A professor at the University of Florida, Yuheng Bu, is developing an innovative invisible watermark technology designed to detect content generated by artificial intelligence (AI).
This technology aims to address the growing challenge faced by academics and employers: discerning whether a text was written by a human or by AI.
In a study conducted in the United Kingdom, it was revealed that 94% of AI-generated texts went undetected, highlighting the urgency of finding effective solutions.
The need to identify AI-generated text
Bu’s proposal implies that the entity applying the watermark also possesses the key necessary for its detection, raising questions about intellectual property and access to such keys. According to Bu, the quality of the text generated by AI will be maintained after applying the watermark, ensuring that it is robust against modifications such as paraphrasing and synonym replacement.
Working with watermarks could become a crucial tool to ensure trust and authenticity, facilitating verification in academic and digital environments, and helping to differentiate genuine content from misinformation.
This claims that its method offers an improvement over other developments, such as those from Google DeepMind, by applying watermarks only to a subset of text during generation.
However, a significant challenge lies in how end users obtain that watermark key. Currently, a teacher must contact the entity that has embedded the watermark to access the key, raising questions about the control and ethical implications that this approach generates.
In this regard, Bu emphasizes the need to establish an ecosystem that ensures the effective use of watermarks and the distribution of keys in the future.