Google has introduced SynthID, a brand new technology for tracing AI-generated text. Now, by adding an invisible watermark, AI-generated content will be marked. In other words, the source of AI-generated texts can be identified thanks to this hidden mark. Here are the details…
Google’s secret signature on AI texts: SynthID will mark AI-generated content
SynthID, developed as part of Google’s Responsible GenAI Toolkit, is currently available for free download on the Hugging Face platform. This technology adds a hidden marker to the content without disrupting the natural flow of the text. In other words, this invisible marker can be used to determine whether a text was written by an artificial intelligence model or a human. So how does it work?
The working principle of artificial intelligence is based on probability calculations when creating text. Each word or character has a probability value and the AI creates the text based on these values. SynthID plays with these probability values and adds an invisible watermark to the text generated by the AI. In short, when a text is written by AI, SynthID marks that text, leaving a kind of digital fingerprint.
But as with any technology, SynthID has its limitations. Google admits that the technology is not yet effective, especially with short texts, reorganized or translated content. It should also not be used in content about real events, because in such cases there is a risk of distorting the accuracy of the text. For example, if an AI is providing information about a historical event, interfering with the tokens used to ensure the accuracy of that event could undermine the accuracy of the text.
In fact, SynthID technology has been used in Google’s Gemini AI model since spring 2023. With this technology, Google says, AI content can be detected more transparently. By the way, Google is not alone in this. OpenAI is also working on similar technologies and plans to add hidden markers to AI-generated content.
In short, SynthID is especially attractive for organizations and platforms that want transparency in content production. Although not yet perfect, AI-generated texts will no longer only be expressive and creative, but also traceable and identifiable.
DeepL ile çevrildi (https://www.deepl.com/app/?utm_source=ios&utm_medium=app&utm_campaign=share-translation)
