Future AI chatbots may secretly dob on their users
You might not be able to pass generative AI writing off as your own for much longer: data scientists have developed a way for the technology to dob on its users. or signup to continue reading.
Technological Solution for Watermarking Text
Researchers from Google DeepMind released their findings on Thursday, introducing a technical solution that could watermark text generated by large-language models like Google Gemini and ChatGPT. This breakthrough in artificial intelligence has significant implications for various fields, including education.
The research project, known as SynthID-Text, was detailed in Springer's Nature journal. Google DeepMind researchers explored the concept of biasing word choice in AI-generated text to create a signature recognizable by AI detection programs.
Distinct Approaches
The study delved into two main approaches: a "distortionary" method that impacted the writing quality of AI tools and a "non-distortionary" approach with minimal impact on the text quality. Testing conducted on over 20 million Gemini chatbot interactions revealed that both methods made the AI-generated text easier to distinguish as machine-created.
The paper stated, "Our work provides proof of the real-world viability of generative text watermarks," emphasizing the importance of accountability and transparency in deploying large-language models.
Challenges and Benefits
Despite the efficacy of text-based watermarks, the study acknowledged their vulnerability to weakening if users edited or paraphrased the content. However, this watermarking technology presents a significant step forward in addressing the challenges posed by generative AI.
Computer science lecturer Erica Mealy from the University of the Sunshine Coast highlighted the potential of AI watermarking in educational settings for detecting AI-generated contributions in studies and research papers. The technology could help mitigate false positives and negatives associated with generative AI use.
Implications and Considerations
While the research demonstrates the promise of SynthID-Text in enhancing accountability, Google's plans to incorporate this technology into its AI tools remain undisclosed. Nevertheless, the development of text-based watermarks marks a critical advancement in the evolving landscape of artificial intelligence.
