'Humanizer' tool can erase signs of AI-written text (www.nature.com)

A new tool called "Humanizer," developed by Jie Ding from the University of Minnesota, aims to help researchers revise AI-generated text to better match their own writing style, particularly for academic papers and grant proposals. Released on June 20, the tool focuses on "stripping AI tells" while maintaining the text's clarity and authenticity. Despite its potential utility, notably praised by some researchers for assisting with tasks like email writing, the tool has sparked significant debate. Critics argue that it risks encouraging scientists to use AI inappropriately, potentially obscuring the role of AI in their work. Max Spero, CEO of the AI-detection firm Pangram, noted that initial tests showed the Humanizer could still be detected in modified text, indicating that while the tool is not overly sophisticated, it raises ethical concerns about transparency in academic writing. In response to such concerns, Ding has updated the tool's description to emphasize its role as an editing aid and added an ethical note highlighting the obligation to disclose AI assistance. As AI tools become increasingly prevalent in academic environments, the discourse surrounding their use, implications for scientific integrity, and the role of disclosure remains a critical issue for the AI/ML community.