Listen to the audio version of the article
Google, with the introduction of a watermark on images generated by artificial intelligence, is the first tech company to launch a tool to distinguish content produced using artificial intelligence. The announcement of Deepmind, the division of Alphabet specializing in artificial intelligence, proceeds in the direction of the voluntary agreement that the seven leading companies in the field of artificial intelligence signed last July with the US presidency to preserve the integrity of information avoiding distortions such as deepfakes and protecting copyright and respecting the boundaries of human creativity.
A will, which already last May, during the annual I / O conference of Google, the CEO Sundar Pichai had declared to carry on arguing that the company was committed from the beginning to building its models to include watermarks and make the contents To the “recognizable”.
The tool, called SynthId, will initially be available to users of Google’s Ai Imagen image generator on the Vertex machine learning platform. Users will be able to generate images and choose whether or not to add the watermark. This particular type of watermark – explained the developer – incorporates changes to individual pixels in the images, resulting in fact invisible to the human eye. “It doesn’t change the image, the image quality or the resulting experience, but it is resistant to various transformations: cropping, resizing and all the things you might do to try to work around normal, traditional and simple watermarks,” he said. declared Demis Hassabis, CEO of Google DeepMind describing SynthId, which at the time of release is in Beta phase. In fact, the developer said that he is constantly working on improving the watermark, with the aim of making the watermark less and less perceptible but more easily detectable by detection tools.