Google releases AI watermark tool SynthID for public use

Google has made its SynthID tool, a technology that embeds invisible watermarks into AI-generated content, publicly available. This tool allows developers and businesses to easily detect whether their content was generated by AI, helping to increase transparency and counter the spread of misinformation.

SynthID works by subtly adjusting the probability of word choices in AI-generated texts. For example, if a user asks a chatbot, "What is your favorite fruit?", the watermark technology can increase the likelihood that the chatbot responds with "mango" rather than "apple" or "banana." This pattern of probability adjustments serves as the watermark, which is imperceptible to humans but detectable by machines.

This technique can be applied to a single sentence, which may contain several adjusted "probability scores." In longer texts, SynthID’s accuracy increases as a full page could contain hundreds of such adjustments. By analyzing these probability patterns, SynthID can determine whether a text was generated by an AI model. The system also works well on texts that have been edited, shortened, or paraphrased, though it struggles more with short or translated texts.

In addition to text, SynthID can watermark AI-generated images, videos, and audio. This allows users to verify if content, such as an image or video, was created by an AI model without compromising the visual or auditory quality.