Google made a series of artificial intelligence (AI) announcements during its I/O 2024 keynote late Tuesday. This includes new AI models, improvements to existing core models, and the integration of AI features into Google products.
The tech giant has also focused on AI security, expanding the use of a native watermarking technology for AI-generated content called SynthID. This new toolkit now embeds watermarks in text created by Gemini apps and web clients, as well as videos created by Veo.
SynthID was first launched by Google DeepMind in August 2023 as a beta project with the goal of correctly tagging AI-generated content. The need for such a solution has been felt due to the increasing number of cases where these man-made media are being passed off as authentic. These were used to spread false information and cybercrimes such as phishing.
The tech giant first used the technology in November 2023 when it was used to watermark AI-generated audio data created using the Lyria model. This toolkit adds a waveform watermark to your audio, making it subtle but recognizable.
Now Google is expanding the use of SynthID to include text and video generation. Text generated with Gemini apps and websites is now watermarked. For this reason, the toolbox itself is aimed at the production process. All text-based AI models use cues (words, syllables, phrases, etc.) for training. Part of the learning process includes understanding the process of using these cues and the order in which they should be followed to produce the most consistent response.
SynthID introduces “additional information for token distribution at the time of generation by setting the probability of a token being generated.” In this way, numbers are assigned to specific words in the generated block of text. When determining whether AI was used to generate the text, we compare the score to an adjusted probability score to determine whether the source is likely to be an AI model.
DeepMind highlighted in the post that this technique is useful when AI generates long creative texts, as it makes it easier for probabilistic models to identify how the text was created. However, if the actual responses are short, the detection may not be as accurate.
The company is also expanding SynthID to AI-generated videos in the recently launched Veo. According to Google, the technology embeds a watermark directly into the pixels of each video frame, making it imperceptible to the human eye but visible when using recognition systems.
Google plans to open-source SynthID text watermarking through its Responsible Generator AI toolkit in the coming months. A detailed research report describing text watermarking technology will also be published.
Bhupendra Singh Chundawat is a seasoned technology journalist with over 22 years of experience in the media industry. He specializes in covering the global technology landscape, with a deep focus on manufacturing trends and the geopolitical impact on tech companies. Currently serving as the Editor at Udaipur Kiran, his insights are shaped by decades of hands-on reporting and editorial leadership in the fast-evolving world of technology.
