allows for the evaluation of the idea’s source, regardless of who authored the final draft. The system has been trained on approximately one million documents previously classified as either human-generated or AI-generated.

Performance and Accuracy

Tests have shown that IdeaLens successfully identifies 91% of works created by artificial intelligence. In contrast, the error rate for classifying human texts does not exceed 0.6%. The tool also outperformed traditional detection methods when analyzing texts that underwent minor AI modifications. It classified 95% of human documents as human, compared to an average of 44% achieved by the Pangram tool.

Distinguishing Human and AI Ideas

The results indicate that human ideas often feature concrete details, verifiable sources, and acknowledgment of limitations. In contrast, AI-generated ideas tend to be more general, make broad claims, and present figures that are difficult to verify.

Future Use and Limitations

Despite the tool’s accuracy, researchers emphasize that it is not yet ready for widespread use in scientific journals or educational institutions. The tool faces challenges with some modern AI models, necessitating further testing before official adoption.

To follow the news in Arabic