With government demand for AI-generated content being flagged, Google DeepMind has released a potential solution: a watermarking tool.
Called SynthIDThe system uses deep learning to embed digital stamps into the pixels of images. According to DeepMind, the watermarks are imperceptible to the human eye, but recognizable with an algorithm.
The UK-based research lab said the watermark doesn’t affect image quality. It remains recognizable even after modifications like adding filters, changing colors, and cropping.
The concept is similar to watermarks applied to film and stock photos. But in this case it is not about protecting the rights holders from forgery, but the viewer from deception.
“While SynthID is not a panacea for misinformation, but a promising technical solution to some of today’s pressing AI security problems.” Demis Hassabis, co-founder and CEO of Google DeepMind, said on X.
SynthID appears amid growing concerns about Generative AI. Governments around the world have sounded the alarm that manipulated images are undermining elections, distorting reality and misleading the public.
The risks were revealed in a recent presidential talk by Ron DeSantis campaign video. At first glance, the footage appeared to show actual photos of Donald Trump hugging and kissing Anthony Fauci, who led the White House response to the COVID-19 pandemic. But forensic scientists assume that the images are deepfakes.
China has already imposed strict rules on this technology. In January, the country made it mandatory Identify content created with artificial intelligence. Similar rules are also implemented in the EU.
“We need to label everything that is AI-generated by watermarking it,” Commissioner Thierry Breton said in June.
The US is also urging companies to identify content created with AI. Following a meeting at the White House last month, Google and six other tech giants met pledged to develop watermarking systems.
SynthID is a step towards fulfilling the commitments – but the system remains experimental. The tool will initially only be available to some paying customers of Google’s cloud computing business. It also only works on content created with the Imagen art generator.
That may change in the future. According to Google, SynthID be extended to other AI models. The company said it also plans to integrate the tool into other products.
Governments and regulators will closely monitor progress. With elections approaching in the US, India and the UK, the AI threat is looming ever closer.