Google has made an AI detector, but you can’t use it

AI detectors, although a potentially useful feature, are currently a waste. Tools like Zerogpt like to do a lot of false positives and false negatives. However, the need for a truly reliable AI detector is real, and that is why efforts continue to be made in this direction. Google has its own, although you cannot yet use it.

Google has just announced the synthetic detector, a new verification portal designed to identify the content created using its artificial intelligence tools, such as Gemini, the Imagen image generation model or the VEO video generation model. The synthetic detector works by scanning the downloaded supports for an imperceptible digital watermark, also named Synthed. Google has developed this watermark technology to integrate directly into the content generated by its AI models, including Gemini (text and multimodal), Imagen (images), Lyria (Audio) and Veo (video). According to the company, more than 10 billion content has already been tedious using this system. It is therefore a tool made by Google looking for this watermark and tells you if something is generated by Ai-Ai or not.

When you download a file – whether it is an image, an audio track, a video or a text document – at the synthetic detector portal, it seeks to see if this integrated watermark is present. And if this is the case, the portal will indicate that the content is probably generated by the AI ​​and, in some cases, highlights specific parts where the filigree is detected the most well. On the one hand, in audio files, it can emphasize the segments containing the watermark, and for images, it can indicate areas where the digital signature is probably present.

Related

6 best gemini features to try on your Google Pixel 9

You will love them all.

What I still don’t like about this is that it always seems to make a lot of assumptions. The detector can be “uncertain” of certain parts, which is not a good omen for a so-called reliable watermark method which can resist modifications and modifications. Just as he can be sure of certain bits, he could detect a watermark where there is none, or he could fail to detect something generated by AI. I would say that it would be more subject to false positives than false negatives, but false positives can always be a problem. I’m sure it will continue to be improved, however. A first part tool like this could be the most reliable way at the moment to find out if something has been generated by AI, but I would not say that there is still a 100% reliable and test of the balls to catch them all.

This detector currently takes place on a few people early, and it will be followed by a limited deployment for journalists, media professionals and researchers via a waiting list.

Source: Google