Previously yr, the massive popularity of generative AI models has also brought with it the proliferation of AI-generated deepfakes, nonconsensual porn, and copyright infringements. Watermarking—a way where you hide a signal in a bit of text or a picture to discover it as AI-generated—has change into probably the most popular ideas proposed to curb such harms. 

In July, the White House announced it had secured voluntary commitments from leading AI firms reminiscent of OpenAI, Google, and Meta to develop watermarking tools in an effort to combat misinformation and misuse of AI-generated content. 

At Google’s annual conference I/O in May, CEO Sundar Pichai said the corporate is constructing its models to incorporate watermarking and other techniques from the beginning. Google DeepMind is now the primary Big Tech company to publicly launch such a tool.

Traditionally images have been watermarked by adding a visual overlay onto them, or adding information into their metadata. But this method is “brittle” and the watermark might be lost when images are cropped, resized, or edited, says Pushmeet Kohli, vice chairman of research at Google DeepMind.

SynthID is created using two neural networks. One takes the unique image and produces one other image that appears almost similar to it, but with some pixels subtly modified. This creates an embedded pattern that’s invisible to the human eye. The second neural network can spot the pattern and can tell users whether it detects a watermark, suspects the image has a watermark, or finds that it doesn’t have a watermark. Kohli said SynthID is designed in a way which means the watermark can still be detected even when the image is screenshotted or edited—for instance, by rotating or resizing it. 

Google DeepMind is just not the just one working on these styles of watermarking methods,  says Ben Zhao, a professor on the University of Chicago, who has worked on systems to forestall artists’ images from being scraped by AI systems. Similar techniques already exist and are utilized in the open-source AI image generator Stable Diffusion. Meta has also conducted research on watermarks, even though it has yet to launch any public watermarking tools. 

Kohli claims Google DeepMind’s watermark is more proof against tampering than previous attempts to create watermarks for images, although still not perfectly immune.  

But Zhao is skeptical. “There are few or no watermarks which have proven robust over time,” he says. Early work on watermarks for text has found that they’re easily broken, often inside just a few months.