Google DeepMind Launches SynthID: New AI Watermarking Tool for Text

Google DeepMind has open-sourced a new technology called SynthID, designed to watermark AI-generated text. Announced on Wednesday, this artificial intelligence (AI) watermarking tool can be utilized across various formats, including text, images, videos, and audio. Currently, the focus is on providing text watermarking capabilities for businesses and developers. The company aims to promote broader use of this tool to help identify AI-generated content easily. Users can access SynthID through Google’s updated Responsible Generative AI Toolkit.

In a post on X (formerly Twitter), Google DeepMind shared that the text watermarking feature of SynthID is now freely available to developers and businesses. Along with the Responsible GenAI Toolkit, it can also be downloaded from Google’s Hugging Face listing.

As AI-generated text increasingly populates the Internet, concerns about its implications are rising. A recent study by Amazon Web Services AI lab revealed that approximately 57.1 percent of all sentences online that have been translated into two or more languages might be produced using AI tools.

While the spread of AI-generated text may seem like harmless spamming, there are significant risks involved. Bad actors could exploit AI technology to generate large quantities of misinformation or misleading content. Given that much of social discourse happens online, such actions could affect real-world events, such as elections, and serve as propaganda against public figures.

Identifying AI-generated text has proven to be particularly challenging. Traditional watermarking methods are ineffective for words, and even if they could be applied, malicious users might simply rephrase the text to evade detection.

SynthID takes a novel approach to watermark AI-generated text. The tool leverages machine learning to predict the words that could follow specific words in a sentence. For example, in the sentence “John was feeling extremely tired after working the entire day,” only a limited selection of words can follow “extremely.” By analyzing the content generation styles of various AI models, SynthID predicts the next word after “extremely” and replaces it with a synonym from its database. This modified word is then embedded throughout the text. When SynthID checks for AI-generated content, it counts these altered words to assess authenticity.

For images and videos, SynthID embeds a watermark directly into the pixels of the frames, making it invisible yet detectable with the tool. In the case of audio, the sound waves are first transformed into a spectrogram, where the watermark is integrated into the visual representation. However, these advanced capabilities are currently exclusive to Google.

OpenAI's Orion Set to Revolutionize AI with 100x Performance Boost Over GPT-4

Global Scam Alert: Fake Trading Apps Exploit Users Through ‘Pig Butchering’ Fraud

Google's Strategic Shift: New Leadership Aims to Revolutionize AI Technology