Nature Podcast: Detecting the Effects of Brain Scans and High-Altitude Tree Planting on Statistically Marked Text
Never miss an episode. The Nature Podcast is accessible on a variety of audio and video applications. An RSS feed for the Nature Podcast
is available too.
The researcher found out how birth-control pills respond to her brain scans and how high-altitude tree planting could offer refuge to a butterfly species.
Sumanth Dathathri and his colleagues at the DeepMind laboratory in London tested a new approach to marking text that uses a statistical signature. The word watermark comes from the era of paper and print, and describes a variation in paper thickness, not usually immediately obvious to the naked eye, that does not change the printed text. A watermark in digitally generated text or images should be similarly invisible to the user — but immediately evident to specialized software.
Two remote cities hidden high in the mountains of Central Asia are revealed, as is how a digital watermark could help identify artificial intelligence-generated text.
The Impact of Artificial Intelligence in Watermarking the Outputs of SynthID-Text: Highlights from Scott Aaronson and John Kirchenbauer
In a welcome move, DeepMind has made the model and underlying code for SynthID-Text free for anyone to use. The work is an important step forwards, but the technique itself is in its infancy. We need it to grow up fast.
There is an urgent need for technological improvements to combat the misuse of generative artificial intelligence, as well as to understand how malicious actors use it, what a trustworthy information environment looks like and how people interact with it. Researchers need to study these questions.
However, even if the technical hurdles can be overcome, watermarking will only be truly useful if it is acceptable to companies and users. Regulation is probably going to force companies to take action, but it’s not certain whether users will trust watermarking or similar technologies.
In contrast to the United States, the European Union has adopted a legislative approach, with the passage in March of the EU Artificial Intelligence Act, and the establishment of an AI Office to enforce it. China’s government has already introduced mandatory watermarking, and the state of California is looking to do the same.
The approach of watermarking the outputs is not new by the authors. Openai, a San Francisco, California-based company, is testing a version of it. But there is limited literature on how the technology works and its strengths and limitations. One of the most important contributions came in 2022, when Scott Aaronson, a computer scientist at the University of Texas at Austin, described, in a much-discussed talk, how watermarking can be achieved. Others have also made valuable contributions — among them John Kirchenbauer and his colleagues at the University of Maryland in College Park, who published a watermark-detection algorithm last year3.