Googles invisible AI watermark to identify generative text and video.
World’s go to search engine and a global leader in innovation, Google, has made quite some AI-related announcements at the Google I/O developer conference this week which include stronger security measures in its AI models to clamp down on the spread of misinformation via deepfakes and problematic outputs.
Google’s Advanced AI Watermarking
In order to enable the insertion of invisible watermarks on AI-generated video and text, the company expanded its SynthID line of watermarking technologies.
This way the documents can be traced back to their original sources and the SynthID already applies watermarks to AI-generated images and audio.
James Manyika, senior vice president at Google, at Google I/O said that “We are … developing new tools to prevent the misuse of our models”.
As AI is becoming the cynosure of developments, there is grown importance for watermarking AI-generated content in order to create various types of content.
There have been multiple cases across the globe of Deepfake video and audio have already been used to spread misinformation and for business email compromise.
Two new AI models were announced at the I/O by Google:
- Veo: AI model that generates realistic videos
- Imagen 3: AI model which generates life-like images.
Safeguarding Against Misinformation and Ensuring Accountability
Manyika said that the new watermarking techniques will be implemented in both models to easily identify fakes and prevent the spread of misinformation.
For starters, all the videos that are generated by Veo on VideoFX will be watermarked by SynthID. He said that “We’re doing a lot of research in this area, including the potential for harm and misuse”.
A watermark by SynthID, could mean a block of text or an invisible statistical pattern. This is then used in a scoring system in order to identify the uniqueness of that watermark pattern to see whether the text was AI-generated or came from another source.
Google is open-sourcing SynthID text watermarking to other vendors.
In a blog post, Google wrote that “SynthID for text watermarking works best when a language model generates longer responses, and in diverse ways — like when it’s prompted to generate an essay, a theatre script or variations on an email”.
The company further noted how various AI-assisted red-teaming techniques are employed to protect AI models.
In order to improve and expand their red-team capabilities, these AI agents are trained to compete with each other to improve and expand their red-team capabilities.
He added the “We test our own models and try to break them by identifying weaknesses. Building AI responsibility means both addressing the risks and maximizing the benefits of people and society.”
