Open Source Google DeepMind SynthID AI Watermarking Technology for Enterprises and Developers

Google DeepMind open-sourced a new technology for watermarking AI-generated text on Wednesday. Called SynthID, the artificial intelligence (AI) watermarking tool can be used in different modalities, including text, images, videos and audio. However, currently, it offers the text watermark tool only to businesses and developers. The company aims for broader adoption of the tool so that AI-generated content can be easily detected. Individuals and businesses can access the tool through the Mountain View-based tech giant’s updated Responsible Generative AI Toolkit.

Google DeepMind’s Open Source AI Text Watermarking Technology

In a post on X (formerly known as Twitter), the official Google handle DeepMind announced the free availability of SynthID’s text watermarking feature to developers and enterprises. In addition to the Responsible GenAI Toolkit, it can also be downloaded from Google’s Hugging Face list.

AI-generated text has already started to flood the Internet. Amazon Web Services’ AI lab published a study earlier this year that claimed that up to 57.1% of all online sentences that have been translated into two or more languages ​​can be generated using AI tools.

While AI chatbots filling the internet with AI-generated gibberish may seem like a case of harmless spam, there is a darker side to it. In the hands of bad actors, AI tools can be used to mass generate misinformation or misleading content. With a significant portion of social discourse occurring online, such actions can impact real-life events, such as elections, and be used to create propaganda against public figures.

Of all modalities, evaluating AI-generated text has proven to be the most difficult task so far. This is mainly because it’s not possible to watermark words, and even if it were, bad actors could always rework the content using a second output loop.

However, Google DeepMind’s SynthID uses a new way to mark up AI-generated text. The tool uses machine learning to predict the words that might appear after a specific word in a sentence. For example, consider the sentence “John was feeling extremely tired after working all day.” Here, only a limited number of words can appear after the word “extremely”.

Based on analysis of the content generation styles of various AI models, SynthID can predict the word that will appear after “extremely” and replace it with another synonym that exists in its database. The watermark tool will embed these words throughout the content. Later, when the tool scans AI-generated content, it looks for the number of these words to determine their authenticity.

Notably, for images and videos, SynthID adds a watermark directly to the pixels of the frames so they remain invisible but can still be detected in the tool. For audio, the audio waves are first converted in a spectrograph and the watermark is added to the visual data. These features are currently not available to anyone outside of Google.